阅读说明：本技术 一种基于大数据的磷酸铁锂电池soc充电在线校正方法 (Lithium iron phosphate battery SOC charging online correction method based on big data ) 是由 来翔 彭勇俊 *** 王晓东 于 2019-11-11 设计创作，主要内容包括：本发明公开了一种基于大数据的磷酸铁锂电池SOC充电在线校正方法,属于汽车电池领域。针对现有电动汽车经常使用在浅充浅放的工况,而磷酸铁锂电池在此时没有较好的校正方法,导致SOC不准确从而影响驾驶体验性的问题,本发明提供一种基于大数据的磷酸铁锂电池SOC充电在线校正方法,使用数据过滤模块基于已有的充电数据得到表征电池特性的特征值,搭建包含BP神经网络模型的电池模型；根据后台电池当前充电情况和特征值,通过SOC校正模块得到估算SOC,从而实现SOC的在线校正。本发明能够根据已有的数据分析结果,在车辆充电过程中校正SOC,防止SOC长时间没有得到校正而造成较大累积误差,数据过滤模块对大数据进行筛选,保证了SOC的准确性和精度。(The invention discloses a lithium iron phosphate battery SOC charging online correction method based on big data, and belongs to the field of automobile batteries. Aiming at the problem that the driving experience is affected due to inaccurate SOC (system on chip) caused by the fact that the conventional electric automobile is frequently used in a shallow charging and shallow discharging working condition and the lithium iron phosphate battery does not have a better correction method at the moment, the invention provides a lithium iron phosphate battery SOC charging online correction method based on big data, wherein a characteristic value representing the battery characteristic is obtained by using a data filtering module based on the existing charging data, and a battery model containing a BP (back propagation) neural network model is built; and obtaining the estimated SOC through an SOC correction module according to the current charging condition and the characteristic value of the background battery, thereby realizing the online correction of the SOC. According to the invention, the SOC can be corrected in the vehicle charging process according to the existing data analysis result, so that a large accumulated error caused by long-time non-correction of the SOC is prevented, and the data filtering module screens the big data, so that the accuracy and precision of the SOC are ensured.)

1. The method is characterized in that a system collects charging parameters of the current working condition, a data filtering module is used for filtering the parameters, a BP neural network module carries out data analysis on the selected filtering parameters, and an SOC correction module corrects the analyzed data to realize online correction of the charging process.

2. The lithium iron phosphate battery SOC charging online correction method based on big data as claimed in claim 1, characterized in that the data filtering module firstly groups the collected data and then filters and selects each group of data.

3. The lithium iron phosphate battery SOC charging online correction method based on big data as claimed in claim 2, characterized in that the charging parameters include current, temperature, voltage, battery health SOH and battery SOC.

4. The lithium iron phosphate battery SOC charging online correction method based on big data according to claim 3, characterized in that the grouping conditions of the data filtering module include:

judging whether the charging current is stable;

the SOC of the initial charging battery is smaller than a threshold value, and the deviation of the SOC of the battery is smaller than a set value during full charge correction; calculating a battery capacity differential curve dQ/dv of the acquired data, wherein the calculated time interval is equal potential interval, and adopting sliding average filtering on the dQ/dv;

wherein dt represents a charging time; dQ represents the accumulated charge capacity at dt times; i represents a charging current; dv represents the voltage change over dt time;

and grouping according to the SOH and the battery temperature to obtain the average voltage and the average dQ/dv value of each group.

5. The lithium iron phosphate battery SOC charging online correction method based on big data according to claim 4, characterized in that the difference between the current and the last uploaded current is within 5%, and the charging is considered to be stable; calculating a battery capacity differential curve dQ/dv value of the acquired data when the initial charging SOC is less than 30% and the SOC deviation is less than 2% during full charge correction; the dQ/dv is calculated once at an interval of 5mv, and the filtering frequency of the moving average filtering is 10 times; the parameters are grouped according to SOH phase difference of 2% and temperature phase difference of 5 ℃.

