阅读说明：本技术 台区用户辨识与电压影响评估方法 (Transformer area user identification and voltage influence evaluation method ) 是由 熊文 李欣 刘艳萍 曾顺奇 吴杰康 蔡志宏 于 2021-07-30 设计创作，主要内容包括：本发明公开了一种台区用户辨识与电压影响评估方法,包括：构建台区用户有功值的数据矩阵以及变压器低侧电压时间序列矩阵；对所述数据矩阵进行预处理,得到降维后的数据；基于所述数据矩阵和时间序列矩阵,计算出每一位用户与低压台区电压波动之间的关系,以进行台户关联因子的构建；以隶属度为欧式距离建立目标函数,以皮尔逊相关系数和隶属度和为约束条件进行聚类,从而得到在特定影响因子范围内,不同用户等级的影响程度分类情况,继而得到台户之间用户有功值与台区变压器低压侧电压值之间的关系,根据所述关系进行影响评估。本方法对于认识配电网拓扑、建设智能电网具有重要意义,为实行进一步电能质量管理提供必要的技术支撑。(The invention discloses a transformer area user identification and voltage influence evaluation method, which comprises the following steps: constructing a data matrix of active values of users in a transformer area and a time sequence matrix of low-side voltage of a transformer; preprocessing the data matrix to obtain data after dimensionality reduction; based on the data matrix and the time sequence matrix, calculating the relation between each user and the voltage fluctuation of the low-voltage transformer area so as to construct a transformer association factor; and establishing a target function by taking the membership as an Euclidean distance, and clustering by taking the Pearson correlation coefficient and the membership sum as constraint conditions, thereby obtaining the classification condition of the influence degrees of different user grades in a specific influence factor range, further obtaining the relation between the user success value between the users and the voltage value of the low-voltage side of the transformer in the transformer area, and carrying out influence evaluation according to the relation. The method has important significance for knowing the topology of the power distribution network and building the smart power grid, and provides necessary technical support for further power quality management.)

1. A method for identifying users in a distribution room and evaluating voltage influence is characterized by comprising the following steps:

acquiring active power data of a user side at a set resolution and time point, and constructing a data matrix of active values of users in a distribution area; acquiring a time sequence active value under the transformer area voltage in a unidirectional way, and constructing a time sequence matrix of the voltage at the low side of the transformer;

preprocessing the data matrix by adopting a multi-dimensional scaling method to obtain data after dimension reduction;

based on the data matrix and the time sequence matrix, calculating the relation between each user and the voltage fluctuation of the low-voltage transformer area so as to construct a transformer association factor;

and establishing a target function by taking the membership as an Euclidean distance, and clustering by taking the Pearson correlation coefficient and the membership sum as constraint conditions, thereby obtaining the classification condition of the influence degrees of different user grades in a specific influence factor range, further obtaining the relation between the user success value between the users and the voltage value of the low-voltage side of the transformer in the transformer area, and carrying out influence evaluation according to the relation.

2. The method of claim 1, wherein the preprocessing the data matrix by using a multidimensional scaling method to obtain dimension-reduced data comprises:

for m users, each user acquires D-dimensional data, and calculates a distance matrix B epsilon R in the original space^m×DIts ith row and j column element dist_ijFor any one user sample x_iTo x_jThe goal is to obtain a representation matrix Z ∈ R in d' dimensional space^m×d′D 'is less than or equal to D, and the Euclidean distance of any two samples in the D' dimensional space is equal to the distance of the original space: | | z_i-z_j||＝dist_ij，z_i,z_jRepresenting the reduced dimension sample, sample x_i、x_jAs a data matrix X_pThe user data in (1);

let E be Z^TZ∈R^m×mAnd decentralizing the matrix Z after dimension reduction, simplifying the matrix Z by combining decentralization constraint conditions, calculating E, and performing characteristic decomposition on the E to obtain E ═ eta^TWherein Λ ═ diag [ λ ═ d [ lambda ] ]₁,λ₂,…,λ_n]A diagonal matrix formed by the eigenvalues, wherein eta is an eigenvector matrix; selecting an eigenvector matrix W [ [ eta ] η ] corresponding to the first 95% contribution eigenvalue according to the contribution of the eigenvalue₁,η₂,…,η_d′]And decomposing the characteristic of the E matrix to obtain a characteristic value matrix, and sorting the characteristic value matrix according to the first d' numbers from large to small: lambda [ alpha ]₁≥λ₂≥…≥λ_q≥λ_d′To obtain the finalAnd D, reducing the dimension of the matrix Z.

