A power battery capacity evaluation algorithm based on big data technology

Abstract

The present invention provides a power battery capacity algorithm based on big data technology evaluation. Based on a data-driven online battery capacity evaluation method, it can realize the capacity estimation of power batteries in all regions, all working conditions, and the entire life cycle. The algorithm fully considers the problem that it is difficult to obtain complete SOC data using only ampere-hour integration, and provides an effective compensation method, so as to provide a relatively complete and accurate correspondence between cumulative mileage and capacity. Based on the present invention, the rapidity, timeliness and robustness of retired battery screening can be significantly improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of power battery status monitoring and management, and specifically relates to a power battery capacity algorithm evaluated based on big data technology. Background Art

[0002] With the popularization of new energy vehicles and the development of related industries, the number of retired batteries has reached a considerable scale in recent years. According to the current average retirement level of 20% attenuation of power batteries in 8 years, the number of retired power batteries in my country is already in a rapid growth stage. The forecast results show that by 2025, the number of retired power batteries in my country will reach 93GWh, and the number of retired battery packs will exceed 1 million. After the power batteries are retired, if they can be effectively screened and processed and used in appropriate fields, there is still an opportunity to create huge social and economic value.

[0003] At present, due to differences in brands, models, parameters, and remaining lifespans, and the number of retired batteries each year has exceeded one million, how to quickly classify and detect battery packs is the key to whether they can be smoothly recycled. The existing technology still mainly relies on offline vehicle testing to measure the capacity of retired batteries. The testing process is time-consuming and labor-intensive, and the investment in manpower, equipment, and venues also increases sharply with the increase in the number of batteries. The investment cost is high and the return is slow. Therefore, how to provide a time-saving and efficient capacity assessment method to address the defects of large-scale testing and screening of existing retired power batteries is a technical problem that needs to be solved urgently in this field. Summary of the invention

[0004] Different from traditional fuel vehicles, the data of new energy vehicles in driving conditions, power battery status and other aspects are easier to collect and integrate, which helps to solve the supervision of a large number of vehicles, batteries and other components by using on-board big data. In view of this, the present invention provides a power battery capacity evaluation algorithm based on big data technology to address the technical problems existing in the field, which specifically includes the following steps:

[0005] Step 1: Use the in-vehicle intelligent terminal to collect the charging habits of a specific vehicle model and upload it to the new energy vehicle big data monitoring platform;

[0006] Step 2: The monitoring platform pre-processes the received charging habit-related characteristic data and calculates the SOC segments of multiple different charging behaviors using the ampere-hour integration method;

[0007] Step 3: crawl the four features corresponding to the calculated SOC fragment, including vehicle model related features, battery type related features, single charge related features, and vehicle environment related features, to form a data strip of the vehicle model containing the above five features;

[0008] Step 4: Perform kmeans clustering and linear regression algorithm training on the data strips to obtain a partial correspondence between the cumulative mileage and the battery capacity of the vehicle model;

[0009] For the complete correspondence between accumulated mileage and battery capacity, compensation is performed for missing parts other than the partial correspondence, including: if there is complete data of the correspondence in a certain class after clustering, the complete data in the class is used for compensation; if there is no complete data in the class but there is complete data globally, the global complete data outside the class is used for compensation.

[0010] Step 5: For the discrete points of the relationship between the cumulative mileage and the battery capacity of each compensated data strip, a complete curve of the relationship between the cumulative mileage and the battery capacity is fitted based on linear regression.

[0011] Step 6: Output the retired battery capacity evaluation result of the target vehicle according to the fitting curve and the actual vehicle data of the target vehicle of the same model;

[0012] Step 7: Perform an offline battery capacity test on the same vehicle model, compare it with the battery capacity evaluation result obtained based on the previous steps, and make feedback adjustments to the fitted curve based on the comparison result.