6. The lithium iron phosphate battery SOC charging online correction method based on big data according to claim 4 or 5, characterized in that the filtering condition of the data filtering module on the grouped data comprises:

comparing the voltage parameter in the group with the voltage average value of the selected battery model;

the differential curve dQ/dv of the battery capacity is the maximum value of the battery state of charge SOC within a set threshold value, and the difference between the maximum value and the dQ/dv average value of the existing battery model is within a set range.

7. The lithium iron phosphate battery SOC charging online correction method based on big data according to claim 6, characterized in that the difference between the voltage parameter in the group and the voltage average value of the selected battery model is within 10 mv; the differential curve dQ/dv of the battery capacity is a maximum value of the battery state of charge SOC at 5%, and the difference between the maximum value and the dQ/dv average value of the existing battery model is within 20%.

8. The lithium iron phosphate battery SOC charging online correction method based on big data according to claim 3, characterized in that the BP neural network module divides the screened packets into a training set and a test set; establishing three layers of neural networks with inputs of dQ/dv, temperature, SOH and voltage and outputs of SOC and a hidden layer, applying a training set to obtain the trained neural network, using a test set to test the established neural network, and if the average error of the test set is within a set value, using the neural network; otherwise, wait for new charging data to increase the number of training sets.

9. The lithium iron phosphate battery SOC charging online correction method based on big data according to claim 8, characterized in that if the BP neural network is not available, new data are uploaded at fixed time intervals to increase the number of training sets, and the training is repeated until the requirements of the test set are met; and the BP neural network is updated regularly, the updated data replaces the training set and the test set in the same time, and the training and the testing of the BP neural network are carried out again.

10. The lithium iron phosphate battery SOC charging online correction method based on big data according to claim 3, characterized in that the correction strategy of the SOC correction module is as follows:

if the difference between the corrected SOC obtained by the selected battery model and the current battery SOC is within a set value, the correction is not carried out;

if the corrected SOC obtained by the selected battery model is larger than the current SOC by more than a set value, subtracting the set value from the SOC correction;

and if the corrected SOC obtained by the selected battery model is smaller than the current SOC by more than a set value, increasing the SOC correction by the set value.

Technical Field

The invention relates to the field of automobile batteries, in particular to a lithium iron phosphate battery SOC charging online correction method based on big data.

Background

The power battery is a power source for providing power source for the tool, and is a storage battery for providing power for electric automobiles, electric trains, electric bicycles and golf carts. The power battery is a device for converting chemical energy into electric energy, and can replace gasoline and diesel oil to be used as a running power supply of an electric automobile or an electric bicycle in the future, and has the characteristics of super-long service life, generally 5-10 years of service life, support of quick charge and discharge, high temperature resistance, large capacity, small volume, light weight, safe use and the like. The power battery is also an environment-friendly green power supply, does not use any toxic substances such as mercury, chromium, lead and the like, and has no pollution to the environment.

The conversion process of the power battery is a complex physical and chemical reaction process, and the peak value of a differential curve dQ/dv of the capacity voltage can well represent the phase change of the conversion process. For lithium iron phosphate batteries, the platform area range is large, the OCV correction or the SOC calculation by the extended Kalman filtering method is not easy, generally, an ampere-hour integration method is used, the service life of the battery is greatly prolonged by shallow charging, so that the accumulation is larger and larger by the pure ampere-hour integration method under the working condition, the phase change of the battery in the charging process can be well reflected by the peak value of the dQ/dv, but the fluctuation of a dQ/dv curve is often caused by considering the sampling error in the actual operation process, and the direct use of the dQ/dv peak value correction is not easy.