3. The method of claim 2, wherein the value of d' is selected according to the following expression of contribution degree:

4. the method of claim 1, wherein the calculating the relationship between each user and the voltage fluctuation of the low voltage transformer area comprises:

and sequentially calculating the overall mean value and the overall covariance between each user and the voltage fluctuation of the low-voltage area, and then calculating to obtain the overall Pearson correlation coefficient.

5. The method of claim 1, wherein the step of constructing the fuzzy matrix center during clustering comprises:

data Z epsilon after dimensionality reduction is R^m×d′Three users are randomly selected as a clustering center, and original user data is divided into three categories: large users, medium users and small users, and constructing a clustering center vector of the data set: c. C_i＝{c_i,1,c_i,2,…,c_i,kWhere c is 1,2,3_i,kAnd representing the characteristic value of the k-th dimension of the ith cluster center.

6. The method of claim 1, wherein the objective function and constraint conditions are set as follows:

where γ is a membership factor, m represents the number of all samples, and the membership u exists assuming that each sample j belongs to a class i_ijThe relationship of (1); c denotes the center of the cluster, c_iDenotes the ith cluster center, d_ijRepresenting the distance, X, of a sample point from a central point_jRepresents X_pThe vector of the user in (1) is,representing the overall pearson correlation coefficient.

7. The transformer area user identification and voltage influence evaluation method according to claim 6, wherein a Lagrange multiplier and a relaxation variable are introduced from an original objective function in order to obtain a minimum value of the objective function under a constraint condition, an inequality constraint is changed into an equality constraint, the problem of solving the minimum value of the original problem is converted into a convex optimization problem of solving a quadratic plan, and the objective function is reconstructed by integrating the original objective function and the constraint condition:

zeta in the formula_jRepresenting a lagrange multiplier; h (-) represents a membership function; mu.s_jRepresents a relaxation variable; g (-) represents a correlation coefficient function.

8. The station area user identification and voltage impact evaluation method of claim 1, wherein the objective function satisfies the following KKT condition:

in the formulaRepresenting the derivation of the target function;representing an equality constraint in the objective function;an inequality constraint representing an objective function;representing a solution that makes the objective function partial derivative 0.

9. The method of claim 1, wherein the obtaining a relationship between a user activity value between users and a voltage value at a low-voltage side of a transformer of the transformer:

large users and the influence factors are weakly correlated above 0.2; large users and the influence factors are more than 0.4 and are related to a medium degree; the large users have strong correlation when the influence factor is more than 0.6; the influence factors of medium users are weakly correlated above 0.2; moderate users and the influence factors are moderately correlated above 0.4; medium users and strong correlation of influence factors above 0.6; small users and the influence factors are weakly correlated above 0.2; small users and the influence factors are moderately correlated above 0.4; small users and impact factors are strongly correlated above 0.6.

10. The method of claim 1, wherein performing impact evaluation based on the relationship comprises:

obtaining which users belong to large users, medium users, small users and which users have strong influence factors according to clustering; users with influence factors above 0.6 need to take electric measures to defend the users, so as to prevent accidents; the users with the influence factors above 0.4 need to increase the monitoring strength; users with influence factors above 0.2 need to periodically check the system operation to see whether the system is abnormal or not, and provide certain attention to the users; users with an impact factor above 0.2 do not need to intervene.

Technical Field

The invention relates to the field of transformer area relation identification research and electric energy management, in particular to a transformer area user identification and voltage influence evaluation method.

