阅读说明：本技术 企业数据推演计算系统及方法 (Enterprise data deduction computing system and method ) 是由 叶凯文 于 2021-09-03 设计创作，主要内容包括：本发明提供一种企业数据推演计算系统,包括业绩趋势建模模块、未来序列计算模块、现值计算模块、内部收益率计算模块、股权策略仿真模块、算法流程控制模块、云计算服务平台和数据存储模块,能够适应企业数据复杂多变、样本稀疏的特点,实现基于企业业绩的动态量化分析。所述企业数据推演计算系统涉及本发明提供的一种时间序列推演分析方法及其补充,通过分析差分值序列、差分变化率序列识别常态趋势方向形态及当前状态趋势方向形态建立未来序列推演模型,能够高效适应时间序列结构多变且样本稀疏量少的情况。所述企业数据推演计算系统还涉及本发明提供的一种联立时间序列集推演分析方法,按关系建立维度模型并构建联合推演模型,能够广泛描述时间序列系统并可灵活结合时间序列方法。(The invention provides an enterprise data deduction computing system which comprises a performance trend modeling module, a future sequence computing module, a present value computing module, an internal earning rate computing module, a share right strategy simulation module, an algorithm flow control module, a cloud computing service platform and a data storage module, and can adapt to the characteristics of complexity and changeability of enterprise data and sparse samples and realize dynamic quantitative analysis based on enterprise performance. The enterprise data deduction computing system relates to a time sequence deduction analysis method and supplement thereof, and a future sequence deduction model is established by analyzing a difference value sequence and a difference change rate sequence to identify a normal state trend direction form and a current state trend direction form, so that the enterprise data deduction computing system can be efficiently suitable for the conditions of variable time sequence structures and small sample sparse amount. The enterprise data deduction computing system also relates to a joint time sequence set deduction analysis method provided by the invention, a dimension model is built according to the relation, a joint deduction model is built, a time sequence system can be widely described, and the time sequence method can be flexibly combined.)

1.A time sequence deduction analysis method is characterized in that a difference value sequence and a difference change rate sequence of a time sequence are calculated, normal trend directions and forms of samples in an NS step length of the time sequence are identified according to the difference value sequence, the difference change rate sequence and a statistical significance level alpha, continuous samples with similar change characteristics with a last time step sample and current state trend directions and forms of the continuous samples are identified according to the difference value sequence, the difference change rate sequence, a confidence step number S and the statistical significance level alpha, MS step length time sequence model deduction parameters are calculated according to a time weight method of an annealing rate anr, a future sequence deduction model is built, and inertia processing is carried out on the future sequence deduction model according to the consistency of the current state trend directions and forms and the normal trend directions and forms.

2. The time series deduction analysis method according to claim 1, specifically comprising the following 6 steps:

step 1, calculating a difference value sequence and a difference change rate sequence of a time sequence; in the case of analyzing seasonal time series, a quarterly difference value series differentiated by year and a quarterly difference change rate series compared by year are used;

step 2, eliminating abnormal samples in the difference value sequence in the step 1;

step 3, identifying the normal trend direction and the form of the time sequence in the NS step according to the difference value sequence, the difference change rate sequence and the statistical significance level alpha; the method specifically comprises the following 2 substeps:

step 3.1, using the difference value sequence processed in the step 2, distributing the difference values according to t, wherein the significance level is alpha (0< alpha <0.5), and respectively performing single-tailed hypothesis test of a virtual hypothesis that the difference value < ═ 0, an opposite hypothesis that the difference value >0 and the virtual hypothesis that the difference value > -0 and the opposite hypothesis that the difference value <0 are respectively performed; accepting the opposite hypothesis if the single-tailed hypothesis testing virtual hypothesis is rejected; if the rejection difference value is less than 0, the acceptance difference value is greater than 0, and the normal trend is in an increase direction; if the reject difference value > is 0, the reject difference value <0 is accepted, and the normal trend direction is descending; when the virtual hypothesis with the difference value of 0 and the difference value of 0 is not rejected, the normal trend direction is random;

step 3.2, when the normal trend direction of the time sequence is identified to be increasing or decreasing according to the step 3.1, the difference value sequence processed in the step 2 is used, and if the sample data of the difference value sequence is not both positive values and not both negative values, the normal trend form is identified to be a linear state; if the sample data of the sequence of the difference values are all positive values or all negative values, calculating a logarithm sequence of the change rate sequence of the difference values, distributing the logarithm sequence according to t, wherein the significance level is alpha (0< alpha <0.5), and respectively carrying out virtual hypothesis that the logarithm value is 0 and the opposite hypothesis: single-tailed hypothesis test of logarithm value >0, virtual hypothesis that logarithm value >0, and opposite hypothesis that logarithm value < 0; accepting the opposite hypothesis if the single-tailed hypothesis testing virtual hypothesis is rejected; if the rejection logarithm value is less than 0, the acceptance logarithm value is greater than 0, and the normal trend form is an exponential form; if the rejection logarithm value > is 0, the acceptance logarithm value <0, and the normal trend is a gentle state; when the virtual assumptions of the logarithm value <0 and the logarithm value > <0 are not rejected, the normal trend form is a linear form;

step 4, identifying continuous samples which are similar to the change characteristics of the last time step sample and are continuous with the last time step in the time sequence and the current state trend direction and form of the continuous part according to the difference value sequence, the difference change rate sequence, the confidence step number S and the statistical significance level alpha, and specifically comprising the following 2 substeps:

step 4.1, using the difference value sequence of step 1; if the length of the differential value sequence is more than or equal to the confidence step S and the last S step data of the differential value sequence is positive or negative, establishing a new sequence as a current state differential sequence at the part which is continuous with the last data of the differential value sequence and has the same positive or negative sign in the differential value sequence, wherein if the current state differential sequence data is regular, the current state trend direction is increased, and if the current state differential sequence data is negative, the current state trend direction is decreased; if the length of the differential value sequence is more than or equal to the confidence setting step S and the last S step data of the differential value sequence are not both positive and not both negative, establishing a new sequence as the current state differential sequence at the part of the differential value sequence, which is continuous with the last data of the differential value sequence and has a sample which is not the same number as the previous S-1 step data; if the length of the current state differential sequence is more than or equal to the signal setting step S, the sequence is used for being distributed according to t and the significance level is alpha (0< alpha <0.5), and single-tail hypothesis test of virtual hypothesis that the differential value is 0, the opposite hypothesis that the differential value is 0 and the virtual hypothesis that the differential value is 0 and the opposite hypothesis that the differential value is 0 is respectively carried out; accepting the opposite hypothesis if the single-tailed hypothesis testing virtual hypothesis is rejected; if the rejection difference value is less than 0, the acceptance difference value is greater than 0, and the trend direction of the current state is increasing; if the rejection difference value > is 0, the rejection difference value <0 is accepted, and the trend direction of the current state is descending; when the virtual hypothesis with the difference value of 0 and the difference value of 0 is not rejected, the trend direction of the current state is random; if the length of the differential value sequence or the current state differential sequence is smaller than the signal receiving step S, the trend direction of the current state is random;

step 4.2, when the trend direction of the current state of the time sequence is identified to be increasing or decreasing according to the step 4.1, the current state differential sequence of the step 4.1 is used, if the current state differential sequence is positive or negative, the differential change rate sequence of the step 1 is used, and the logarithmic sequence of the differential change rate sequence in the period of the current state differential sequence is calculated; if the length of the logarithm sequence is less than the signal-placing step S, the trend form of the current state is a linear state; if the length of the log sequence is more than or equal to the confidence step S and the last S step data of the log sequence are both positive or negative, intercepting a differential change rate sequence sample according to the time period of the part, continuous with the last data of the log sequence, of which the positive and negative are the same in sign, of the log sequence to establish a new differential change rate sequence as a current state differential change rate sequence, wherein the current state differential change rate sequence is more than 1, the current state trend form is an index state, and the current state differential change rate sequence is less than 1, and the current state trend form is a flat state; if the length of the log sequence is more than or equal to the confidence step S and the last S step data of the log sequence are not both positive and not both negative, establishing a new differential change rate sequence as a current state differential change rate sequence according to a time period of a part of the log sequence, which is continuous with the last data of the log sequence and has a sample which is not of the same number as the previous S-1 step data; if the length of the current state difference change rate sequence is larger than or equal to the confidence step S, calculating a logarithm sequence of the current state difference change rate sequence, using the logarithm sequence to be distributed according to t and the significance level is alpha (0< alpha <0.5), and respectively carrying out single-tail hypothesis test on a virtual hypothesis that the logarithm value is 0, an opposite hypothesis that the logarithm value is 0 and an opposite hypothesis that the logarithm value is 0; accepting the opposite hypothesis if the single-tailed hypothesis testing virtual hypothesis is rejected; if the rejection logarithm value is less than 0, the acceptance logarithm value is greater than 0, and the current state trend form is an exponential form; if the rejection logarithm value > is 0, the acceptance logarithm value <0, and the current state trend form is a gentle state; when the virtual assumptions of the logarithm value <0 and the logarithm value > <0 are not rejected, the current state trend form is a linear state; if the length of the current state differential change rate sequence is less than the signal receiving step S, the current state trend form is a linear form;

step 5, calculating a time series model deduction parameter in the MS step length according to a time weight method of the annealing rate anr and establishing a future series deduction model, wherein the calculation is carried out under the following 8 conditions:

condition 5.1, the trend direction of the current state is random or unidentified trend form, the time weight mean value of the annealing rate anr of the differential sequence of the current state does not accord with the direction of the current state, and the time sequence is the condition of seasonal time sequence; under the condition, the time period of the current state differential sequence is traced back to 4 quarters as the current state time period, the weight mean value of each quarter sub-time sequence in the time period is calculated according to the annealing rate anr to be used as a quarter parameter, and the quarter with the sequence length of 0 uses the mean values of other quarter parameters; the future sequence deduction model is that the quaternary value of each season in the future is a quaternary value parameter of the current season;

condition 5.2, the current state trend direction is a condition of increasing or decreasing and no trend morphology is identified and the current state differential sequence coincides with the current state direction by the time weighted mean of the annealing rate anr and the time series is a seasonal time series; under the condition, the last sample of each quarter of the time sequence is used as a quarter parameter, and the quarter with the sequence length of 0 uses the mean value of other quarter parameters; calculating the weighted mean value of each quaternary subsequence of the differential sequence in the current state as a quaternary differential parameter according to the annealing rate anr, wherein the mean value of other quaternary differential parameters is used in the quaternary with the sequence length of 0; the future sequence deduction model is that the quaternary difference of each quaternary in the future is a quaternary difference parameter of the season, and the quaternary value of each quaternary in the future is the sum of the quaternary value parameter of the season and the accumulated value of the quaternary difference subsequence of the future sequence from the season to the season;

condition 5.3, the current trend form is linear, or the trend form is exponential or flat but the current state difference change rate sequence does not accord with the current state form according to the weight average value of the annealing rate anr, and the time sequence is the condition of seasonal time sequence; under the condition, the last sample of each quarter of the time sequence is used as a quarter parameter, and the quarter with the sequence length of 0 uses the mean value of other quarter parameters; tracing the sample time point of the current state differential change rate sequence back to 4 quarters as a current state time period, calculating the weight mean value of each quarter subsequence of the differential value sequence in the time period according to the annealing rate anr as a quarter differential parameter, and using the mean value of other quarter differential parameters in the quarter with the sequence length of 0; the future sequence deduction model is that the quaternary difference of each quaternary in the future is a quaternary difference parameter of the season, and the quaternary value of each quaternary in the future is the sum of the quaternary value parameter of the season and the accumulated value of the quaternary difference subsequence of the future sequence from the season to the season;

the condition 5.4 is that the current trend form is an exponential state or a flat state, the current state difference change rate sequence accords with the current state form according to the weight average value of the annealing rate anr, and the time sequence is a seasonal time sequence; under the condition, the last sample of each quarter of the time sequence is used as a quarter parameter, and the quarter with the sequence length of 0 uses the mean value of other quarter parameters; taking the last sample of each quarter of the difference value sequence as a quarter difference parameter, wherein the quarter with the sequence length of 0 uses the mean value of other quarter difference parameters; calculating the weight mean value of each quaternary subsequence of the current state difference change rate sequence according to the annealing rate anr to be used as a quaternary change rate parameter, wherein the mean value of other quaternary change rate parameters is used in the quaternary with the sequence length of 0; the future sequence deduction model is that the seasonal change rate of each seasonal in the future is a seasonal change rate parameter of the season, the seasonal difference of each seasonal in the future is the product of a seasonal difference parameter of the season and the cumulative value of the seasonal change rate subsequence of the future sequence to the season, and the seasonal value of each seasonal in the future is the sum of the seasonal value parameter of the season and the cumulative value of the seasonal difference subsequence of the future sequence to the season;

condition 5.5, the trend direction of the current state is random or unidentified trend form, the current state differential sequence does not accord with the current state direction according to the time weight mean value of the annealing rate anr, and the time sequence is a condition of non-seasonal time sequence; under the condition, the time period of the current state differential sequence is traced back to 1 step as the current state time period, and the weight mean value of the time sequence in the time period is calculated according to the annealing rate anr and is used as a value parameter; the future sequence deduction model is that the value of each sequence step in the future is a value parameter;