Chinese patent application for a lithium ion battery SOC prediction method based on big data and a bp neural network, application No. 201910377462.8, published 2019, 5 months and 7 days, discloses a lithium ion battery SOC prediction method based on big data and a bp neural network, and collects external characteristic parameters of a battery to establish a data set of battery SOC big data; establishing a training set and a test set; constructing a multi-bp neural network prediction model; respectively putting the data sets into bp neural network prediction models with different parameters to obtain measurement precision; and analyzing the measurement precision obtained by the bp neural network prediction model with different parameters to obtain a prediction result. According to the lithium ion battery SOC prediction method based on the big data and the bp neural network, the data can be conveniently and effectively mined through the big data set of SOC prediction, and the prediction precision is ensured; the method can accurately predict the SOC of the battery by the external parameters of the battery through the distributed bp neural network prediction model, has higher precision, particularly under the condition of large data, the external parameters of the battery continuously change, still can accurately predict the value of the SOC of the battery, has high usability, and can be widely applied in practice.

Disclosure of Invention

1. Technical problem to be solved

Aiming at the problems that the SOC of the existing lithium iron phosphate battery is inaccurate in calculation, the accumulated error is larger and larger, and error correction can occur in the prior art, the invention provides the online correction method for the SOC of the lithium iron phosphate battery based on big data.

2. Technical scheme

The purpose of the invention is realized by the following technical scheme.

A lithium iron phosphate battery SOC charging online correction method based on big data comprises the steps that a system collects charging parameters of the current working condition, a data filtering module is used for filtering the parameters, a BP neural network module carries out data analysis on the selected filtering parameters, and an SOC correction module corrects the analyzed data to achieve online correction of the charging process. The system builds a battery model according to the existing charging historical data, the battery model comprises a data filtering module, a BP neural network module and an SOC correction module, in the charging process, the system acquires and calculates charging parameters in real time and inputs the charging parameters into the data filtering module, when the conditions of the data filtering module are met, the system enters the BP neural network module again to calculate the SOC needing to be corrected, the SOC is input into the SOC correction module again, whether the SOC is corrected or not is determined by combining the current SOC, and if the SOC needs to be corrected, a corrected SOC value is given.

Furthermore, the data filtering module firstly groups the collected data and then performs filtering selection on each group of data.

Further, the charging parameters include current, temperature, voltage, battery state of health, SOH, and battery state of charge, SOC.

Further, the grouping condition of the data filtering module comprises:

judging whether the charging current is stable;

the SOC of the initial charging battery is smaller than a threshold value, and the deviation of the SOC of the battery is smaller than a set value during full charge correction; calculating a battery capacity differential curve dQ/dv of the acquired data, wherein the calculated time interval is equal potential interval, and adopting sliding average filtering on the dQ/dv;

wherein dt represents a charging time; dQ represents the accumulated charge capacity at dt times; i represents a charging current; dv represents the voltage change over dt time;

and grouping according to the SOH and the battery temperature to obtain the average voltage and the average dQ/dv value of each group.

The current, the temperature, the voltage, the battery health degree SOH and the battery state of charge SOC are basic parameters of the battery, a battery capacity differential curve dQ/dv can be calculated according to the battery charging parameters according to limited experimental data analysis, the battery capacity differential curve can well reflect the phase change of the battery in the charging process, the sampling error is reduced, data are selected through a data filtering module, BP neural network calculation is carried out according to the peak value of dQ/dv and the temperature, the voltage and the SOH, and the SOC can be well corrected in the charging process.

Furthermore, the difference between the current and the current uploaded last time is within 5 percent, and the charging is considered to be stable; calculating a battery capacity differential curve dQ/dv value of the acquired data when the initial charging SOC is less than 30% and the SOC deviation is less than 2% during full charge correction; the dQ/dv is calculated once at an interval of 5mv, and the filtering frequency of the moving average filtering is 10 times; the parameters are grouped according to SOH phase difference of 2% and temperature phase difference of 5 ℃.

Further, the filtering condition of the grouped data by the data filtering module comprises:

comparing the voltage parameter in the group with the voltage average value of the selected battery model;

the differential curve dQ/dv of the battery capacity is the maximum value of the battery state of charge SOC within a set threshold value, and the difference between the maximum value and the dQ/dv average value of the existing battery model is within a set range.

Furthermore, the difference between the voltage parameter in the battery pack and the voltage average value of the selected battery model is within 10 mv; the differential curve dQ/dv of the battery capacity is a maximum value of the battery state of charge SOC at 5%, and the difference between the maximum value and the dQ/dv average value of the existing battery model is within 20%.