Background

Topological information in the power distribution system has important significance for platform load balance management, and uneven load distribution not only can increase operating line loss, but also can seriously affect the service life of equipment. With the advance of the digitization process of the power grid, power grid companies start to perform partition management on low-voltage distribution areas of the power distribution network. In recent years, the low-voltage transformer area has realized the full coverage of the intelligent electric meter, the quality of the electricity consumption measurement data of the transformer area is improved, and the possibility is provided for the data-driven user topology identification. With the access of a large number of Distributed Generation (DG) of an active power Distribution network and an electric automobile, the load Distribution is uneven and the DG output is difficult to predict; in addition, the structure and the operation mode of the power system with the expanded power distribution network node size, the enhanced load dynamic performance and the like are gradually complicated, the traditional data acquisition method is difficult to meet the data analysis requirement under the large background of the smart power grid, the defects of data loss, low data precision and the like are not easily existed, for example, the SCADA acquisition resolution is 5min, the phase and the accurate value of the data cannot be accurately obtained, and the RTU and the AMI are all functional modules which are packaged based on the smart power meter and have the function of acquiring real-time data. However, the resolutions of the data are 10min and 15min respectively, and the data analysis requirements of modern smart power grids cannot be met. In addition to the data measurement system provided above, mass user load data collected by using a PMU (phasor measurement unit) can also provide data support for operations and maintenance tasks of a distribution area such as line loss analysis and load prediction.

Three indexes of evaluating the power quality, namely voltage, frequency and waveform, have the prominent low-voltage problem in a power distribution network, and the power supply voltage limit values of different voltage grades are regulated in the power quality power supply voltage deviation (GB/T12325-2008): the deviation of three-phase voltage of 20kv and below is the nominal voltage; the 220V single-phase power supply voltage is deviated to a nominal voltage. The reasons for the low voltage problem mainly include three aspects of power supply, power grid and load.

Aiming at the load side, along with the rapid development of urban and rural economic construction in China in recent years, the power utilization load is increased rapidly, some small users are developed into high-power users, the power utilization load exceeds the reserved space of a line, some small users are still small-power users, and some small users are medium-load type users. The contradiction between the power supply capacity of the power distribution network and the rapidly-increased power consumption demand is more and more prominent, so that the problem that the voltage of a transformer area and the voltage of a user side are low frequently occurs, the power consumption experience of the user is seriously influenced, the power supply reliability is reduced, and a severe test is brought to a power supply company. The power load levels of different subscribers can cause the voltage levels on both the platform and the subscriber to be higher or lower. The voltage influence of high-power users on the transformer area in the peak and the valley of power load may cause the increase of line loss, and the overall voltage of the transformer area is reduced; secondly, medium load users may have secondary effects; there are some small users, and although the influence of a single user on the platform area is not obvious in some peak electricity utilization intervals, if there is a small user with similar electricity utilization behavior, the line loss or voltage of the platform area may be influenced to some extent in peak electricity utilization or valley electricity utilization. Aiming at the problem that a smart power grid is established at present, real-time and effective management of a load end by a management department is realized, and the power consumption quality of a user is ensured, a clear relation is not established in the relation identification of a platform house in the traditional low-voltage platform area power quality management, and the relation identification research, the power management and the line loss management of the platform house are not facilitated.

Disclosure of Invention

The invention aims to provide a transformer area user identification and voltage influence evaluation method, which provides technical support for power quality management.

In order to realize the task, the invention adopts the following technical scheme:

a method for identifying users in a distribution room and evaluating voltage influence comprises the following steps:

acquiring active power data of a user side at a set resolution and time point, and constructing a data matrix of active values of users in a distribution area; acquiring a time sequence active value under the transformer area voltage in a unidirectional way, and constructing a time sequence matrix of the voltage at the low side of the transformer;

preprocessing the data matrix by adopting a multi-dimensional scaling method to obtain data after dimension reduction;

based on the data matrix and the time sequence matrix, calculating the relation between each user and the voltage fluctuation of the low-voltage transformer area so as to construct a transformer association factor;

and establishing a target function by taking the membership as an Euclidean distance, and clustering by taking the Pearson correlation coefficient and the membership sum as constraint conditions, thereby obtaining the classification condition of the influence degrees of different user grades in a specific influence factor range, further obtaining the relation between the user success value between the users and the voltage value of the low-voltage side of the transformer in the transformer area, and carrying out influence evaluation according to the relation.