[0013] Furthermore, the battery type-related characteristics include: battery material, capacity density, manufacturer and other characteristic parameters; vehicle type-related characteristics include but are not limited to vehicle type classification characteristics such as sedans, SUVs, trucks, vehicle levels, vehicle weights, different configurations of the same vehicle type and other characteristic parameters; the single charge-related characteristics include:

[0014] The five characteristic parameters include the mileage after the current charge, the charging rate, the charging depth, the charging time, and the ambient temperature during the current charge; the charging habit-related characteristics include: the number of days the battery has been put into use, the mileage difference of the vehicle after being put into use, the total number of charging times, the total charging capacity, the average single charging capacity, the total charging hours, the average charging rate, the average starting SOC of charging, the average ending SOC, the average charging SOC depth, the median SOC at the start of charging, the median SOC at the end of charging, and the median charging SOC depth, which are 13 characteristic parameters; the vehicle environment-related characteristics include: the annual average temperature of the environment, the annual median temperature, the annual temperature variance, the annual temperature range, and the number of effective temperatures.

[0015] Furthermore, the preprocessing in step 2 and step 6 both include preliminary elimination and sorting of vehicle models based on the vehicle identification code VIN and the legal status of the vehicle.

[0016] Furthermore, the compensation in step 4 specifically includes: first removing outliers from the clustered data strips and then taking the mean of the complete data for compensation, or first taking the mean of the clustered complete data strips for compensation and then removing outliers.

[0017] Furthermore, in step 5, the corresponding relationship curve between cumulative mileage and battery capacity based on linear regression fitting is specifically fitted by a univariate linear model, with cumulative mileage as the independent variable and battery capacity as the dependent variable, and the law of battery capacity change with attenuation is output; and / or some high-quality vehicle samples are used as a backup to perform linear fitting.

[0018] The power battery capacity algorithm based on big data technology provided by the present invention is based on a data-driven online battery capacity evaluation method, which can realize the capacity estimation of power batteries in all regions, all working conditions, and the entire life cycle. The algorithm fully considers the problem that it is difficult to obtain complete SOC data using only ampere-hour integration, and provides an effective compensation method, so as to provide a relatively complete and accurate correspondence between accumulated mileage and capacity. Based on the present invention, the rapidity, timeliness and robustness of retired battery screening can be significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A capacity algorithm process established based on the present invention;

[0020] Figure 2 Example of battery compensation scatter plot and fitted curve drawing. DETAILED DESCRIPTION

[0021] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] The present invention provides a power battery capacity algorithm based on big data technology evaluation, such as Figure 1 As shown, the specific steps include:

[0023] Step 1: Use the in-vehicle intelligent terminal to collect the charging habits of a specific vehicle model and upload it to the new energy vehicle big data monitoring platform;

[0024] Step 2: The monitoring platform pre-processes the received charging habit-related characteristic data and calculates the SOC segments of multiple different charging behaviors using the ampere-hour integration method;

[0025] Step 3: crawl the four features corresponding to the calculated SOC fragment, including vehicle model related features, battery type related features, single charge related features, and vehicle environment related features, to form a data strip of the vehicle model containing the above five features;

[0026] Step 4: Perform kmeans clustering and linear regression algorithm training on the data strips to obtain a partial correspondence between the cumulative mileage and the battery capacity of the vehicle model;

[0027] For the complete correspondence between accumulated mileage and battery capacity, compensation is performed for missing parts other than the partial correspondence, including: if there is complete data of the correspondence in a certain class after clustering, the complete data in the class is used for compensation; if there is no complete data in the class but there is complete data globally, the global complete data outside the class is used for compensation.

[0028] Step 5: For the discrete points of the relationship between the accumulated mileage and the battery capacity of each compensated data strip, a complete curve of the relationship between the accumulated mileage and the battery capacity is fitted based on linear regression. The ordinate is the kwh of electricity, the abscissa is the kilometers, and the nominal capacity announced by the vehicle is initially determined as the original electricity.