condition 5.6, the current state trend direction is a condition of increasing or decreasing and no trend morphology is identified and the current state differential sequence coincides with the current state direction by the time weighted mean of the annealing rate anr and the time sequence is a non-seasonal time sequence; under this condition, the last sample of the time series is taken as a value parameter; calculating the weight mean value of the difference sequence of the current state as a difference parameter according to the annealing rate anr; the future sequence deduction model is that the difference of each sequence step in the future is a difference parameter, and the value of each sequence step in the future is the sum of a value parameter and the accumulated value of the difference sequence of the future until the step;

condition 5.7, the current trend form is linear, or the trend form is exponential or flat but the current state difference change rate sequence does not accord with the current state form according to the weight average value of the annealing rate anr, and the time sequence is a non-seasonal time sequence condition; under this condition, the last sample of the time series is taken as a value parameter; backtracking the time period of the current state differential sequence by 1 step to be used as the current state time period, and calculating the weighted average of the differential value sequence in the time period according to the annealing rate anr to be used as a differential parameter; the future sequence deduction model is that the difference of each sequence step in the future is a difference parameter, and the value of each sequence step in the future is the sum of a value parameter and the accumulated value of the difference sequence of the future until the step;

the current trend form is an exponential state or a flat state, the current state difference change rate sequence is consistent with the current state form according to the weight average value of the annealing rate anr, and the time sequence is a non-seasonal time sequence condition; under this condition, the last sample of the time series is taken as a value parameter; taking the last sample of the difference value sequence as a difference parameter; calculating the weight mean value of the current state difference change rate sequence according to the annealing rate anr to be used as a change rate parameter; the future sequence deduction model is that the change rate of each future time step is a change rate parameter, the difference of each future time step is the product of a difference parameter and the accumulated value of the future sequence change rate sequence to the step, and the value of each future time step is the sum of a value parameter and the accumulated value of the future sequence difference sequence to the step;

step 6, carrying out inertia processing on the future sequence deduction model according to the consistency of the trend direction and the form of the current state and the trend direction and the form of the normal state; specifically, the following 2 conditions were used for calculation:

condition 6.1, if the current state trend direction is consistent with the normal state trend direction and if the current state trend form is consistent with the normal state trend form, the future sequence deduction model is maintained as the future sequence deduction model in the step 5;

if the trend direction of the current state is inconsistent with the normal trend direction, the future sequence deduction model deduces according to a method of a neutral trend direction, namely a random trend direction after the inertia year IY; if the time sequence is a seasonal time sequence, the seasonal value of each quarter after the inertia year IY is the last seasonal value of each seasonal value sequence of the future sequence in the inertia year IY; if the time sequence is a non-seasonal time sequence, the value of each sequence step after the inertia year IY is the value of the last sequence step of the future sequence in the inertia year IY;

if the current state trend direction is consistent with the normal state trend direction but the current state trend form is inconsistent with the normal state trend form, the future sequence deduction model deduces according to a method of a neutral trend form, namely a linear trend form after the inertia year IY under the condition of 6.3; if the time sequence is a seasonal time sequence, the quarterly difference of each quarter after the inertia number IY is the last quarterly difference of each quarterly difference sequence of the future sequence in the inertia number IY, and the quarterly value of each quarter is the sum of the last quarterly value of each quarterly value sequence of the future sequence in the inertia number IY and the accumulated value of the quarterly difference subsequence of the future sequence after the inertia number IY when the season difference subsequence reaches the season; if the time sequence is a non-seasonal time sequence, the difference of each sequence step after the inertia year IY is the last sequence step value of the future sequence difference sequence in the inertia year IY, and the value of each sequence step is the sum of the last sequence step value of the future sequence in the inertia year IY and the accumulated value of the step after the inertia year IY of the future sequence difference sequence.

3. A time series deduction analysis method, characterized by specifically comprising 6 steps, wherein step 1, step 2, step 3, step 4, and step 6 are the same as step 1, step 2, step 3, step 4, and step 6 of the time series deduction analysis method according to claim 1, and step 5 specifically calculates under the following 6 conditions:

condition 5.1, the trend direction of the current state is random or unidentified trend form, the weight mean value of each quaternary subsequence in the last RS step of the difference value sequence is calculated according to the annealing rate anr and is not consistent with the direction of the current state, and the time sequence is the condition of seasonal time sequence; under the condition, calculating the weight mean value of each quarterly sub-time sequence in the last RS step of the time sequence as a quarterly parameter according to the annealing rate anr, wherein the quarterly with the sequence length of 0 uses the mean values of other quarterly parameters; the future sequence deduction model is that the quaternary value of each season in the future is a quaternary value parameter of the current season;

condition 5.2, the trend direction of the current state is increasing or decreasing and no trend form is identified, when the weight mean value of each quaternary subsequence in the last RS step of the differential value sequence is calculated according to the annealing rate anr and the direction of the current state is consistent, or the current trend form is linear, or the trend form is exponential or flat, but the weight mean value of each quaternary subsequence in the last RS step of the differential change rate sequence is calculated according to the annealing rate anr and the form of the current state are not consistent, and the time sequence is a condition of seasonal time sequence; under the condition, the last sample of each quarter of the time sequence is used as a quarter parameter, and the quarter with the sequence length of 0 uses the mean value of other quarter parameters; calculating the weighted mean value of each quaternary subsequence in the last RS step of the differential value sequence according to the annealing rate anr by using the differential value sequence in the step 1 as a quaternary differential parameter, wherein the quaternary with the sequence length of 0 uses the mean value of other quaternary differential parameters; the future sequence deduction model is that the quaternary difference of each quaternary in the future is a quaternary difference parameter of the season, and the quaternary value of each quaternary in the future is the sum of the quaternary value parameter of the season and the accumulated value of the quaternary difference subsequence of the future sequence from the season to the season;

condition 5.3, the current trend form is exponential or flat, the weight average value of each quaternary subsequence in the last RS step of the difference change rate sequence is calculated according to the annealing rate anr and is consistent with the form of the current state, and the time sequence is the condition of seasonal time sequence; under the condition, the last sample of each quarter of the time sequence is used as a quarter parameter, and the quarter with the sequence length of 0 uses the mean value of other quarter parameters; taking the last sample of each quarter of the difference value sequence as a quarter difference parameter, wherein the quarter with the sequence length of 0 uses the mean value of other quarter difference parameters; calculating the weight mean value of each quaternary subsequence in the last RS step of the differential change rate sequence as a quaternary change rate parameter according to the annealing rate anr by using the differential change rate sequence in the step 1, wherein the quaternary with the sequence length of 0 uses the mean value of other quaternary change rate parameters; the future sequence deduction model is that the seasonal change rate of each seasonal in the future is a seasonal change rate parameter of the season, the seasonal difference of each seasonal in the future is the product of a seasonal difference parameter of the season and the cumulative value of the seasonal change rate subsequence of the future sequence to the season, and the seasonal value of each seasonal in the future is the sum of the seasonal value parameter of the season and the cumulative value of the seasonal difference subsequence of the future sequence to the season;

condition 5.4, the trend direction of the current state is random or unidentified trend form, the weight mean value of each quaternary subsequence in the last RS step of the difference value sequence is calculated according to the annealing rate anr and is not consistent with the direction of the current state, and the time sequence is a condition of non-seasonal time sequence; under the condition, calculating the weight average value in the last RS step of the time sequence as a value parameter according to the annealing rate anr; the future sequence deduction model is that the value of each sequence step in the future is a value parameter;

condition 5.5, the trend direction of the current state is increasing or decreasing, no trend form is identified, the weight mean value of each quaternary subsequence in the last RS step of the differential value sequence is calculated according to the annealing rate anr and is consistent with the direction of the current state, or the current trend form is linear, or the trend form is exponential or flat, the weight mean value of each quaternary subsequence in the last RS step of the differential change rate sequence is calculated according to the annealing rate anr and is not consistent with the form of the current state, and the time sequence is a condition of non-seasonal time sequence; under this condition, the last sample of the time series is taken as a value parameter; calculating the weight mean value of the difference value sequence in the last RS step as a difference parameter according to the annealing rate anr by using the difference value sequence in the step 1; the future sequence deduction model is that the difference of each sequence step in the future is a difference parameter, and the value of each sequence step in the future is the sum of a value parameter and the accumulated value of the difference sequence of the future until the step;

the current trend form is exponential or flat, the weight average value of each quaternary subsequence in the last RS step of the difference change rate sequence is calculated according to the annealing rate anr and is consistent with the form of the current state, and the time sequence is a non-seasonal time sequence condition; under this condition, the last sample of the time series is taken as a value parameter; taking the last sample of the difference value sequence as a difference parameter; calculating a weight mean value in the last RS step of the difference change rate sequence according to the annealing rate anr by using the difference change rate sequence in the step 1 as a change rate parameter; the future sequence deduction model is that the change rate of each future time step is a change rate parameter, the difference of each future time step is the product of a difference parameter and the accumulated value of the future sequence change rate sequence to the step, and the value of each future time step is the sum of a value parameter and the accumulated value of the future sequence difference sequence to the step.

4. A joint time sequence set deduction analysis method is characterized in that a time sequence set with an incidence relation is divided into a self-increment time sequence, a driving time sequence, a quota time sequence and a composite time sequence to form a joint time sequence set, a self-increment dimensional model, a driving dimensional model, a quota dimensional model and a composite dimensional model are respectively built according to time sequence classification, and a joint deduction model of the joint time sequence set is formed.

5. The joint time sequence set deduction analysis method according to claim 4, specifically comprising the following 6 steps:

step 1, dividing a time sequence set with an association relationship into a self-increment time sequence, a driving time sequence, a quota time sequence and a composite time sequence to form a simultaneous time sequence set;

step 2, calculating a future sequence deduction model of all the self-increment time sequences in the simultaneous time sequence set to serve as a self-increment dimension model; a future sequence deduction model of the self-increment time sequence is a self-increment dimension model;

step 3, defining traction dimension time sequences of all driving time sequences in the simultaneous time sequence set, wherein the traction dimension time sequences are time sequences in the simultaneous time sequence set, calculating ratio sequences of all driving time sequences and the traction dimension time sequences of the sequences, and calculating future sequence deduction models of all the ratio sequences; the driving dimension model is that the value of each future sequence step is the product of the value of each future sequence step of the future sequence deduction model of the ratio sequence and the value of each future sequence step of the future sequence deduction model of the traction dimension time sequence; a future sequence deduction model of the driving time sequence is a driving dimension model;

step 4, constructing quota dimensional models of all quota time sequences, defining a reference dimension time sequence of the quota time sequences and a quota of the quota time sequences, wherein the reference dimension time sequence is a time sequence in a simultaneous time sequence set; the quota dimensionality model is that the value of each future sequence step is the product of the quota time sequence and the value of each future sequence step of the future sequence deduction model of the reference dimensionality time sequence of the quota time sequence; a future sequence deduction model of the quota time sequence is a quota dimensionality model;

step 5, constructing a composite dimension model of all composite time sequences, defining an dimension adding time sequence set and a dimension reducing time sequence set of the composite time sequences, wherein the time sequences of the dimension adding time sequence set and the dimension reducing time sequence set are the time sequences in the simultaneous time sequence set; the composite dimension model is that the value of each sequence step in the future is the difference value of the sum of the values of the sequence steps of all time sequences of the dimension-added time sequence set of the composite time sequence and the sum of the values of the sequence steps of all time sequences of the dimension-subtracted time sequence set of the composite time sequence; a future sequence deduction model of the composite time sequence is a composite dimension model;

and 6, taking the future sequence deduction models of all time sequences in the simultaneous time sequence set as submodels to jointly form a joint deduction model of the simultaneous time sequence set.