Further, the preprocessing the data matrix by using a multidimensional scaling method to obtain the data after dimensionality reduction includes:

for m users, each user acquires D-dimensional data, and calculates a distance matrix B epsilon R in the original space^m×DIts ith row and j column element dist_ijFor any one user sample x_iTo x_jThe goal is to obtain a representation matrix Z ∈ R in d' dimensional space^m×d′D 'is less than or equal to D, and the Euclidean distance of any two samples in the D' dimensional space is equal to the distance of the original space: | | z_i-z_j||＝dist_ij，z_i,z_jRepresenting the reduced dimension sample, sample x_i、x_jAs a data matrix X_pThe user data in (1);

let E be Z^TZ∈R^m×mAnd decentralizing the matrix Z after dimension reduction, simplifying the matrix Z by combining decentralization constraint conditions, calculating E, and performing characteristic decomposition on the E to obtain E ═ eta^TWherein Λ ═ diag [ λ ═ d [ lambda ] ]₁,λ₂,…,λ_n]A diagonal matrix formed by the eigenvalues, wherein eta is an eigenvector matrix; selecting an eigenvector matrix W [ [ eta ] η ] corresponding to the first 95% contribution eigenvalue according to the contribution of the eigenvalue₁,η₂,…,η_d′]Obtaining an eigenvalue matrix by E matrix eigen decompositionAnd sorting the first d' pieces from big to small: lambda [ alpha ]₁≥λ₂≥…≥λ_q≥λ_d′And obtaining a matrix Z after final dimensionality reduction as follows:

Z＝W^TX。

further, the value of d' is chosen according to the following contribution expression:

further, the calculating the relationship between each user and the voltage fluctuation of the low-voltage transformer area comprises:

and sequentially calculating the overall mean value and the overall covariance between each user and the voltage fluctuation of the low-voltage area, and then calculating to obtain the overall Pearson correlation coefficient.

Further, when clustering is performed, the construction process for the fuzzy matrix center includes:

data Z epsilon after dimensionality reduction is R^m×d′Three users are randomly selected as a clustering center, and original user data is divided into three categories: large users, medium users and small users, and constructing a clustering center vector of the data set: c. C_i＝{c_i,1,c_i,2,...,c_i,kWhere c is 1,2,3_i,kAnd representing the characteristic value of the k-th dimension of the ith cluster center.

Further, when clustering is performed, the set objective function and constraint condition are expressed as:

where γ is a membership factor, m represents the number of all samples, and the membership u exists assuming that each sample j belongs to a class i_ijThe relationship of (1); c denotes the center of the cluster, c_iDenotes the ith cluster center, d_ijRepresenting the distance, X, of a sample point from a central point_jRepresents X_pUser vector of (1) | phi_Xi,VAnd | represents the overall pearson correlation coefficient.

Furthermore, in order to obtain the minimum value of the target function under the constraint condition, the original target function introduces a Lagrange multiplier and a relaxation variable, changes inequality constraint into equality constraint, converts the problem of solving the minimum value of the original problem into the problem of solving the convex optimization of quadratic programming, and reconstructs the target function by integrating the original target function and the constraint condition:

zeta in the formula_jRepresenting a lagrange multiplier; h (-) represents a membership function; mu.s_jRepresents a relaxation variable; g (-) represents a correlation coefficient function.

Further, the objective function satisfies the following KKT condition:

in the formulaRepresenting the derivation of the target function;representing an equality constraint in the objective function;an inequality constraint representing an objective function;representing a solution that makes the objective function partial derivative 0.

Further, the clustering with the pearson correlation coefficient and the sum of membership as constraint conditions includes:

and (3) adopting a fuzzy clustering analysis method, taking the data subjected to dimensionality reduction as a clustering object through an iterative computation mode, and obtaining a final classification cluster through iteration by combining with an improved fuzzy optimal constraint condition KKT.