Further, the BP neural network module divides the screened packets into a training set and a test set; establishing three layers of neural networks with inputs of dQ/dv, temperature, SOH and voltage and outputs of SOC and a hidden layer, applying a training set to obtain the trained neural network, using a test set to test the established neural network, and if the average error of the test set is within a set value, using the neural network; otherwise, wait for new charging data to increase the number of training sets. Wherein 70% of the data of each packet is used as a training set and 30% of the data is used as a test set; in the invention, if the average error of the test set is within 5 percent, the neural network can be used; otherwise, wait for new charging data to increase the number of training sets.

Furthermore, if the BP neural network is not available, new data are uploaded at fixed time intervals to increase the number of training sets, and the training is carried out again until the requirements of the test set are met; and the BP neural network is updated regularly, the updated data replaces the training set and the test set in the same time, and the training and the testing of the BP neural network are carried out again. Generally, new data is uploaded every 3 days to increase the number of training sets; the updating is carried out once every month, and the training set and the testing set of the same time period before one year are advanced.

Further, the correction strategy of the SOC correction module is as follows:

if the difference between the corrected SOC obtained by the selected battery model and the current battery SOC is within a set value, the correction is not carried out;

if the corrected SOC obtained by the selected battery model is larger than the current SOC by more than a set value, subtracting the set value from the SOC correction;

and if the corrected SOC obtained by the selected battery model is smaller than the current SOC by more than a set value, increasing the SOC correction by the set value.

When the current parameter does not meet the SOC correction, the calculation of the SOC continues to use the original method, such as ampere-hour integration or Kalman filtering.

The correction parameter setting value of the SOC correction module is 3%, the SOC correction module only corrects the SOC data once, and if the SOC correction module corrects the SOC data once in the charging process, the SOC correction module does not correct the SOC data. In practical tests, the change curve of dQ/dv from empty to full generally has only two peaks. The deviation range of the corrected value and the true value is generally +/-2.5%, so that in the correction method, 3% of correction indexes are adopted, and only one correction is needed.

3. Advantageous effects

Compared with the prior art, the invention has the advantages that:

(1) the invention provides a method for extracting characteristic parameters of a rechargeable battery based on existing data and building a battery model, judging whether SOC correction is needed or not through the model, and if so, issuing the corrected SOC to a controller ECU for SOC correction. Compared with the prior art, the method can eliminate the accumulated error of the ampere-hour integral calculation SOC under the shallow charging and shallow discharging working condition, and improve the accuracy of the SOC;

(2) because the battery generates a lot of data during charging, the battery model built by the invention comprises a technology and data filtering module, a BP neural network module and an SOC correction module, so that the filtering, calculation and correction of the data are realized, and the noise data are effectively filtered. Several typical cell characterization parameters were selected simultaneously: the peak values of the temperature, the voltage, the SOH and the dQ/dv are used as the input of the BP neural network, so that the excessive input is prevented, and the BP neural network is prevented from being over-fitted; the method selects the parameter dQ/dv, the peak value of the dQ/dv can well reflect the phase change of the battery in the charging process, the method has good representativeness to the same battery, and the method is easy to obtain and calculate in the charging process. The dQ/dv has a little deviation along with the difference of temperature, current and SOH, and the current is almost fixed during charging, so that different neural network models are selected according to the difference of the current and the SOH, and the deviation problem of the dQ/dv can be effectively solved;

(3) the invention stores the training set and the test set of the BP neural network of the last year in the background system, iterates the training set and the test set once every month, can ensure the coverage of the training set and the test set, and can avoid the overstaffed background data and the serious resource consumption caused by excessive storage of working condition data due to more vehicles and long time needing to be managed by the background.

Drawings

FIG. 1 is a flow chart of the method of the present invention;

fig. 2 is a flow of the charging calibration strategy according to the present invention.

Detailed Description

The invention is described in detail below with reference to the drawings and specific examples.