6. The utility model provides an enterprise data deduction computing system, includes achievement trend modeling module, future sequence calculation module, present value calculation module, internal rate of return calculation module, share right strategy simulation module, algorithm flow control module, cloud computing service platform and data storage module, its characterized in that:

the performance trend modeling module comprises a time series model modeling module and a simultaneous time series set modeling module; the time sequence model modeling module is used for establishing a future sequence deduction model of historical time sequence data related to each dimension of the enterprise quarterly report profit list, and comprises a state updating model module, a state buffering model module and an ARIMA state model module; the state update model module is used for calculating a future sequence deduction model of a time sequence according to the time sequence deduction analysis method of claim 1; the state buffer model module is used for calculating a future sequence deduction model of a time sequence according to the time sequence deduction analysis method of claim 3; the ARIMA state model module is a future sequence deduction model for calculating a time sequence according to an ARIMA model in combination with the improved time sequence deduction analysis method of step 1-3 of claim 2; the simultaneous time sequence set modeling module is used for calculating a joint deduction model of an enterprise simultaneous time sequence set, and comprises a common enterprise simultaneous time sequence set modeling module, a bank enterprise simultaneous time sequence set modeling module, a security enterprise simultaneous time sequence set modeling module and an insurance enterprise time sequence set modeling module;

the future sequence calculation module is used for calculating a future sequence within a specific sequence step of a joint deduction model of the enterprise simultaneous time sequence set calculated by the performance trend modeling module;

the present value calculating module is used for calculating infinite step length future sequence present values of each sub-model in the joint deduction model of the enterprise simultaneous time sequence set calculated by the performance trend modeling module according to specific present rates; the system is used for calculating the discount value of the specific finite-step-length future sequence calculated by the future sequence calculation module according to the specific discount rate;

the internal yield rate calculation module is used for calculating the discount rate when the discount value of the infinite step length future sequence calculated by the search current value calculation module is equal to the specific price value; the calculating module is used for calculating the discount rate when the concrete finite-step-length future sequence discount value calculated by the searching module is equal to the concrete price value;

the stock right strategy simulation module comprises a factor value calculation module, a factor logic judgment module, a strategy logic integration module and a transaction simulation module; the factor value calculation module is used for calculating a value factor, a technical factor and a financial factor, wherein the value factor is a factor value obtained or derived based on the calculation result of the current value calculation module or the internal rate of return calculation module; the factor logic judgment module is used for carrying out true and false logic calculation according to the factor value or the factor sequence calculated by the factor value calculation module; the strategy logic integration module is used for integrating the factor logic judgment result calculated by the factor logic judgment module; the transaction simulation module is used for calculating historical simulation transaction operation data and historical simulation transaction income data according to specific transaction type transaction conditions and factor logic judgment integration results calculated by the strategy logic integration module;

the algorithm flow control module is used for transmitting instructions and parameters to the performance trend modeling module, the future sequence calculating module, the present value calculating module, the internal rate of return calculating module and the share right strategy simulating module, arranging a calculating flow and acquiring and processing a calculating result; the cloud computing service platform is used for receiving and processing computing requests of the cloud computing service platform and feeding back computing results; the data storage module is used for requesting data required by calculation to the data storage module and processing data feedback; for requesting external data and processing external data feedback;

the cloud computing service platform is used for providing an interface interaction function and an HTTPS/HTTP API interaction function, and comprises an interface function module, an API function module and an interaction control module; the interface function module comprises a performance trend modeling interface function module, a value calculation analysis interface function module and a strategy history simulation interface function module; the performance trend modeling interface function module comprises a state updating model calculation interface function module, a state buffer model calculation interface function module and an ARIMA state model calculation interface function module; the state updating model computing interface function module, the state buffer model computing interface function module and the ARIMA state model computing interface function module respectively comprise a computing parameter template setting function, a specific enterprise specific time simultaneous time sequence set joint deduction model computing function and a specific enterprise specific time simultaneous time sequence set future sequence set deduction computing function; the value calculation analysis interface function module comprises a performance present value calculation interface function module and a performance internal rate of return calculation interface function module; the performance calculation interface functional module comprises a calculation parameter template setting function and a performance calculation function of a specific enterprise in a specific period; the performance internal rate of return calculation interface function module comprises a calculation parameter template setting function and a performance internal rate of return calculation function of a specific enterprise in a specific period; the strategy history simulation interface function module comprises a common strategy simulation interface function module, a financing strategy simulation interface function module and a financing strategy simulation interface function module; the common strategy simulation interface function module, the financing strategy simulation interface function module and the financing strategy simulation interface function module all comprise strategy parameter template setting and editing and specific enterprise strategy simulation calculation functions;

the API function module comprises a state updating model calculation API module, a state buffer model calculation API module, an ARIMA state model calculation API module, a state updating deduction calculation API module, a state buffer deduction calculation API module, an ARIMA state deduction calculation API module, a performance present value calculation API module, a performance internal yield calculation API module and a strategy history simulation API module; the state updating model calculation API module, the state buffer model calculation API module and the ARIMA state model calculation API module all provide a calculation function of a joint deduction model of a joint time sequence set at a specific enterprise specific time period; the state updating deduction calculation API module, the state buffering deduction calculation API module and the ARIMA state deduction calculation API module provide a deduction calculation function of a future sequence set of the specific enterprise specific time simultaneous time sequence set; the achievement value calculation API module provides the function of calculating the achievement value of a specific enterprise in a specific period; the performance internal rate of return calculation API module provides the function of calculating the performance internal rate of return of specific enterprise in specific period; the strategy historical simulation API module provides a strategy simulation calculation function for a specific enterprise at a specific period;

the interaction control module is used for receiving and processing the instruction of setting the template by the interface function module and storing the template setting data into the data storage module; the interface function module is used for receiving and processing a data request of the interface function module, acquiring data from the data storage module and feeding back a data result to the interface function module; the system comprises an algorithm flow control module, an interface function module, an API function module and a control module, wherein the algorithm flow control module is used for receiving and processing a calculation request of the interface function module and the API function module, providing the calculation request to the algorithm flow control module, acquiring a calculation result, processing and converting the calculation result and feeding back the calculation result to the interface function module and the API function module;

the data storage module is used for storing calculation parameter template data set by the state updating model calculation interface function module, the state buffer model calculation interface function module, the ARIMA state model calculation interface function module, the performance present value calculation interface function module, the performance internal yield calculation interface function module and other modules, and storing strategy parameter template data set and edited by the common strategy simulation interface function module, the financing strategy simulation interface function module and other modules.

Technical Field

The invention relates to an enterprise data deduction computing system, in particular to a time sequence deduction analysis method, a supplement method thereof and a joint time sequence set deduction analysis method.

Background

For the analysis and deduction method of time series, statistical analysis methods such as ARIMA model, exponential smoothing method, moving average method, etc., and LSTM (long short term memory network) algorithm designed by combining neural network algorithm, etc. are commonly used in the industry and the academia. Because the time series samples have the characteristic of timeliness, exogenous variables influencing the time series data are often obviously and complexly changed in a time series period, and particularly under the conditions of long time interval of the time series samples, long time series period and small sample sparse amount, the currently and commonly used statistical method and neural network algorithm are difficult to adapt. The specific defects are as follows:

statistical analysis methods such as an ARIMA model, an exponential smoothing method, a moving average method and the like cannot directly identify changes and select proper time sequence samples for analysis under the conditions that the structure of an exogenous variable is changed rapidly and is difficult to monitor, and the sample sparsity is small.

The ARIMA model needs to consume large computing resources when searching for optimal P, D, Q and optimal seasons P, D and Q, and needs to acquire dozens of effective sample quantities under a stable exogenous variable structure to carry out reasonable parameter estimation, so that the ARIMA model cannot adapt to the conditions of long time period, large exogenous change and small sample dilution quantity, such as the condition of enterprise seasonal time sequence analysis.

The neural network algorithm including the LSTM has larger randomness and subjectivity in input form and arrangement of neural nodes, thousands of data samples are required for training even if few nodes are arranged, and the method is a violent computing method consuming a large amount of computing resources and obviously not suitable for situations of scarce data and complex change.

In summary, the existing method cannot efficiently solve the time series analysis deduction problems of long time period, complex exogenous influence change and small sample sparse amount.

The following problems and challenges continue to exist with respect to enterprise data deduction calculations and equity value assessments:

the enterprise operation is periodic, and is influenced by complex and changeable external factors all the time, so that the enterprise data has the characteristics of long data interval and complex exogenous variable change.

The data volume of the enterprise standardized public data is small, and the enterprise standardized data which can be legally contacted by the public is limited to quarterly reported data which is regularly reported by listed companies. Non-enterprise published data often presents confidence problems.

The models of market profitability, market equity and market sales rate commonly used for enterprise valuation are limited to the current performance of the enterprise and cannot reflect the performance growth of the enterprise.

The qualitative analysis of enterprises one by one is usually carried out manually on the performance growth of the enterprises, and the automatic quantitative improvement of the efficiency, the transparency and the objectivity of a standard algorithm cannot be realized.

The enterprise standardized time sequence data embody a dynamic system of correlated time sequence sets, and a deduction algorithm of the simultaneous time sequence sets needs to be established for analysis and calculation.

The quantitative calculation method for the equity trading strategy of the listed enterprise usually mainly takes short-term technical indexes and static financial indexes, and lacks systematic dynamic deduction analysis of enterprise performance data.

In summary, the enterprise data deduction calculation and the stock right value evaluation need to solve the problems of long time period of the enterprise time sequence data, complex exogenous influence change and less sample sparse amount, a reasonable simultaneous time sequence set model is needed to perform systematic dynamic deduction analysis on the enterprise performance data, and automatic quantification is needed to be realized through a standard algorithm so as to improve the working efficiency, transparency and objectivity of related work.

Disclosure of Invention

The invention provides a time series deduction analysis method, aiming at the problems of long time period, complex exogenous influence change and less sample sparsity in time series deduction calculation. The method is realized by the following technical scheme:

calculating a difference value sequence and a difference change rate sequence of the time sequence; identifying the normal trend direction and the form of the time sequence in the NS step length according to the difference value sequence, the difference change rate sequence and the statistical significance level alpha; according to the difference value sequence, the difference change rate sequence, the confidence step number S and the statistical significance level alpha, identifying continuous samples with time sequence and last time step sample change characteristics similar and the current state trend direction and form of the continuous part; calculating a time series model deduction parameter in the MS step length according to a time weight method of the annealing rate anr and establishing a future series deduction model; carrying out inertial processing on the future sequence deduction model according to the consistency of the current state trend direction and form and the normal state trend direction and form; the method specifically comprises the following steps:

step 1, calculating a difference value sequence and a difference change rate sequence of a time sequence; in the case of analyzing seasonal time series, a quarterly difference value series differentiated by year and a quarterly difference change rate series compared by year are used.

And step 2, eliminating the samples judged as abnormal samples in the difference value sequence in the step 1.

And 3, identifying the normal trend direction and the form of the sample of the time sequence in the NS step length according to the difference value sequence, the difference change rate sequence and the statistical significance level alpha. The method specifically comprises the following steps:

step 3.1, using the difference value sequence processed in the step 2, distributing the difference values according to t, wherein the significance level is alpha (0< alpha <0.5), and respectively performing single-tailed hypothesis test of a virtual hypothesis that the difference value < ═ 0, an opposite hypothesis that the difference value >0 and the virtual hypothesis that the difference value > -0 and the opposite hypothesis that the difference value <0 are respectively performed; if the single-tailed hypothesis test virtual hypothesis is rejected, the opponent hypothesis is accepted. If the rejection difference value is less than 0, the acceptance difference value is greater than 0, and the normal trend is in an increase direction; if the reject difference value > is 0, the reject difference value <0 is accepted, and the normal trend direction is descending; when the dummy hypothesis with the difference value <0 and the difference value > <0 is not rejected, the normal trend direction is random.

Step 3.2, when the normal trend direction of the time sequence is identified to be increasing or decreasing according to the step 3.1, the difference value sequence processed in the step 2 is used, and if the sample data of the difference value sequence is not both positive values and not both negative values, the normal trend form is identified to be a linear state; if the sample data of the sequence of the difference values are all positive values or all negative values, calculating a logarithm sequence of the change rate sequence of the difference values, distributing the logarithm sequence according to t, wherein the significance level is alpha (0< alpha <0.5), and respectively carrying out virtual hypothesis that the logarithm value is 0 and the opposite hypothesis: single-tailed hypothesis test of logarithm value >0, virtual hypothesis that logarithm value >0, and opposite hypothesis that logarithm value < 0; if the single-tailed hypothesis test virtual hypothesis is rejected, the opponent hypothesis is accepted. If the rejection logarithm value is less than 0, the acceptance logarithm value is greater than 0, and the normal trend form is an exponential form; if the rejection logarithm value > is 0, the acceptance logarithm value <0, and the normal trend is a gentle state; when the virtual assumption that the logarithm value is less than 0 and the logarithm value is greater than 0 is not rejected, the normal trend behavior is linear.

And 4, identifying continuous samples of which the time sequences are similar to the change characteristics of the samples of the last time step and are continuous with the last time step and the current state trend direction and the current state trend form of the continuous samples according to the difference value sequence, the difference change rate sequence, the confidence step number S and the statistical significance level alpha. The method specifically comprises the following steps:

step 4.1, the sequence of difference values of step 1 is used. And if the length of the differential value sequence is more than or equal to the confidence step S and the last S step data of the differential value sequence is positive or negative, establishing a new sequence as the current state differential sequence at the part which is continuous with the last data of the differential value sequence and has the same positive or negative sign in the differential value sequence, wherein if the current state differential sequence data is regular, the current state trend direction is increased, and if the current state differential sequence data is negative, the current state trend direction is decreased. And if the length of the differential value sequence is more than or equal to the confidence setting step S and the last S step data of the differential value sequence are not both positive and not both negative, establishing a new sequence as the current state differential sequence for the part, which is continuous with the last data of the differential value sequence and has a sample which is not the same number as the previous S-1 step data, in the differential value sequence. If the length of the current state differential sequence is more than or equal to the signal setting step S, the sequence is used for being distributed according to t and the significance level is alpha (0< alpha <0.5), and single-tail hypothesis test of virtual hypothesis that the differential value is 0, the opposite hypothesis that the differential value is 0 and the virtual hypothesis that the differential value is 0 and the opposite hypothesis that the differential value is 0 is respectively carried out; if the single-tailed hypothesis test virtual hypothesis is rejected, the opponent hypothesis is accepted. If the rejection difference value is less than 0, the acceptance difference value is greater than 0, and the trend direction of the current state is increasing; if the rejection difference value > is 0, the rejection difference value <0 is accepted, and the trend direction of the current state is descending; when the virtual hypothesis with the difference value <0 and the difference value > <0 is not rejected, the current state trend direction is random. If the length of the difference value sequence or the current state difference sequence is smaller than the signal receiving step S, the trend direction of the current state is random.