Further, in the fuzzy clustering process, with the objective function as a convergence condition, a specific iteration process includes:

1) setting a membership factor gamma, an iteration stop error epsilon and the maximum iteration times;

2) calculating an initial distance matrix;

3) updating the membership degree between the user and the clustering center, if the distance between the user and the clustering center is 0, the membership degree is 1, otherwise, determining the membership degree according to a derivation formula, wherein the membership degree updating formula is as follows:

wherein (t) represents the t-th iteration, and d () represents the distance from the sample point to the cluster center;

4) updating a clustering center:

5) recalculating the distance formula and calculating a target function;

6) comparing whether the target function is smaller than a set error epsilon or whether the iteration times meet an iteration ending condition, otherwise, turning to the step 3) to recalculate the membership degree until a constraint condition is met and jumping out of an iteration loop; and obtaining preset class clusters after the iteration is finished, wherein each class cluster has a corresponding correlation coefficient value.

Further, obtaining a relationship between a user active value and a transformer low-voltage side voltage value of the transformer area between the users comprises:

large users and the influence factors are weakly correlated above 0.2; large users and the influence factors are more than 0.4 and are related to a medium degree; the large users have strong correlation when the influence factor is more than 0.6; the influence factors of medium users are weakly correlated above 0.2; moderate users and the influence factors are moderately correlated above 0.4; medium users and strong correlation of influence factors above 0.6; small users and the influence factors are weakly correlated above 0.2; small users and the influence factors are moderately correlated above 0.4; small users and impact factors are strongly correlated above 0.6.

Further, performing impact evaluation according to the relationship, including:

obtaining which users belong to large users, medium users, small users and which users have strong influence factors according to clustering; users with influence factors above 0.6 need to take electric measures to defend the users, so as to prevent accidents; the users with the influence factors above 0.4 need to increase the monitoring strength; users with influence factors above 0.2 need to periodically check the system operation to see whether the system is abnormal or not, and provide certain attention to the users; users with an impact factor above 0.2 do not need to intervene.

A terminal device comprises a memory, a processor and a computer program which is stored in the memory and can run on the processor, wherein the processor executes the computer program to realize the steps of the station area user identification and voltage influence evaluation method.

A computer-readable storage medium, in which a computer program is stored, which, when being executed by a processor, implements the steps of the aforementioned station area user identification and voltage impact evaluation method.

Compared with the prior art, the invention has the following technical characteristics:

the method can distinguish large, medium and small clusters of a certain phase of users under a single transformer area, simultaneously reflects the influence factor of the users in each cluster by combining the Pearson correlation coefficient, can directly judge the relation between the users in a certain grade and the voltage fluctuation of the low-voltage side of the transformer, has important significance for knowing the topology of a power distribution network and building a smart power grid, and provides necessary technical support for further power quality management.

Drawings

FIG. 1 is a schematic flow diagram of the process of the present invention;

FIG. 2 is a detailed flow chart of an improved optimal fuzzy clustering algorithm.

Detailed Description

Referring to fig. 1, the invention provides a method for identifying users in a distribution room and evaluating voltage influence, which is based on Multi-Dimensional Scaling (MDS) combined with pearson influence factors to improve a fuzzy C-means clustering method, can quickly distinguish user grades and identify influence degrees of users with different grades on voltage fluctuation of a low-voltage distribution room, and finally obtains a cluster with a large influence degree of different users on the distribution room. The power management is normalized and continuous, and the user types cannot be judged according to data of a certain day or two days by distinguishing the user grades, so that the relationship of the users is unknown, the power management is distorted, and management redundancy or insufficient management strength is generated. The data mining is generally carried out on the electricity utilization historical data in the same station area with the collection interval of 7d and different users 7d, so that N users are divided into three types, namely large users, medium users and small users. The technical solution of the present invention will be further described in detail with reference to the accompanying drawings.

A transformer area user identification and voltage influence evaluation method comprises the following steps:

construction of S1 data matrix

The station-user relationship identification provided by the invention is based on historical data of the station area and users, wherein the influence of the power peak-valley interval on the voltage of the station area is based on the historical data of the station area and the users. The user active power data used by the invention is obtained by historical collected data of a synchronous vector Measurement Unit (PMU), and m user side active power data of a certain phase of the transformer at D time points are obtained according to the resolution ratio of Mmin. Data matrix X of active values of station users acquired from historical data of power utilization acquisition system_P∈R^m×D(R represents the real number domain), where m × D represents the m users co-samplingD time points are collected, and are specifically expressed as follows:

whereinRespectively being user end 1,2, m number users at t_jThe unit of the active measurement value at a moment is as follows: kw, j ═ 1, 2.