[0029] Step 6: Output the retired battery capacity evaluation result of the target vehicle according to the fitting curve and the actual vehicle data of the target vehicle of the same model;

[0030] Step 7: Perform an offline battery capacity test on the same vehicle model, compare it with the battery capacity evaluation result obtained in the previous step, and make feedback adjustments to the fitted curve based on the comparison result. In a preferred embodiment of the present invention, the charging habit-related features X are defined respectively. charge-related 、Single charge related features X battery-related 、Environment-related characteristics X environment , charging start time t start , charging end time t end , instantaneous charging voltage U, instantaneous charging current I, time t, perform the following clustering process:

[0031]

[0032] Group=arg min∑||xc||,x∈[x charge-related , x battery-related , x environment ]

[0033] y2=f(y1,K∈Group)

[0034] arg min||y2-(w×M±b)||

[0035] Among them, x represents the data strip sample, M is the cumulative mileage, c is the class center, w is the decay rate, K is the cluster category, f is the selected clustering rule, and b is a constant.

[0036] Since the same brand of vehicle batteries are widely used in the market for different types of batteries due to vehicle models, grades, configuration levels, etc., the environment in which the vehicle is located, such as the annual temperature and traffic congestion, will have different degrees of influence on the battery capacity. Therefore, it is necessary to comprehensively consider these influencing factors to improve the accuracy of capacity estimation. Therefore, in a preferred embodiment of the present invention, the collected battery type-related features include: battery material, capacity density, manufacturer and other characteristic parameters; vehicle model-related features include but are not limited to vehicle model classification features such as sedans, SUVs, trucks, vehicle grades, vehicle weights, different configurations of the same vehicle model and other characteristic parameters; the single charge-related features include: mileage after the current charge, charging rate, charging depth, charging time, and ambient temperature during the current charge.

[0037] The characteristics related to charging habits include: the number of days the battery has been put into use, the mileage difference of the vehicle after being put into use, the total number of charging times, the total charging capacity, the average single charging capacity, the total charging hours, the average charging rate, the average charging start SOC, the average charging end SOC, the average charging SOC depth, the median SOC at the start of charging, the median SOC at the end of charging, and the median charging SOC depth, 13 characteristic parameters; the characteristics related to the vehicle's environment include: 5 characteristic parameters: the annual average temperature of the environment, the annual median temperature, the annual temperature variance, the annual temperature range, and the number of effective temperatures.

[0038] In a preferred embodiment of the present invention, the preprocessing in step 2 and step 6 both include preliminary elimination and sorting of vehicle models based on the vehicle identification code VIN and the legal status of the vehicle.

[0039] In actual operation, since the complete data strips reflecting the corresponding relationship may exist in normal clustered samples and may also appear in a small amount in outliers, in order to compensate more flexibly, in a preferred embodiment of the present invention, the compensation in step four specifically includes: first removing outliers from the clustered data strips and then taking the mean of the complete data for compensation, or first taking the mean of the complete data strips after clustering for compensation and then removing outliers.

[0040] In step 5, the corresponding relationship curve between cumulative mileage and battery capacity based on linear regression fitting is specifically fitted by a univariate linear model, with cumulative mileage as the independent variable and battery capacity as the dependent variable, outputting the law of battery capacity changing with attenuation; and / or using some high-quality vehicle samples to perform linear fitting. In some cases, if there is no complete data within the class, it can be further compensated by the complete data of other models with the same battery-related characteristics and the same model vehicles in similar environments.

[0041] Figure 2 The corresponding relationship curve between cumulative mileage and battery capacity based on linear regression fitting implemented by the present invention is shown. The ordinate is the power in kWh, and the abscissa is the kilometers. The nominal capacity announced by the vehicle is initially determined as the original power. The top and bottom two of the three curves in the figure define the acceptable boundaries of each discrete point in the figure, and the middle curve is finally fitted.

[0042] In the example based on the present invention, taking the BYD e5-300 real vehicle as an example, an offline real vehicle battery capacity charging test was carried out. The offline test evaluation capacity of the whole vehicle battery pack was 40.73kwh, and the online capacity big data evaluation result was 40.9333kwh, which met the initial goal. For this single vehicle, the online evaluation result has high accuracy.