And 4.2, when the trend direction of the current state of the time sequence is identified to be increasing or decreasing according to the step 4.1, using the current state differential sequence of the step 4.1, and if the current state differential sequence is positive or negative, using the differential change rate sequence of the step 1 to calculate the logarithmic sequence of the differential change rate sequence in the period of the current state differential sequence. And if the length of the logarithmic sequence is less than the confidence step S, the trend form of the current state is a linear state. And if the length of the log sequence is more than or equal to the confidence step S and the last S step data of the log sequence is positive or negative, intercepting a differential change rate sequence sample according to the time period of the part, continuous with the last data of the log sequence, of positive, negative and same sign in the log sequence to establish a new differential change rate sequence as a current state differential change rate sequence, wherein the current state differential change rate sequence is more than 1, the current state trend form is an exponential state, and the current state differential change rate sequence is less than 1, and the current state trend form is a flat state. And if the length of the log sequence is more than or equal to the confidence step S and the last S step data of the log sequence are not both positive and not both negative, establishing a new differential change rate sequence as the current state differential change rate sequence according to the time period of a part of the log sequence, which is continuous with the last data of the log sequence and has a sample with a non-uniform number with the previous S-1 step data. If the length of the current state difference change rate sequence is larger than or equal to the confidence step S, calculating a logarithm sequence of the current state difference change rate sequence, using the logarithm sequence to be distributed according to t and the significance level is alpha (0< alpha <0.5), and respectively carrying out single-tail hypothesis test on a virtual hypothesis that the logarithm value is 0, an opposite hypothesis that the logarithm value is 0 and an opposite hypothesis that the logarithm value is 0; if the single-tailed hypothesis test virtual hypothesis is rejected, the opponent hypothesis is accepted. If the rejection logarithm value is less than 0, the acceptance logarithm value is greater than 0, and the current state trend form is an exponential form; if the rejection logarithm value > is 0, the acceptance logarithm value <0, and the current state trend form is a gentle state; when the virtual hypothesis with the logarithm value of 0 and the logarithm value of 0 is not rejected, the current state trend form is a linear form. And if the length of the current state differential change rate sequence is less than the signal receiving step S, the current state trend form is a linear form.

And step 5, calculating a time series model deduction parameter in the MS step length according to a time weight method of the annealing rate anr and establishing a future series deduction model. Specifically, the calculation was carried out under the following conditions:

condition 5.1, the current state trend direction is random, or no trend morphology identified and the current state differential sequence does not match the current state direction by the time weighted mean of the annealing rate anr, and the time sequence is a condition of a seasonal time sequence. Under the condition, the time period of the current state differential sequence is traced back to 4 quarters as the current state time period, the weighted mean value of each quarter sub-time sequence in the time period is calculated according to the annealing rate anr to be used as a quarter parameter, and the quarter with the sequence length of 0 uses the mean values of other quarter parameters. The future sequence deduction model is that the quaternary value of each season in the future is a quaternary value parameter of the current season.

Condition 5.2, the current state trend direction is a condition of increasing or decreasing and no trend morphology identified and the current state differential sequence coincides with the current state direction by a time-weighted mean of the annealing rate anr and the time series is a seasonal time series. Under this condition, the last sample of each quarter of the time series is taken as a quarter parameter, and the quarter with a sequence length of 0 uses the mean of the other quarter parameters. And calculating the weighted mean value of each quaternary subsequence of the differential sequence in the current state as a quaternary differential parameter according to the annealing rate anr, wherein the mean value of other quaternary differential parameters is used in the quaternary with the sequence length of 0. The future sequence deduction model is that the quaternary difference of each quaternary in the future is a quaternary difference parameter of the season, and the quaternary value of each quaternary in the future is the sum of the quaternary value parameter of the season and the accumulated value of the quaternary difference subsequence of the future sequence until the season.

And 5.3, the current trend morphology is a linear morphology, or the trend morphology is an exponential morphology or a flat morphology but the current state differential change rate sequence does not accord with the current state morphology according to the weighted average of the annealing rate anr, and the time sequence is a seasonal time sequence condition. Under this condition, the last sample of each quarter of the time series is taken as a quarter parameter, and the quarter with a sequence length of 0 uses the mean of the other quarter parameters. And tracing the sample time point of the current state differential change rate sequence back to 4 quarters as a current state time period, calculating the weight mean of each quarter subsequence of the differential value sequence in the time period according to the annealing rate anr to be used as a quarter differential parameter, and using the mean of other quarter differential parameters in the quarter with the sequence length of 0. The future sequence deduction model is that the quaternary difference of each quaternary in the future is a quaternary difference parameter of the season, and the quaternary value of each quaternary in the future is the sum of the quaternary value parameter of the season and the accumulated value of the quaternary difference subsequence of the future sequence until the season.

And 5.4, the current trend morphology is exponential or flat and the current state differential rate of change sequence conforms to the current state morphology according to the weighted mean of the annealing rate anr, and the time sequence is a condition of seasonal time sequence. Under this condition, the last sample of each quarter of the time series is taken as a quarter parameter, and the quarter with a sequence length of 0 uses the mean of the other quarter parameters. And taking the last sample of each quarter of the differential value sequence as a quarter differential parameter, wherein the quarter with the sequence length of 0 uses the mean value of other quarter differential parameters. And calculating the weighted mean value of each quaternary subsequence of the current state difference change rate sequence according to the annealing rate anr as a quaternary change rate parameter, wherein the mean value of other quaternary change rate parameters is used in the quaternary with the sequence length of 0. The future sequence deduction model is that the seasonal change rate of each seasonal in the future is a seasonal change rate parameter of the season, the seasonal difference of each seasonal in the future is the product of a seasonal difference parameter of the season and the cumulative value of the seasonal change rate subsequence of the future sequence to the season, and the seasonal value of each seasonal in the future is the sum of the seasonal value parameter of the season and the cumulative value of the seasonal difference subsequence of the future sequence to the season.

Condition 5.5, the current state trend direction is random, or no trend morphology identified and the current state differential sequence does not match the current state direction by the time weighted mean of the annealing rate anr, and the time sequence is a condition of a non-seasonal time sequence. Under the condition, the time period of the current state differential sequence is traced back to 1 step as the current state time period, and the weight average value of the time sequence in the time period is calculated according to the annealing rate anr and is used as a value parameter. The future sequence deduction model is that the value of each sequence step in the future is a value parameter.

Condition 5.6, the current state trend direction is a condition of increasing or decreasing and no trend morphology identified and the current state differential sequence coincides with the current state direction by a time weighted mean of the annealing rate anr and the time sequence is a non-seasonal time sequence. Under this condition, the last sample of the time series is taken as the value parameter. The difference parameter is calculated as the mean of the current state difference sequence weights at the annealing rate anr. The future sequence deduction model is that the difference of each sequence step in the future is a difference parameter, and the value of each sequence step in the future is the sum of a value parameter and the accumulated value of the difference sequence of the future until the step.

And 5.7, the current trend morphology is a linear morphology, or the trend morphology is an exponential morphology or a flat morphology but the current state difference change rate sequence does not accord with the current state morphology according to the weighted average of the annealing rate anr, and the time sequence is a non-seasonal time sequence condition. Under this condition, the last sample of the time series is taken as the value parameter. And (3) tracing the time period of the current state differential sequence back to 1 step to serve as the current state time period, and calculating the weighted average of the differential value sequence in the time period according to the annealing rate anr to serve as a differential parameter. The future sequence deduction model is that the difference of each sequence step in the future is a difference parameter, and the value of each sequence step in the future is the sum of a value parameter and the accumulated value of the difference sequence of the future until the step.

And 5.8, the current trend form is an exponential form or a flat form, the current state differential change rate sequence accords with the current state form according to the weighted average value of the annealing rate anr, and the time sequence is a non-seasonal time sequence condition. Under this condition, the last sample of the time series is taken as the value parameter. The last sample of the sequence of difference values is taken as the difference parameter. The mean of the weights of the current state differential rate of change sequence is calculated as the rate of change parameter by the anneal rate anr. The future sequence deduction model is that the change rate of each future time step is a change rate parameter, the difference of each future time step is the product of a difference parameter and the accumulated value of the future sequence change rate sequence to the step, and the value of each future time step is the sum of a value parameter and the accumulated value of the future sequence difference sequence to the step.

And 6, carrying out inertia processing on the future sequence deduction model according to the consistency of the current state trend direction and form and the normal state trend direction and form. Specifically, the calculation was carried out under the following conditions:

conditional 6.1, if the current state trend direction is consistent with the normal state trend direction and if the current state trend morphology is consistent with the normal state trend morphology, the future sequence deduction model is maintained as the future sequence deduction model of step 5.

If the trend direction of the current state is inconsistent with the normal trend direction, the future sequence deduction model deduces according to a method of a neutral trend direction, namely a random trend direction after the inertia year IY; if the time sequence is a seasonal time sequence, the seasonal value of each quarter after the inertia year IY is the last seasonal value of each seasonal value sequence of the future sequence in the inertia year IY; if the time series is a non-seasonal time series, the value of each sequence step after the inertia year IY is the value of the last sequence step of the future sequence within the inertia year IY.

If the current state trend direction is consistent with the normal state trend direction but the current state trend form is inconsistent with the normal state trend form, the future sequence deduction model deduces according to a method of a neutral trend form, namely a linear trend form after the inertia year IY under the condition of 6.3; if the time sequence is a seasonal time sequence, the quarterly difference of each quarter after the inertia number IY is the last quarterly difference of each quarterly difference sequence of the future sequence in the inertia number IY, and the quarterly value of each quarter is the sum of the last quarterly value of each quarterly value sequence of the future sequence in the inertia number IY and the accumulated value of the quarterly difference subsequence of the future sequence after the inertia number IY when the season difference subsequence reaches the season; if the time sequence is a non-seasonal time sequence, the difference of each sequence step after the inertia year IY is the last sequence step value of the future sequence difference sequence in the inertia year IY, and the value of each sequence step is the sum of the last sequence step value of the future sequence in the inertia year IY and the accumulated value of the step after the inertia year IY of the future sequence difference sequence.

Compared with the prior art, the time series deduction analysis method of the invention content 1 can better adapt to the situation that the time series trend structure is changeable due to the fact that external factors are changed in a complex mode, can sensitively discover abnormal changes and verify a stable and significant trend and conduct automatic deduction processing, can conduct effective deduction calculation on the time series even under the limited condition that the sample dilution amount is small, and consumes far lower amount of calculation resources than an automatic ARIMA algorithm and a neural network algorithm.

Summary of the inventionthe present invention provides a time series deduction analysis method similar to the time series deduction analysis method of the summary of the invention 1 provided by the present invention, as a supplement to the time series deduction analysis method of the summary of the invention 1. The specific steps 1, 2, 3, 4, and 6 are the same as the steps 1, 2, 3, 4, and 6 of the time series deduction analysis method of the invention content 1, and the step 5 is calculated under the following conditions:

and 5.1, the trend direction of the current state is random or unidentified trend form, the weight mean value of each quaternary subsequence in the last RS step of the difference value sequence is calculated according to the annealing rate anr and does not accord with the direction of the current state, and the time sequence is the condition of seasonal time sequence. Under this condition, the weighted mean of each quarterly sub-time series in the last RS step of the time series is calculated as the quarterly parameter according to the annealing rate anr, and the quarterly with the sequence length of 0 uses the mean of other quarterly parameters. The future sequence deduction model is that the quaternary value of each season in the future is a quaternary value parameter of the current season.

And 5.2, the trend direction of the current state is an increasing or decreasing condition, no trend morphology is identified, when the weight mean value of each quaternary subsequence in the last RS step of the differential value sequence calculated according to the annealing rate anr is consistent with the direction of the current state, or the current trend morphology is a linear morphology, or the trend morphology is an exponential morphology or a flat morphology, but the weight mean value of each quaternary subsequence in the last RS step of the differential change rate sequence calculated according to the annealing rate anr is not consistent with the morphology of the current state, and the time sequence is a seasonal time sequence. Under this condition, the last sample of each quarter of the time series is taken as a quarter parameter, and the quarter with a sequence length of 0 uses the mean of the other quarter parameters. And (3) calculating the weighted mean value of each quaternary subsequence in the last RS step of the differential value sequence as a quaternary difference parameter according to the annealing rate anr by using the differential value sequence in the step 1, wherein the mean value of other quaternary difference parameters is used in the quarter with the sequence length of 0. The future sequence deduction model is that the quaternary difference of each quaternary in the future is a quaternary difference parameter of the season, and the quaternary value of each quaternary in the future is the sum of the quaternary value parameter of the season and the accumulated value of the quaternary difference subsequence of the future sequence until the season.