Obtaining the time sequence active value under the single-phase of the platform area voltage, and synchronously obtaining the single-phase voltage data according to the resolution and the acquisition interval, wherein the unit is as follows: volts. Constructing a voltage time sequence matrix of the low-voltage side of the transformer:

in the formula V_TA matrix formed by D voltage values collected in a period under a certain transformer single phase (for example, A phase) in the transformer area;and the active measurement voltage value of the single phase at the time D is shown.

Data preprocessing of S2 multidimensional scaling method

The Multi-Dimensional Scaling (MDS) is a typical dimension reduction algorithm, which reduces the dimension of original data and minimizes the phenomenon of data "distortion" on the principle of keeping the characteristics of the original data to the maximum. Therefore, the calculation amount of the data can be reduced, and the original characteristics of the data can be kept as much as possible. For m users, each user acquires D-dimensional data, and a distance matrix B epsilon R in the original space can be calculated^m×DIts ith row and j column element dist_ijFor a sample (any one user) x_iTo x_jThe goal is to obtain a representation matrix Z ∈ R in d' dimensional space^m×d′D 'is less than or equal to D, and the Euclidean form of any two samples in D' dimension spaceThe distance is equal to the distance of the original space, i.e. | | z_i-z_j||＝dist_ij，z_i,z_jThe reduced dimension samples are represented. Wherein the sample x_i、x_jIs X_pThe user data in (1) isAnd the like.

Let E be Z^TZ∈R^m×mWherein E is an inner product matrix of the reduced samples,comprises the following steps:

wherein the content of the first and second substances,

making the matrix Z after dimension reduction belong to R^m×d′Decentralization, i.e.

And (3) simplifying the formula by combining the decentralized constraint condition:

wherein, dist_i.、dist_j.、dist_ijThe average distance is indicated.

Obtained by the above formula

Respectively calculating E-Z from the above formula^TZ∈R^m×mAnd performing characteristic decomposition on the E to obtain E ═ eta Λ eta^TWherein Λ ═ diag [ λ ═ d [ lambda ] ]₁,λ₂,…,λ_n]And the eta is a feature vector matrix. In reality, for effective dimension reduction, the distance after dimension reduction is often only required to be as close as possible to the distance of the original space, and is not necessarily strictly equal. Selecting an eigenvector matrix W [ [ eta ] η ] corresponding to the first 95% contribution eigenvalue according to the contribution of the eigenvalue₁,η₂,…,η_d′]. And (3) obtaining an eigenvalue matrix by E matrix eigen decomposition, and sequencing the eigenvalue matrix from large to small (d' first): lambda [ alpha ]₁≥λ₂≥…≥λ_q≥λ_d′. The value of d' is expressed according to the contribution degree:

the matrix Z after dimension reduction belongs to R^m×d′Can be expressed as:

Z＝W^TX

wherein W is ∈ R^D×d′Is a transition matrix, Z ∈ R^m×d′Is a sample space X_pAnd (5) reducing the expression of the new space.

S3 construction of platform relationship factor

For X_pM sets of user vectors { x }₁,x₂,x₃,...,x_m(for convenience of representation, subscript t is omitted)_j) And corresponding low-voltage platform voltage matrixAnd calculating the relation between each user and the voltage fluctuation of the low-voltage transformer area through the Pearson correlation coefficient:

overall mean value:

overall covariance:

wherein, X_i、V_iRespectively representing the active numerical value and the voltage vector of the transformer area of a single user;

overall pearson correlation coefficient:

σ_X，σ_Vare each X_pAnd V_TStandard deviation of (a):

s4 improved fuzzy clustering method station area user identification

The target object of the clustering algorithm obtained in the step S2 is the data feature with 95% of original data after dimensionality reduction. At present, the common clustering algorithms in China are roughly divided into two categories: direct and indirect processes. The direct method is used for directly clustering data, and is usually subjected to K-means, FCM, SOM and the like, but with the continuous increase of data scale and the influence of data noise, data residual errors, default values and the like, the direct method not only brings the challenges of poor clustering effect, large storage capacity, low calculation efficiency and the like. Clustering is carried out according to a traditional clustering algorithm, only users with set clusters can be obtained, the relation between the users and the distribution area voltage cannot be rapidly identified, and further Pearson correlation coefficient analysis needs to be combined.