And 5.3, the current trend form is exponential or flat, the weight average value of each quaternary subsequence in the last RS step of the difference change rate sequence calculated according to the annealing rate anr is consistent with the form of the current state, and the time sequence is the condition of seasonal time sequence. Under this condition, the last sample of each quarter of the time series is taken as a quarter parameter, and the quarter with a sequence length of 0 uses the mean of the other quarter parameters. And taking the last sample of each quarter of the differential value sequence as a quarter differential parameter, wherein the quarter with the sequence length of 0 uses the mean value of other quarter differential parameters. And (3) calculating the weighted mean value of each quaternary subsequence in the last RS step of the differential change rate sequence as a quaternary change rate parameter according to the annealing rate anr by using the differential change rate sequence in the step 1, wherein the mean value of other quaternary change rate parameters is used in the quaternary with the sequence length of 0. The future sequence deduction model is that the seasonal change rate of each seasonal in the future is a seasonal change rate parameter of the season, the seasonal difference of each seasonal in the future is the product of a seasonal difference parameter of the season and the cumulative value of the seasonal change rate subsequence of the future sequence to the season, and the seasonal value of each seasonal in the future is the sum of the seasonal value parameter of the season and the cumulative value of the seasonal difference subsequence of the future sequence to the season.

And 5.4, the trend direction of the current state is random or unidentified trend form, the weight mean value of each quaternary subsequence in the last RS step of the differential value sequence is calculated according to the annealing rate anr and does not accord with the direction of the current state, and the time sequence is a non-seasonal time sequence condition. Under this condition, the weight average value in the last RS step of the time series was calculated as a value parameter in accordance with the annealing rate anr. The future sequence deduction model is that the value of each sequence step in the future is a value parameter.

And 5.5, the trend direction of the current state is increasing or decreasing, no trend form is identified, the weight mean value of each quaternary subsequence in the last RS step of the differential value sequence is calculated according to the annealing rate anr and is consistent with the direction of the current state, or the current trend form is linear, or the trend form is exponential or flat, the weight mean value of each quaternary subsequence in the last RS step of the differential change rate sequence is calculated according to the annealing rate anr and is not consistent with the form of the current state, and the time sequence is a non-seasonal time sequence condition. Under this condition, the last sample of the time series is taken as the value parameter. And (3) calculating the weight mean value in the last RS step of the differential value sequence as a differential parameter according to the annealing rate anr by using the differential value sequence in the step 1. The future sequence deduction model is that the difference of each sequence step in the future is a difference parameter, and the value of each sequence step in the future is the sum of a value parameter and the accumulated value of the difference sequence of the future until the step.

And 5.6, the current trend form is exponential or flat, the weight average value of each quaternary subsequence in the last RS step of the difference change rate sequence calculated according to the annealing rate anr is consistent with the form of the current state, and the time sequence is a non-seasonal time sequence condition. Under this condition, the last sample of the time series is taken as the value parameter. The last sample of the sequence of difference values is taken as the difference parameter. And (3) calculating a weight mean value in the last RS step of the difference change rate sequence according to the annealing rate anr by using the difference change rate sequence in the step 1 as a change rate parameter. The future sequence deduction model is that the change rate of each future time step is a change rate parameter, the difference of each future time step is the product of a difference parameter and the accumulated value of the future sequence change rate sequence to the step, and the value of each future time step is the sum of a value parameter and the accumulated value of the future sequence difference sequence to the step.

The time series deduction analysis method of the invention content 2 provided by the invention content 1 has the beneficial technical effects that the excessive influence of abnormal factors on the time series deduction analysis can be buffered, and the time series deduction analysis which is obviously stable is more time-efficient.

The invention content 3 provides a joint time sequence set deduction analysis method aiming at the deduction analysis problem of a dynamic system of a time sequence set which is mutually associated. The method is realized by the following technical scheme:

dividing a time sequence set with an incidence relation into a self-increment time sequence, a driving time sequence, a quota time sequence and a composite time sequence to form a simultaneous time sequence set, and respectively establishing a self-increment dimensional model, a driving dimensional model, a quota dimensional model and a composite dimensional model according to time sequence classification to form a joint deduction model of the simultaneous time sequence set; the method specifically comprises the following steps:

step 1, dividing a time sequence set with an association relationship into a self-increment time sequence, a driving time sequence, a quota time sequence and a composite time sequence to form a simultaneous time sequence set.

And 2, calculating a future sequence deduction model of all the self-increment time sequences in the simultaneous time sequence set to serve as a self-increment dimension model. The future sequence deduction model of the self-increment time sequence is a self-increment dimension model.

And 3, defining traction dimension time sequences of all driving time sequences in the simultaneous time sequence set, wherein the traction dimension time sequences are the time sequences in the simultaneous time sequence set, calculating a ratio sequence of all the driving time sequences and the traction dimension time sequences of the sequences, and calculating a future sequence deduction model of all the ratio sequences, wherein the driving dimension model is that the value of each future sequence step is the product of the value of each future sequence step of the future sequence deduction model of the ratio sequence and the value of each future sequence step of the future sequence deduction model of the traction dimension time sequences. The future sequence deduction model of the driving time sequence is a driving dimension model.

And 4, constructing quota dimensional models of all quota time sequences, defining a reference dimension time sequence of the quota time sequences and a quota of the quota time sequences, wherein the reference dimension time sequence is a time sequence in a simultaneous time sequence set. The quota dimension model is that the value of each future sequence step is the product of the quota time series and the value of each future sequence step of the future sequence deduction model of the reference dimension time series of the quota time series. The future sequence deduction model of the quota time sequence is a quota dimension model.

And 5, constructing a composite dimension model of all the composite time sequences, defining an dimension-added time sequence set of the composite time sequences and a dimension-subtracted time sequence set of the composite time sequences, wherein the time sequences of the dimension-added time sequence set and the dimension-subtracted time sequence set are the time sequences in the simultaneous time sequence sets. The composite dimension model is that the value of each sequence step in the future is the difference value of the sum of the values of the sequence steps of all time sequences of the dimension-added time sequence set of the composite time sequence and the sum of the values of the sequence steps of all time sequences of the dimension-subtracted time sequence set of the composite time sequence. And the future sequence deduction model of the composite time sequence is a composite dimension model.

And 6, taking the future sequence deduction models of all time sequences in the simultaneous time sequence set as submodels to jointly form a joint deduction model of the simultaneous time sequence set.

The simultaneous time sequence set deduction analysis method of the invention content 3 provided by the invention has the beneficial technical effect that dynamic systems with correlated time sequences can be widely described, including dynamic systems with correlated time sequence sets in enterprise data. The method can be flexibly combined with various time sequence deduction analysis methods to carry out the joint model deduction analysis of the joint time sequence set.

The invention 4 provides an enterprise data deduction computing system aiming at the problems and challenges of enterprise data deduction computing and stock right value evaluation. The method is realized by adopting the following technical scheme:

an enterprise data deduction computing system comprises a performance trend modeling module, a future sequence computing module, a present value computing module, an internal rate of return computing module, a share right strategy simulation module, an algorithm flow control module, a cloud computing service platform and a data storage module. The performance trend modeling module comprises a time series model modeling module and a simultaneous time series set modeling module. The time sequence model modeling module is used for establishing a future sequence deduction model of historical time sequence data related to each dimension of the enterprise quarterly report profit list, and comprises a state updating model module, a state buffering model module and an ARIMA state model module; the state updating model module is a future sequence deduction model for calculating the time sequence based on the time sequence deduction analysis method of the invention content 1 provided by the invention; the state buffer model module is a future sequence deduction model for calculating a time sequence based on the time sequence deduction analysis method of the invention content 2 provided by the invention; the ARIMA state model module is a future sequence deduction model for calculating the time sequence based on an ARIMA model and an improved time sequence deduction analysis method carried out in combination with the steps 1-3 of the invention content 1 provided by the invention. The simultaneous time sequence set modeling module is used for calculating a joint deduction model of an enterprise simultaneous time sequence set and comprises a common enterprise simultaneous time sequence set modeling module, a bank enterprise simultaneous time sequence set modeling module, a security enterprise simultaneous time sequence set modeling module and an insurance enterprise time sequence set modeling module.

The future sequence calculation module is used for calculating the future sequence within a specific sequence step of the joint deduction model of the enterprise simultaneous time sequence set calculated by the performance trend modeling module.

The present value calculating module is used for calculating infinite step length future sequence present values of each sub-model in the joint deduction model of the enterprise simultaneous time sequence set calculated by the performance trend modeling module according to specific present rates; the module is used for calculating the discount value of the specific finite-step-length future sequence calculated by the future sequence calculation module according to the specific discount rate.

The internal yield rate calculation module is used for calculating the discount rate when the discount value of the infinite step length future sequence calculated by the search current value calculation module is equal to the specific price value; the calculating module is used for calculating the discount rate when the concrete finite-step-length future sequence discount value calculated by the searching module is equal to the concrete price value.

The stock right strategy simulation module comprises a factor value calculation module, a factor logic judgment module, a strategy logic integration module and a transaction simulation module; the factor value calculation module is used for calculating a value factor, a technical factor and a financial factor, wherein the value factor is a factor value obtained or derived based on the calculation result of the current value calculation module or the internal rate of return calculation module; the factor logic judgment module is used for carrying out true and false logic calculation according to the factor value or the factor sequence calculated by the factor value calculation module; the strategy logic integration module is used for integrating the factor logic judgment result calculated by the factor logic judgment module; the transaction simulation module is used for calculating historical simulation transaction operation data and historical simulation transaction income data according to specific transaction type transaction conditions and factor logic judgment integration results calculated by the strategy logic integration module.

The algorithm flow control module is used for transmitting instructions and parameters to the performance trend modeling module, the future sequence calculating module, the present value calculating module, the internal rate of return calculating module and the share right strategy simulating module, arranging a calculating flow and acquiring and processing a calculating result; the cloud computing service platform is used for receiving and processing computing requests of the cloud computing service platform and feeding back computing results; the data storage module is used for requesting data required by calculation to the data storage module and processing data feedback; for requesting external data and handling external data feedback.

The cloud computing service platform is used for providing an interface interaction function and an HTTPS/HTTP API interaction function and comprises an interface function module, an API function module and an interaction control module. The interface function module comprises a performance trend modeling interface function module, a value calculation analysis interface function module and a strategy history simulation interface function module. The performance trend modeling interface function module comprises a state updating model calculation interface function module, a state buffer model calculation interface function module and an ARIMA state model calculation interface function module. The state updating model computing interface function module, the state buffer model computing interface function module and the ARIMA state model computing interface function module respectively comprise a computing parameter template setting function, a specific enterprise specific time simultaneous time sequence set joint deduction model computing function and a specific enterprise specific time simultaneous time sequence set future sequence set deduction computing function. The value calculation analysis interface function module comprises a performance present value calculation interface function module and a performance internal rate of return calculation interface function module. The performance calculation interface functional module comprises a calculation parameter template setting function and a performance calculation function of a specific enterprise in a specific period. The performance internal rate of return calculation interface function module comprises a calculation parameter template setting function and a performance internal rate of return calculation function of a specific enterprise in a specific period. The strategy history simulation interface function module comprises a common strategy simulation interface function module, a financing strategy simulation interface function module and a financing strategy simulation interface function module. The common strategy simulation interface function module, the financing strategy simulation interface function module and the financing strategy simulation interface function module all comprise strategy parameter template setting and editing and specific enterprise strategy simulation calculation functions.

The API function module comprises a state updating model calculation API module, a state buffer model calculation API module, an ARIMA state model calculation API module, a state updating deduction calculation API module, a state buffer deduction calculation API module, an ARIMA state deduction calculation API module, a performance present value calculation API module, a performance internal yield calculation API module and a strategy history simulation API module. The state updating model calculation API module, the state buffering model calculation API module and the ARIMA state model calculation API module all provide a calculation function of a joint deduction model of a joint time sequence set at a specific enterprise specific time period. The state updating deduction calculation API module, the state buffering deduction calculation API module and the ARIMA state deduction calculation API module provide a deduction calculation function of a future sequence set of the specific enterprise specific time simultaneous time sequence set. The performance achievement value calculation API module provides a function of calculating the performance achievement value of a specific enterprise in a specific period. The performance internal rate of return calculation API module provides the performance internal rate of return calculation function of concrete period of concrete enterprise. The strategy history simulation API module provides a strategy simulation calculation function for a specific enterprise at a specific period.

The interaction control module is used for receiving and processing the instruction of setting the template by the interface function module and storing the template setting data into the data storage module; the interface function module is used for receiving and processing a data request of the interface function module, acquiring data from the data storage module and feeding back a data result to the interface function module; and the system is used for receiving and processing the calculation request of the interface function module and the API function module, providing the calculation request to the algorithm flow control module, acquiring the calculation result, processing and converting the calculation result and feeding the calculation result back to the interface function module and the API function module.

The data storage module is used for storing calculation parameter template data set by the state updating model calculation interface function module, the state buffer model calculation interface function module, the ARIMA state model calculation interface function module, the performance present value calculation interface function module, the performance internal yield calculation interface function module and other modules, and storing strategy parameter template data set and edited by the common strategy simulation interface function module, the financing strategy simulation interface function module and other modules.

Compared with the prior art, the enterprise data deduction computing system of invention content 4 provided by the invention has the beneficial technical effects that: the enterprise data deduction calculation can better adapt to the characteristics of complex and variable influence of external factors of enterprise time sequence data and less data sparse quantity, the simultaneous dynamic deduction analysis of the enterprise data time sequence set which is mutually associated is realized, a user is helped to carry out dynamic marketing enterprise equity trading quantitative calculation based on enterprise performance factors, and the automatic quantitative enterprise data deduction calculation function is realized to improve the enterprise data analysis efficiency, the transparency and the objectivity.

Drawings

FIG. 1 is a flow chart of the algorithm associated with inventive concept 1

FIG. 2 is a flow chart of the algorithm associated with step 3.1 of inventive subject matter 1

FIG. 3 is a flow chart of the algorithm associated with step 3.2 of inventive subject matter 1

FIG. 4 is a flow chart of the algorithm associated with step 4.1 of summary of the invention 1

FIG. 5 is a flow chart of the algorithm associated with step 4.2 of summary of the invention 1

FIG. 6 is a flow chart of the algorithm for correlating seasonal time series with the time series of step 5 of inventive content 1

FIG. 7 is a flow chart of the correlation algorithm of the invention in which the time series of step 5 of the summary 1 is a non-seasonal time series

FIG. 8 is a flow chart of the algorithm for correlating seasonal time series with the time series of step 6 of inventive content 1

FIG. 9 is a flow chart of the correlation algorithm of the invention for a non-seasonal time series at step 6 of section 1

FIG. 10 is a flow chart of the correlation algorithm of the invention for step 5 of inventive content 2 with a seasonal time series

FIG. 11 is a flow chart of the correlation algorithm of the invention for step 5 of inventive content 2 with a time series that is a non-seasonal time series

FIG. 12 is a flow chart of the algorithm associated with inventive section 3

FIG. 13 is a schematic diagram of the system operation of inventive concept 4

Detailed Description

In order that the technical solution of the present invention can be easily understood, the present invention will be further explained with reference to the accompanying drawings, which are illustrative of the present invention and the scope of the present invention is not limited thereto.

A. As shown in fig. 1, fig. 2, fig. 3, fig. 4, fig. 5, fig. 6, fig. 7, fig. 8, and fig. 9, the inventive content 1 is a time-series deduction analysis method including 6 steps.

Step 1, calculating a difference value sequence 1 and a difference change rate sequence 2 of a time sequence 0; in the case of analyzing seasonal time series, a quarterly difference value series differentiated by year and a quarterly difference change rate series compared by year are used.

And 2, calculating and obtaining a differential value sequence 3 after the differential value sequence 1 is removed of abnormal samples. The abnormal sample determination step comprises the following steps:

and 2.1, calculating the left and right D quantile values and the median value of the time series 0 in the time range with the sample as the center and the diameter of w in the time series 0.

And 2.2, if the sample data is smaller than the left D quantile value and the absolute value of the difference between the sample data and the left D quantile value is larger than M times of the absolute value of the difference between the left D quantile value and the median value, or the sample data is larger than the right D quantile value and the absolute value of the difference between the sample data and the right D quantile value is larger than M times of the absolute value of the difference between the right D quantile value and the median value, determining that the sample is an abnormal sample.

And 3, identifying the normal trend direction 4 and the normal trend form 5 of the time series 0 in the NS step length. The method specifically comprises the following steps:

and 3.1, identifying the normal trend direction 4 of the time sequence 0 in the NS step length, and using the difference value sequence 1 to remove the difference value sequence 3 after the abnormal sample as shown in figure 2, wherein the significance level is alpha (0< alpha < 0.5). Establishing a virtual hypothesis: the difference value < ═ 0, the opposite assumption: differential value >0, single tailed hypothesis test with t distribution significance level of α:

(1) if the rejection difference value is less than 0, accepting the difference value to be more than 0, and identifying the normal trend direction 4 as increasing;

(2) if the differential value < ═ 0 is not rejected, a virtual hypothesis is established: differential value > -0, contradictory assumption: differential value <0, single tailed hypothesis test with t distribution significance level α:

(2.1) if the reject difference value > is 0, accepting the difference value <0, and recognizing that the normal tendency direction 4 is a fall;

(2.2) if the difference value > is not rejected to 0, the normal tendency direction 4 is identified as random.

The normal tendency direction 4 is output.

Step 3.2, when the normal trend direction 4 is increasing or decreasing, identifying the normal trend form 5 of the time sequence 0 in the NS step size, as shown in fig. 3, the difference value sequence 1 eliminates the difference value sequence 3 of the abnormal sample, and the significance level is set as α (0< α < 0.5):

(1) if the data of the difference value sequence 3 are all positive values or all negative values, calculating a logarithm sequence 3101 of the change rate sequence of the difference value sequence 3, and establishing a virtual hypothesis: logarithm value < ═ 0, contradictory hypothesis: log >0, single tailed hypothesis test with t distribution significance level α:

(1.1) if the rejection logarithm value is 0, accepting the logarithm value >0, and identifying the normal tendency form 5 as an exponential form;

(1.2) if the logarithm value < ═ 0 is not rejected, then a virtual hypothesis is established: logarithm > -0, contradictory hypothesis: for a log value <0, a single tailed hypothesis test was performed with a significant level of t distribution of α:

(1.2.1) if the reject logarithm > is 0, then the opponent hypothesis is accepted: if the logarithm value is less than 0, identifying that the normal trend form 5 is a gentle state;

(1.2.2) if the logarithmic value > is not rejected to be 0, identifying the normal tendency form 5 as a linear form;

(2) if the differential value sequence 3 data are not all positive values and not all negative values, the normal tendency form 5 is identified as a linear form.

The normal tendency form 5 is output.

And 4, identifying the trend direction 7 and the trend form 9 of the current state.

Step 4.1, identifying the current state trend direction 7, and judging as shown in fig. 4:

(1) if the length > of the difference value sequence 1 is equal to the confidence step S, judging:

(1.1) if the last S step of the difference value sequence 1 is positive or negative, the last S step of the difference value sequence 1 constructs a current state difference sequence 6, and starting from n to S, a loop n to n +1, and judging:

(1.1.1) if n < ═ the length of the differential value sequence 1 and the negative and positive of the nth last data and the 1 st last data of the differential value sequence 1 are the same, incorporating the nth last data of the differential value sequence 1 into the differential sequence 6 of the current state, and repeating the loop;

(1.1.2) otherwise, ending the circulation;

after the circulation is finished, judging:

(1.1.a) if the current state differential sequence 6 is positive, identifying the current state trend direction 7 as increasing;

(1.1.b) if the current state differential sequence 6 is negative, identifying the current state trend direction 7 as descending;

(1.2) if the last S steps of the difference value sequence 1 are not both positive and not both negative, the last 1 step of the difference value sequence 1 constructs a current state difference sequence 6, starting from n-1, and cycling n-n +1, judging:

(1.2.1) if n + S-1< ═ difference value sequence 1 length, judge:

(1.2.1.1) if the data of the steps from the n-th to the n + S-1 of the last time of the differential value sequence 1 are not positive and are not negative, incorporating the data of the n-th of the last time of the differential value sequence 1 into the differential sequence 6 of the current state, and repeating the cycle;

(1.2.1.2) otherwise, ending the cycle;

(1.2.2) if n + S-1> length of the differential value sequence 1, using the differential value sequence 1 as a current state differential sequence 6, and ending the cycle;

after the circulation is finished, judging:

(1.2.a) if the current state difference sequence 6 length > is confidence step S, establishing a virtual hypothesis: the difference value < ═ 0, the opposite assumption: differential value >0, single tailed hypothesis test with t distribution significance level of α:

(1.2.a.1) identifying the current state trend direction 7 as increasing if the difference value < ═ 0 is rejected;

(1.2.a.2) if the differential value < ═ 0 is not rejected, establish a virtual hypothesis: differential value > -0, contradictory assumption: differential value <0, single tailed hypothesis test with t distribution significance level α:

(1.2.a.2.1) identifying the current state trend direction 7 as down if the reject difference value > is 0;

(1.2.a.2.2) if the differential value > is not rejected to be 0, identifying the current state trend direction 7 as random;

(1.2.b) if the length of the current state differential sequence 6 is less than the confidence step S, identifying the current state trend direction 7 as random;

(2) if the length of the differential value sequence 1 is < confidence step S, the current state trend direction 7 is identified as random.

The current state trend direction 7 is output.

Step 4.2, when the current state difference sequences 6 are both positive or both negative, identifying the current state trend form 9, as shown in fig. 5, calculating a logarithm sequence 4201 of the difference change rate sequence 2 in the time period of the current state difference sequence 6, and judging:

(1) if the length of the log sequence 4201 is equal to the confidence step S, judging:

(1.1) if the last S step data of the log sequence 4201 is positive or negative, the difference change rate sequence 2 builds the current state difference change rate 8 in the last S step, and starting from n-S, a loop n-n +1, and judging:

(1.1.1) if n < ═ log sequence 4201 length and the nth last data of log sequence 4201 is the same positive or negative as the 1 st last data, then including the nth last data of differential rate of change sequence 2 into the differential rate of change 8 of the current state, repeating the loop;

(1.1.2) otherwise, ending the circulation;

after the circulation is finished, judging:

(1.1.a) if the differential change rates 8 of the current state are both greater than 1, identifying the trend form 9 of the current state as an exponential state;

(1.1.b) if the current state differential change rates 8 are all smaller than 1, identifying the current state trend form 9 as a gentle state;

(1.2) if the last S step data of the log sequence 4201 are not both positive and not both negative, constructing a current state differential change rate 8 in the last 1 step of the differential change rate sequence 2, starting from n-1, and a loop n-n +1, judging:

(1.2.1) if n + S-1< ═ log sequence 4201 length, then judge:

(1.2.1.1) if the steps from the n-th to the n + S-1-th of the last of the log sequence 4201 are not positive and are not negative, the n-th data of the difference change rate sequence 2 is included in the difference change rate 8 of the current state, and the cycle is repeated;

(1.2.1.2) otherwise, ending the cycle;

(1.2.2) if n + S-1> the length of the log sequence 4201, using the differential change rate sequence 2 as the current state differential change rate 8, and ending the loop;

after the circulation is finished, judging:

(1.2.a) if the current state differential change rate 8 length > is the signal step S, then using the log sequence 4202 of the current state differential change rate 8, a virtual hypothesis is established: logarithm value < ═ 0, contradictory hypothesis: log >0, single tailed hypothesis test with t distribution significance level α:

(1.2.a.1) identifying the current state trend shape 9 as an exponential shape if the rejection logarithm value is 0;

(1.2.a.2) if the logarithm value < ═ 0 is not rejected, then a virtual hypothesis is established: logarithm > -0, contradictory hypothesis: for a log value <0, a single tailed hypothesis test was performed with a significant level of t distribution of α:

(1.2.a.2.1) if the rejection logarithm > is 0, identifying that the current state trend form 9 is a gentle state;

(1.2.a.2.2) if the logarithm value > is not rejected to be 0, identifying that the current state trend form 9 is a linear state;

(1.2.b) identifying the current state trend morphology 9 as a linear morphology if the current state differential difference change rate 8 length < confidence step S.

(2) If the log sequence 4201 length < confidence step S, then the current state trend morphology 9 is identified as linear.

The current state trend morphology 9 is output.

Step 5, calculating a time series 0 model deduction parameter in the MS step length, establishing a future sequence deduction model 10, and judging:

(1) if time series 0 is a seasonal time series, as shown in FIG. 6, it is determined that:

(1.1) if the trend direction 7 of the current state is random, backtracking the current state differential sequence 6 by 4 quarters to be used as a current state time period 501, calculating the time weight mean of all quarter subsequences in the time period 501 part of the time sequence 0 according to the annealing rate anr to be a quarter parameter 502, and the future sequence deduction model 10 is as follows:

each quarterly value is the current quarterly value of quarterly parameter 502;

(1.2) if the trend direction 7 of the current state is not random, namely increasing or decreasing, judging:

(1.2.1) if the current state trend morphology 9 is not identified, calculating the time weight mean 503 of each quaternary subsequence of the current state difference sequence 6 according to the annealing rate anr, and judging:

(1.2.1.1) if the time weight mean 503 does not conform to the current state trend direction 7, the current state differential sequence 6 is traced back for 4 quarters as the current state time period 501, the time weight mean of each quarter subsequence in the time period 501 part of the time sequence 0 is calculated as the quarter parameter 502 according to the annealing rate anr, and the future sequence deduction model 10 is:

each quarterly value is the current quarterly value of quarterly parameter 502;

(1.2.1.2) if the time weight mean 503 coincides with the current state trend direction 7, the last sequence step value of each quarter of the time series 0 is the quarter parameter 502, the time weight mean of each quarter subsequence of the current state difference series 6 is calculated as the quarter difference parameter 504 according to the annealing rate anr, and the future sequence deduction model 10 is:

each quarterly difference is the current quarterly value of quarterly difference parameter 504;

each quarterly quarter value is the current quarter value of the quarterly value parameter 502+ the cumulative value of the future sequence current quarter differential subsequence up to the season;

(1.2.2) if the current state trend form 9 is a linear state, the last sequence step value of each quarter of the time series 0 is a quarter parameter 502, the current state difference change rate series 8 is traced back for 4 quarters to be a current state time period 501, the time weight mean value of each quarter subsequence in the time period 501 part of the current state difference series 6 is calculated as a quarter difference parameter 504 according to the annealing rate anr, and the future sequence deduction model 10 is as follows:

each quarterly difference is the current quarterly value of quarterly difference parameter 504;

each quarterly quarter value is the current quarter value of the quarterly value parameter 502+ the cumulative value of the future sequence current quarter differential subsequence up to the season;