Based on the above, an improved fuzzy clustering algorithm is provided, which can quickly obtain the definite relationship among the platform-to-user relationships, maintain the characteristics of the original data, greatly reduce the calculation amount of the clustering algorithm and improve the calculation efficiency. The membership degree is used as Euclidean distance to establish an objective function, and the Pearson correlation coefficient and the membership degree sum are used as constraint conditions to perform clustering, wherein the processing process is as follows:

s4.1 construction of fuzzy clustering center

Adopting a fuzzy clustering analysis method, and enabling data Z after dimensionality reduction to belong to R^m×d′Three users are randomly selected as a clustering center, and original user data is divided into three categories: large users, medium users and small users. Constructing a clustering center vector of the data set: c. C_i＝{c_i,1,c_i,2,…,c_i,kWhere c is 1,2,3_i,kAnd representing the characteristic value of the k-th dimension of the ith cluster center.

S4.2 optimal fuzzy clustering constraint condition setting and reconstruction objective function

The traditional fuzzy clustering algorithm has the constraint condition that the sum of the membership degrees of each particle to each clustering center is always 1. In order to directly identify users in different clusters, for particles with a fixed loudness in the platform region voltage, a Pearson correlation coefficient is added as a constraint condition, so that the user class can be obtained finally, and the user number with a certain influence factor can be identified. Objective function and constraints:

wherein gamma is a membership factor, and m represents the number of all samples, i.e. the number of users; assuming that each sample j belongs to a certain class i, there is a degree of membership u_ijThe relationship of (1); c denotes the center of the cluster, c_iDenotes the ith cluster center, d_ijRepresenting the distance, X, of a sample point from a central point_jRepresents X_pThe user vector of (1).

The traditional fuzzy clustering algorithm has the constraint condition that the sum of the membership degrees of each particle to each clustering center is always 1. In order to directly identify users in different clusters, for particles with a fixed loudness in the platform region voltage, a Pearson correlation coefficient is added as a constraint condition, so that the user class can be obtained finally, and the user number with a certain influence factor can be identified. According to the original objective function, in order to obtain the minimum value of the objective function under the constraint condition, equality constraint and inequality constraint are carried out: introducing a Lagrange multiplier and a relaxation variable, changing inequality constraint into equality constraint, converting the problem of solving the minimum value of the original problem into the problem of solving convex optimization of quadratic programming, and reconstructing an objective function by integrating the original objective function and the constraint condition:

zeta in the formula_jRepresenting a lagrange multiplier; h (-) represents a membership function; mu.s_jRepresents a relaxation variable; g (-) represents a correlation coefficient function.

S4.3 Lagrange multiplier and relaxation variables

Because the original objective function meets the KKT condition, the KKT condition is a sufficient necessary condition for solving the optimization problem. An SMO heuristic can be applied, whose basic idea is: two variables are selected, other variables are fixed, and a quadratic programming problem is constructed aiming at the two variables. The quadratic programming subproblems of the two variables should be closer to the solution of the original quadratic programming problem, because the new variable values can make the original objective function smaller, and more importantly, the subproblems are solved by an analytical method, so that the overall calculation speed of the algorithm is greatly improved. The SMO algorithm continuously decomposes the original problem into sub-problems and solves the sub-problems, thereby achieving the purpose of solving the original problem.

KKT condition:

in the formulaRepresenting the derivation of the target function;representing an equality constraint in the objective function;an inequality constraint representing an objective function; x_j ^*Representing a solution that makes the objective function partial derivative 0.