(1.2.3) if the current trend form 9 is not a linear form, namely an exponential form or a flat form, calculating the time weight mean 505 of the quaternary subsequences of the current state difference change rate sequence 8 according to the annealing rate anr, and judging:

(1.2.3.1) if the time weight mean 505 does not match the current state trend morphology 9, the last sequence step value of each quarter of the time series 0 is the quarter parameter 502, the current state difference change rate series 8 is traced back for 4 quarters to be the current state time period 501, the time weight mean of each quarter subsequence in the time period 501 part of the current state difference series 6 is calculated as the quarter difference parameter 504 according to the annealing rate anr, and the future sequence deduction model 10 is:

each quarterly difference is the current quarterly value of quarterly difference parameter 504;

each quarterly quarter value is the current quarter value of the quarterly value parameter 502+ the cumulative value of the future sequence current quarter differential subsequence up to the season;

(1.2.3.2) if the time-weighted mean 505 is consistent with the current-state trend morphology 9, the last-quarter-sequence step values of the time series 0 are the quarter parameter 502, the last-quarter-sequence step values of the difference value series 1 are the quarter difference parameter 504, the time-weighted mean of the current-state difference rate-of-change series 8 is calculated as the quarter rate-of-change parameter 506 according to the annealing rate anr, and the future sequence deduction model 10 is:

each quarterly change rate is the quarterly change rate parameter 506 which is the quarterly value;

each quarterly difference parameter 504 quarterly difference parameter is a quarterly difference parameter, and a quarterly difference parameter is a cumulative multiplication value of a future sequence quarterly change rate subsequence up to the season;

each quarterly quarter value is the current quarter value of the quarterly value parameter 502+ the cumulative value of the future sequence current quarter differential subsequence up to the season;

(2) if time series 0 is a non-seasonal time series, as shown in FIG. 7, it is determined that:

(2.1) if the trend direction 7 of the current state is random, taking the sequence step of backtracking 1 of the differential sequence 6 of the current state as the time period 501 of the current state, calculating the time weight average value of the time period 501 part of the time sequence 0 as the value parameter 502 according to the annealing rate anr, and the future sequence deduction model 10 is as follows:

each sequence step value is a value parameter 502;

(2.2) if the trend direction 7 of the current state is not random, namely increasing or decreasing, judging:

(2.2.1) if the current state trend form 9 is not identified, calculating the time weight mean 503 of the current state difference sequence 6 according to the annealing rate anr, and judging:

(2.2.1.1) if the time weight mean 503 does not accord with the current state trend direction 7, the current state difference sequence 6 backtracking 1 sequence step is taken as the current state time period 501, the time weight mean of the time sequence 0 time period 501 part is calculated as the value parameter 502 according to the annealing rate anr, and the future sequence deduction model 10 is:

each sequence step value is a value parameter 502;

(2.2.1.2) if the time-weighted mean 503 coincides with the current-state trend direction 7, the time-series 0 last-series step value is the value parameter 502, the time-weighted mean of the current-state difference series 6 is calculated as the difference parameter 504 according to the annealing rate anr, and the future sequence deduction model 10 is:

difference parameter 504 is obtained for each sequence step;

each sequence step value is the value parameter 502+ the accumulated value of the difference sequence of the future sequence until the step;

(2.2.2) if the current state trend form 9 is a linear state, the last sequence step value of the time series 0 is a value parameter 502, the sequence step of the current state difference change rate series 8 backtracking 1 is taken as a current state time period 501, the time weight mean value of the current state difference series 6 time period 501 part is calculated according to the annealing rate anr to be a difference parameter 504, and the future sequence deduction model 10 is as follows:

difference parameter 504 is obtained for each sequence step;

each sequence step value is the value parameter 502+ the accumulated value of the difference sequence of the future sequence until the step;

(2.2.3) if the current trend form 9 is not a linear form, namely an exponential form or a flat form, calculating the time weight mean 505 of the current state difference change rate sequence 8 according to the annealing rate anr, and judging:

(2.2.3.1) if the time-weighted mean 505 does not match the current-state trend morphology 9, the last sequence step of the time series 0 is the value parameter 502, the sequence step of the current-state differential change rate series 8 going back to 1 is the current-state time period 501, the time-weighted mean of the current-state differential series 6 time period 501 is calculated as the difference parameter 504 according to the annealing rate anr, and the future sequence deduction model 10 is:

difference parameter 504 is obtained for each sequence step;

each sequence step value is the value parameter 502+ the accumulated value of the difference sequence of the future sequence until the step;

(2.2.3.2) if the time-weighted mean 505 matches the current-state trend 9, the last sequence step of the time series 0 is the value parameter 502, the last sequence step of the difference value series 1 is the difference parameter 504, the time-weighted mean of the current-state difference change rate series 8 is calculated as the change rate parameter 506 according to the annealing rate anr, and the future sequence deduction model 10 is:

the rate of change of each sequence step is the rate of change parameter 506;

difference parameter 504 is the cumulative value of future sequence rate of change sequence to the step;

each sequence step value is the value parameter 502+ the accumulated value of the future sequence difference sequence up to that step.

And outputting the future sequence deduction model 10 and deduction parameters.

Step 6, performing inertial processing on the future sequence deduction model 10, and judging:

(1) if time series 0 is a seasonal time series, as shown in FIG. 8, it is determined that:

(1.1) if the current state trend direction 7 is not consistent with the normal state trend direction 4, the model and the parameters of the future sequence deduction model 10 are kept unchanged within the inertia year IY, the last quarter value of each quarter value sequence of the future sequence within the inertia year IY is calculated as a quarter value parameter 601 after the inertia year IY, and after the inertia year IY of the future sequence deduction model 10:

each quarterly value is the current quarterly value of quarterly parameter 601;

(1.2) if the current state trend direction 7 is consistent with the normal state trend direction 4, judging:

(1.2.1) if the current state trend form 9 is not consistent with the normal state trend form 5, the model and parameters of the future sequence deduction model 10 are kept unchanged within the IY of the inertial number, the last quarter value of each quarter value sequence of the future sequence within the IY of the inertial number is calculated as a quarter value parameter 601 after the IY of the inertial number, the last quarter difference of each quarter difference sequence of the future sequence within the IY of the inertial number is calculated as a quarter difference parameter 602 after the IY of the inertial number, and after the IY of the future sequence deduction model 10:

each quarterly difference is the current quarterly value of quarterly difference parameter 602;

after the current season value + the inertial number IY of each quartering value parameter 601, cutting the current season difference subsequence of the future sequence to the accumulated value of the season;

(1.2.2) if the current state trend form 9 is consistent with the normal state trend form 5, keeping the future sequence deduction model 10 and the parameters unchanged;

(2) if time series 0 is a non-seasonal time series, as shown in FIG. 9, it is determined that:

(2.1) if the current state trend direction 7 is not consistent with the normal state trend direction 4, the model and the parameters of the future sequence deduction model 10 are kept unchanged within the inertia year number IY, the step value of the final sequence of the future sequence within the inertia year number IY is calculated as a post-inertia year number IY value parameter 601, and after the inertia year number IY of the future sequence deduction model 10:

each sequence step value is 601;

(2.2) if the current state trend direction 7 is consistent with the normal state trend direction 4, judging:

(2.2.1) if the current state trend form 9 is not consistent with the normal state trend form 5, the model and parameters of the future sequence deduction model 10 are kept unchanged within the inertia year IY, the step value of the future sequence within the inertia year IY is calculated as a post-inertia year IY value parameter 601, the step value of the future sequence differential sequence within the inertia year IY is calculated as a post-inertia year IY differential parameter 602, and after the inertia year IY of the future sequence deduction model 10:

each sequence step difference is a difference parameter 602;

after the step value of each sequence is equal to the + inertia year IY of the value parameter 601, the difference sequence of the future sequence is up to the accumulated value of the step;

(2.2.2) if the current state trend morphology 9 is consistent with the normal state trend morphology 5, the future sequence deduction model 10 and the parameters are kept unchanged.

Outputting the future sequence deduction model 10 after inertia processing.

The future sequence deduction model 10 and the parameters are output.

B. The invention content 2 is a time series deduction analysis method, which includes 6 steps, specifically, step 1, step 2, step 3, step 4, step 6 are the same as step 1, step 2, step 3, step 4, and step 6 of the time series deduction analysis method of the invention content 1, and step 5 calculates deduction parameters of a time series 0 model and establishes a future series deduction model 10 as shown in fig. 10 and fig. 11, and the judgment is made:

(1) if time series 0 is a seasonal time series, as shown in FIG. 10, it is determined that:

(1.1) if the trend direction 7 of the current state is random, calculating the time weight mean of each quaternary subsequence in the last RS step of the time series according to the annealing rate anr as a quaternary parameter 502, and the future sequence deduction model 10 is as follows:

each quarterly quarter value is the current quarter value of quarterly parameter 502

(1.2) if the trend direction 7 of the current state is not random, namely increasing or decreasing, judging:

(1.2.1) if the current state trend form 9 is not identified, calculating the time weight mean 503 of each quaternary subsequence in the last RS step of the difference value sequence 1 according to the annealing rate anr, and judging:

(1.2.1.1) if the time-weighted mean 503 does not match the current-state trend direction 7, then the time-weighted mean of each quarter subsequence in the last RS step of the time series is calculated as the quarter parameter 502 according to the annealing rate anr, and the future sequence deduction model 10 is:

each quarterly quarter value is the current quarter value of quarterly parameter 502

(1.2.1.2) if the time-weighted mean 503 coincides with the current-state trend direction 7, the last sequence step of each quarter of time series 0 is taken as a quarter parameter 502, the time-weighted mean of each quarter subsequence in the last RS step of the differential value series 1 is calculated as a quarter difference parameter 504 according to the annealing rate anr, and the future sequence deduction model 10 is:

each quarterly difference is the current quarterly value of quarterly difference parameter 504;

each quarterly quarter value is the current quarter value of the quarterly value parameter 502+ the cumulative value of the future sequence current quarter differential subsequence up to the season;

(1.2.2) if the current state trend form 9 is a linear state, the last sequence step value of each quarter of the time series 0 is a quarter parameter 502, the time weight mean value of each quarter subsequence in the last RS step of the difference value sequence 1 is calculated as a quarter difference parameter 504 according to the annealing rate anr, and the future sequence deduction model 10 is as follows:

each quarterly difference is the current quarterly value of quarterly difference parameter 504;

each quarterly quarter value is the current quarter value of the quarterly value parameter 502+ the cumulative value of the future sequence current quarter differential subsequence up to the season;

(1.2.3) if the current trend form 9 is not a linear form, namely an exponential form or a flat form, calculating the time weight mean 505 of each quaternary subsequence in the last RS step of the differential rate-of-change sequence 2 according to the annealing rate anr, and judging:

(1.2.3.1) if the time-weighted mean 505 does not match the current-state trend morphology 9, the last-quarter-sequence step value of time series 0 is the quarter parameter 502, the time-weighted mean of the quarter subsequences in the last RS step of differential-value series 1 is calculated as the quarter difference parameter 504 according to the annealing rate anr, and the future sequence deduction model 10 is:

each quarterly difference is the current quarterly value of quarterly difference parameter 504;

each quarterly quarter value is the current quarter value of the quarterly value parameter 502+ the cumulative value of the future sequence current quarter differential subsequence up to the season;

(1.2.3.2) if the time-weighted mean 505 matches the current state trend morphology 9, then the last sequence step values in each quarter of time series 0 are the quarter parameter 502, the last sequence step values in each quarter of difference value series 1 are the quarter difference parameter 504, the time-weighted mean of each quarter subsequence in the last RS step of difference rate of change series 2 is calculated as the quarter rate of change parameter 506 according to the annealing rate anr, and the future sequence deduction model 10 is:

each quarterly change rate is the quarterly change rate parameter 506 which is the quarterly value;

each quarterly difference parameter 504 quarterly difference parameter is a quarterly difference parameter, and a quarterly difference parameter is a cumulative multiplication value of a future sequence quarterly change rate subsequence up to the season;

each quarterly quarter value is the current quarter value of the quarterly parameter 502+ the cumulative value of the future sequence quarterly differential subsequence up to the season

(2) If time series 0 is a non-seasonal time series, as shown in FIG. 11, it is determined that:

(2.1) if the trend direction 7 of the current state is random, calculating the time weight mean value in the last RS step of the time series according to the annealing rate anr as a value parameter 502, and the future sequence deduction model 10 is as follows:

value parameter 502 for each sequence step value

(2.2) if the trend direction 7 of the current state is not random, namely increasing or decreasing, judging:

(2.2.1) if the current state trend form 9 is not identified, calculating a time weight mean 503 in the last RS step of the difference value sequence 1 according to the annealing rate anr, and judging:

(2.2.1.1) if the time weight mean 503 does not match the current state trend direction 7, then the time weight mean in the last RS step of the time series is calculated as the value parameter 502 according to the annealing rate anr, and the future series deduction model 10 is:

value parameter 502 for each sequence step value

(2.2.1.2) if the time weight mean 503 is consistent with the current state trend direction 7, the value parameter 502 is the last sequence step value of the time series 0, the time weight mean in the last RS step of the difference value sequence 1 is calculated as the difference parameter 504 according to the annealing rate anr, and the future sequence deduction model 10 is:

difference parameter 504 is obtained for each sequence step;

if the current state trend form 9 is a linear state, the last sequence step value of the time sequence 0 is the value parameter 502, the time weight mean value in the last RS step of the difference value sequence 1 is calculated according to the annealing rate anr as the difference parameter 504, and the future sequence deduction model 10 is:

difference parameter 504 is obtained for each sequence step;

value parameter 502+ accumulated value of future sequence difference sequence up to step

(2.2.3) if the current trend form 9 is not a linear form, namely an exponential form or a flat form, calculating a time weight mean value 505 in the last RS step of the difference change rate sequence 2 according to the annealing rate anr, and judging:

(2.2.3.1) if the time weight mean 505 does not match the current state trend profile 9, the time series 0 last sequence step value is the value parameter 502, the time weight mean in the last RS step of the difference value series 1 is calculated as the difference parameter 504 according to the annealing rate anr, and the future sequence deduction model 10 is:

difference parameter 504 is obtained for each sequence step;

value parameter 502+ accumulated value of future sequence difference sequence up to step

(2.2.3.2) if the time-weighted mean 505 matches the current state trend profile 9, the time series 0 last sequence step value is the value parameter 502, the difference value series 1 last sequence step value is the difference parameter 504, the time-weighted mean in the difference rate of change series 2 last RS step is calculated as the rate of change parameter 506 according to the annealing rate anr, and the future sequence deduction model 10 is:

the rate of change of each sequence step is the rate of change parameter 506;

difference parameter 504 is the cumulative value of future sequence rate of change sequence to the step;

value parameter 502+ accumulated value of future sequence difference sequence up to step

And outputting the future sequence deduction model 10 and deduction parameters.