S4.4 iterative process of optimized fuzzy clustering matrix and optimized fuzzy clustering objective function

By adopting a fuzzy clustering analysis method, obtaining a final classification cluster through iteration by taking a feature matrix subjected to dimensionality reduction obtained by S2 as a clustering object and combining with an improved fuzzy optimal constraint condition KKT in an iterative computation mode; in the fuzzy clustering process, the objective function is taken as a convergence condition, and the specific iteration process is as follows:

1) setting a membership factor gamma, an iteration stop error epsilon and a maximum iteration number (LOOP);

2) calculating an initial distance matrix;

3) updating the membership degree between the user and the clustering center, wherein d (-) is a distance function from the sample point to the sample center, if the distance between the user and the clustering center is 0, the membership degree is 1, otherwise, the membership degree is determined according to a derivation formula, and the updating formula of the membership degree is as follows:

where (t) represents the t-th iteration and d () represents the distance of the sample point to the cluster center.

4) Updating a clustering center:

5) recalculating the distance formula and calculating a target function;

6) and (3) comparing whether the target function is smaller than a set error epsilon or whether the iteration times meet an iteration ending condition, and otherwise, turning to the step 3) to recalculate the membership degree until a constraint condition is met and jumping out of an iteration loop. And obtaining preset class clusters after the iteration is finished, wherein each class cluster has a corresponding correlation coefficient value.

S4.5 clustering results analysis

The traditional fuzzy clustering algorithm is applied to the identification of the relationship between the users, the clustering objects adopt MDS characteristic values, the clustering results only can show the clustering results of different user characteristics, and the user objects with larger influence factors can not be quickly obtained. Based on the constraint conditions of the traditional fuzzy clustering algorithm, the physical meaning of the Pearson correlation coefficient is referred to in Table 1, and the Pearson influence factor is addedAs a constraint condition of user clustering, the method aims to quickly obtain the classification condition of the influence degrees of different user grades in a specific influence factor range and quickly obtain the relationship between the user work value and the transformer low-voltage side voltage value of the transformer in a transformer area between the users. With the previous clustering algorithm, the clustering result is: large users and influence factor at 0More than 2 are weakly correlated; large users and the influence factors are more than 0.4 and are related to a medium degree; the large users have strong correlation when the influence factor is more than 0.6; the influence factors of medium users are weakly correlated above 0.2; moderate users and the influence factors are moderately correlated above 0.4; medium users and strong correlation of influence factors above 0.6; small users and the influence factors are weakly correlated above 0.2; small users and the influence factors are moderately correlated above 0.4; small users and impact factors are strongly correlated above 0.6.

Influence and evaluation: through the clustering result, specific users belonging to a large user, a medium user and a small user can be obtained, and the users having strong influence factors can be obtained. Users with influence factors above 0.6 (including users with large, medium and small grades) need to take electric measures to defend the users, so as to prevent accidents; the users with the influence factors above 0.4 need to increase the monitoring strength; users with influence factors above 0.2 need to periodically check the system operation to see whether the system is abnormal or not, and provide certain attention to the users; users with an impact factor above 0.2 do not need to intervene.

The method adopts an MDS algorithm to extract dimensionality reduction features of data features on the basis of original data, maintains the characteristics of the original data, greatly reduces the calculated amount of a clustering algorithm, improves the calculation efficiency, improves the traditional fuzzy clustering method, combines classification and influence factors, obtains user categories and the influence factors simultaneously from clustering results, and finally adopts proper intervention measures according to the physical meanings of the influence factors.

TABLE 1

The embodiment of the application further provides a terminal device, which can be a computer or a server; the method comprises a memory, a processor and a computer program which is stored in the memory and can run on the processor, wherein the processor realizes the steps of the station area user identification and voltage influence evaluation method when executing the computer program.

The computer program may also be partitioned into one or more modules/units, which are stored in the memory and executed by the processor to accomplish the present application. One or more modules/units may be a series of computer program instruction segments capable of performing specific functions, which are used to describe the execution of a computer program in a terminal device.

The implementation of the present application provides a computer-readable storage medium, which stores a computer program, and the computer program, when executed by a processor, implements the steps of the above-mentioned station user identification and voltage impact evaluation method.

The above embodiments are only used to illustrate the technical solutions of the present application, and not to limit the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; such modifications and substitutions do not substantially depart from the spirit and scope of the embodiments of the present application and are intended to be included within the scope of the present application.