C. In the invention section 3, as shown in fig. 12, a joint-time sequence set deduction analysis method includes 6 steps.

Step 1, dividing a time sequence set with an association relationship into a self-increment time sequence, a driving time sequence, a quota time sequence and a composite time sequence to form a simultaneous time sequence set.

Step 2, setting a self-increment time sequence number A, starting from the time sequence i being 1, circularly calculating a future sequence deduction model 2 according to the self-increment time sequence i 1, defining the future sequence deduction model 2 as a self-increment dimension model 3, and judging:

(1) if n < A, i ═ i +1 repeats the cycle;

(2) otherwise, the loop is ended.

Step 3, setting a driving time sequence number D, circularly defining a traction dimension time sequence 5 of an ith driving time sequence 4 from i to 1, calculating a ratio sequence 6 of the driving time sequence 4 and the traction dimension time sequence 5, calculating a future sequence deduction model 7 according to the ratio sequence 6, constructing a driving dimension model 9 according to the product of the model 7 and a future sequence step value of a future sequence deduction model 8 of the traction dimension time sequence 5, defining the model 9 as the future sequence deduction model of the driving time sequence 4, and judging:

(1) if n < D, i ═ i +1 repeats the cycle;

(2) otherwise, the loop is ended.

Step 4, setting a quota time sequence number Q, circularly defining a reference dimension time sequence 11 and a quota 12 of an ith quota time sequence 10 from the moment that i is equal to 1, constructing a quota dimension model 13 according to the product of a future sequence step value of a future sequence deduction model of the reference dimension time sequence 11 and the quota 12, defining the model 13 as the future sequence deduction model of the quota time sequence 10, and judging:

(1) if n < Q, i +1 repeats the cycle;

(2) otherwise, the loop is ended.

Step 5, setting the number C of the composite time sequences, starting from i equal to 1, defining an dimension-added time sequence set 15 and a dimension-subtracted time sequence set 16 of the ith composite time sequence 14, constructing a composite dimension model 17 according to the difference between the sum of the future sequences of all time sequences of the dimension-added time sequence set 15 and the sum of the future sequences of all time sequences of the dimension-subtracted time sequence set 16, defining the model 17 as a future sequence deduction model of the composite time sequence 14, and judging:

(1) if n < C, i ═ i +1 repeats the cycle;

(2) otherwise, the loop is ended.

And 6, jointly constructing a joint deduction model 18 for constructing a simultaneous time sequence set by using all the self-increment dimensional model 3, the driving dimensional model 9, the quota dimensional model 13 and the composite dimensional model 17.

The joint deduction model 18 is output.

D. The invention content 4 is that as shown in fig. 13, an enterprise data deduction computing system comprises a performance trend modeling module 1, a future sequence computing module 2, a present value computing module 3, an internal rate of return computing module 4, a share right strategy simulation module 6, an algorithm flow control module 7, a cloud computing service platform 7 and a data storage module 8.

The performance trend modeling module 1 includes a time series model modeling module 11 and a simultaneous time series set modeling module 12. The time series model modeling module 11 is used for establishing a future series deduction model of historical time series data related to each dimension of the enterprise quarterly report profit list, and comprises a state updating model module 111, a state buffering model module 112 and an ARIMA state model module 113; the state updating model module 111 is a future sequence deduction model for calculating a time sequence based on the time sequence deduction analysis method of the invention content 1 provided by the present invention; the state buffer model module 112 is a future sequence deduction model for calculating a time sequence based on the time sequence deduction analysis method of the invention content 2 provided by the present invention; the ARIMA state model module 113 is a future sequence deduction model for calculating a time sequence based on an ARIMA model in combination with the improved time sequence deduction analysis method performed in steps 1-3 of the invention 1 provided by the present invention. Each submodule of the time series model modeling module 11 calculates a future series deduction model and parameters for providing a time series according to instructions and parameters transmitted by the algorithm flow control module 6.

The simultaneous time series set modeling module 12 is a joint deduction model for calculating a simultaneous time series set of an enterprise, and includes a common enterprise simultaneous time series set modeling module 121, a bank enterprise simultaneous time series set modeling module 122, a securities enterprise simultaneous time series set modeling module 123, and an insurance enterprise time series set modeling module 124. And each submodule of the simultaneous time sequence set modeling module 12 calculates and provides a joint deduction model and parameters of the simultaneous time sequence set by combining a future sequence deduction model and parameters of the time sequence which are calculated and provided by the time sequence model modeling module 11 according to instructions and parameters which are transmitted by the algorithm flow control module 6.

The future sequence calculation module 2 directly or indirectly obtains the joint deduction model and parameters of the enterprise simultaneous time sequence set calculated by the performance trend modeling module 1 according to instructions and parameters transmitted by the algorithm flow control module 6, and calculates and provides a future sequence within a specific sequence step length. The present value calculation module 3 directly or indirectly obtains the joint deduction model and parameters of the enterprise simultaneous time sequence set calculated by the performance trend modeling module 1 according to instructions and parameters transmitted by the algorithm flow control module 6, and calculates and provides the discount values of infinite step length future sequences of each sub model in the joint deduction model of the simultaneous time sequence set; or, directly or indirectly obtaining the future sequence in the specific sequence step length calculated by the future sequence calculation module 2, and calculating and providing the discount value of the specific finite step length future sequence of each sub-model in the simultaneous time sequence set joint deduction model.

The internal yield calculation module 4 directly or indirectly obtains the future sequence discount value calculated by the present value calculation module 3 according to the instruction and the parameter transmitted by the algorithm flow control module 6, and searches and calculates the discount rate when the discount value of each sub-model future sequence in the joint deduction model of the simultaneous time sequence set is equal to the specific price value.

The stock right strategy simulation module 5 comprises a factor value calculation module 51, a factor logic judgment module 52, a strategy logic integration module 53 and a transaction simulation module 54.

The factor value calculating module 51 calculates and provides a value factor, a technical factor and a financial factor according to the instructions and parameters transmitted by the algorithm flow control module 6, wherein the value factor is a factor value obtained or derived according to the result provided by the calculation of the directly or indirectly acquired present value calculating module 3 or the internal rate of return calculating module 4.

The factor logic judgment module 52 directly or indirectly obtains the value factor, technical factor or financial factor calculated by the factor value calculation module 51 according to the instruction and parameter transmitted from the algorithm flow control module 6, and calculates and provides the true and false logic judgment result of the factor value or factor sequence.

The policy logic integration module 53 integrates all the true and false logic determination results of the factor values or the factor sequences calculated by the factor logic determination module 52 according to the instructions and parameters transmitted from the algorithm flow control module 6, and provides a logic integration result by calculation.

The transaction simulation module obtains a logic integration result calculated by the policy logic integration module 53 according to instructions and parameters, including specific transaction type transaction conditions, input by the algorithm flow control module 6, and calculates and provides historical simulation transaction operation data and historical simulation transaction income data.

The algorithm flow control module 6 is used for transmitting instructions and parameters to the performance trend modeling module 1, the future sequence calculation module 2, the present value calculation module 3, the internal rate of return calculation module 4 and the stock right strategy simulation module 5, arranging a calculation flow, and acquiring and processing a calculation result; the cloud computing service platform 7 is used for receiving and processing computing requests of the cloud computing service platform 7 and feeding back computing results; for requesting the data storage module 8 for the template data required for the calculation and processing the data feedback; for requesting external data and handling external data feedback.

The cloud computing service platform 7 is configured to provide an interface interaction function and an HTTPS/HTTP API interaction function, and includes an interface function module 71, an API function module 72, and an interaction control module 73.

The interface function module 71 includes a performance trend modeling interface function module 711, a value calculation analysis interface function module 712, and a policy history simulation interface function module 713.

The performance trend modeling interface function 711 includes a state update model calculation interface function 7111, a state buffer model calculation interface function 7112, and an ARIMA state model calculation interface function 7113. The state updating model calculation interface function module 7111, the state buffer model calculation interface function module 7112 and the ARIMA state model calculation interface function module 7113 all include a calculation parameter template setting function, a specific enterprise specific time period simultaneous time sequence set joint deduction model calculation function and a specific enterprise specific time period simultaneous time sequence set future sequence set deduction calculation function.

The value calculation analysis interface function 712 includes a performance present value calculation interface function 7121 and a performance internal rate of return calculation interface function 7122. The performance achievement value calculation interface function 7121 includes a calculation parameter template setting function and a performance achievement value calculation function for a particular time period of a particular business. The performance internal rate of return calculation interface function 7122 includes a calculation parameter template setting function and a specific enterprise specific time period performance internal rate of return calculation function.

The strategy history simulation interface function module 713 comprises a common strategy simulation interface function module 7131, a financing strategy simulation interface function module 7132 and a financing strategy simulation interface function module 7133. The common strategy simulation interface function module 7131, the financing strategy simulation interface function module 7132 and the financing strategy simulation interface function module 7133 all comprise the strategy parameter template setting and editing and the strategy simulation calculation function of the specific enterprise period.

The API function module 72 includes a status update model calculation API module 721, a status buffer model calculation API module 722, an ARIMA status model calculation API module 723, a status update deduction calculation API module 724, a status buffer deduction calculation API module 725, an ARIMA status deduction calculation API module 726, a performance value calculation API module 727, a performance internal yield calculation API module 728, and a policy history simulation API module 729.

The state updating model calculation API module 721, the state buffer model calculation API module 722 and the ARIMA state model calculation API module 723 all provide a joint deduction model calculation function of a specific enterprise specific time simultaneous time sequence set.

The state update deduction calculation API module 724, the state buffer deduction calculation API module 725 and the ARIMA state deduction calculation API module 726 each provide enterprise specific time period simultaneous time sequence set future sequence set deduction calculation functions.

The performance value calculation API module 727 provides performance value calculation functions for specific time periods for a particular business. The performance internal rate of return calculation API module 728 provides performance internal rate of return calculation functionality for a particular time period for a particular business. The policy history simulation API module 729 provides enterprise-specific time policy simulation calculation functions.

The interaction control module 73 is used for receiving a template setting instruction of the processing interface function module 71 and storing template setting data into the data storage module 8; the data processing module is used for receiving a data request of the processing interface functional module 71, acquiring data from the data storage module 8 and feeding back a data result to the interface functional module 71; and the processing module is used for receiving and processing the calculation requests of the interface function module 71 and the API function module 72, sending the calculation requests to the algorithm flow control module 6, obtaining the calculation results, processing and converting the calculation results, and feeding the calculation results back to the interface function module 71 and the API function module 72.

The data storage module 8 is configured to store calculation parameter template data set by the modules such as the state update model calculation interface function module 7111, the state buffer model calculation interface function module 7112, the ARIMA state model calculation interface function module 7113, the performance value calculation interface function module 7121, the performance internal rate of return calculation interface function module 7122, and the like, and store policy parameter template data set and edited by the modules such as the general policy simulation interface function module 7131, the financing policy simulation interface function module 7132, the financing policy simulation interface function module 7133, and the like.

The enterprise data deduction computing system provided by the invention can automatically carry out enterprise simultaneous time series set data deduction computing under the conditions of complex and variable external factor influence, complex and compact internal association relation and sparse samples in a high-efficiency manner based on the time series deduction analysis method and the simultaneous time series quartering analysis method provided by the invention, has stronger adaptability compared with an automatic ARIMA algorithm and a neural network algorithm, greatly saves computing resources and realizes dynamic quantitative analysis based on enterprise performance.