[0043] For large quantities of power batteries that are about to be retired, such as models registered between January 2017 and December 2020, and models with a single charge SOC range of 50% to 80% between July 2019 and December 2020, after the battery type-related characteristics and model-related characteristics are input into the system to match accurate battery information, the algorithm can accurately estimate the current battery capacity based on other characteristics of the vehicle.

[0044] It should be understood that the size of the serial numbers of the steps in the embodiment of the present invention does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0045] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A power battery capacity algorithm based on big data technology evaluation, characterized by: The specific steps include: Step 1: Use the in-vehicle intelligent terminal to collect the charging habits of a specific vehicle model and upload it to the new energy vehicle big data monitoring platform; Step 2: The monitoring platform pre-processes the received charging habit-related characteristic data and calculates the SOC segments of multiple different charging behaviors using the ampere-hour integration method; Step 3: crawl the four features corresponding to the calculated SOC fragment, including vehicle model related features, battery type related features, single charge related features, and vehicle environment related features, to form a data strip of the vehicle model containing the above five features; Step 4: Perform kmeans clustering and linear regression algorithm training on the data strips to obtain a partial correspondence between the cumulative mileage and the battery capacity of the vehicle model; The clustering process specifically includes: defining charging habit related features x charge-related , Single charge related features x battery-related 、Environment-related characteristics x environment , charging start time t start , charging end time t end , instantaneous charging voltage U, instantaneous charging current I, time t, perform the following clustering process: Group=argmin∑||x-c||,x∈[x charge-related ,x battery-related ,x environment ] y2=f(y1,K∈Group) arg min||y2-(w×M±b)|| Among them, x represents the data strip sample, M is the cumulative mileage, c is the cluster center, w is the decay rate, f is the selected clustering rule, K is the clustering category, and b is a constant; Compensating for missing parts other than the partial correspondence in the complete correspondence between accumulated mileage and battery capacity, including: if there is complete data of the complete correspondence in a certain class after clustering, using the complete data in the class for compensation; if there is no complete data in the class but there is complete data globally, using the global complete data outside the class for compensation; Step 5: For the discrete points of the relationship between the cumulative mileage and the battery capacity of each compensated data strip, a complete curve of the relationship between the cumulative mileage and the battery capacity is fitted based on linear regression; Step 6: Output the retired battery capacity evaluation result of the target vehicle according to the fitting curve and the actual vehicle data of the target vehicle of the same model; Step 7: Perform an offline battery capacity test on the same vehicle model, compare it with the battery capacity evaluation result obtained based on the previous steps, and make feedback adjustments to the fitted curve based on the comparison result.

2. The algorithm according to claim 1, characterized in that: The collected battery type-related features include: battery material, capacity density, and manufacturer characteristic parameters; vehicle model-related features include sedan, SUV, and truck model classification features, vehicle grade, vehicle weight, and different configuration characteristic parameters of the same model; the single charge-related features include: mileage after the current charge, charging rate, charging depth, charging time, and ambient temperature during the current charge; charging habit-related features include: the number of days the battery has been put into use, the mileage difference of the vehicle after being put into use, the total number of charging times, the total charging capacity, the average single charging capacity, the total charging hours, the average charging rate, the average charging start SOC, the average charging end SOC, the average charging SOC depth, the median SOC at the start of charging, the median SOC at the end of charging, and the median charging SOC depth, 13 characteristic parameters; the vehicle environment-related features include: the annual average temperature of the environment, the annual median temperature, the annual temperature variance, the annual temperature range, and the number of effective temperatures.

3. The algorithm of claim 1, wherein: The preprocessing in step 2 and step 6 both include preliminary elimination and sorting of vehicle models based on the vehicle identification code VIN and the legal status of the vehicle.

4. The algorithm of claim 1, wherein: The compensation in step 4 specifically includes: first removing outliers from the clustered data strips and then taking the mean of the complete data for compensation, or first taking the mean of the complete data strips after clustering for compensation and then removing outliers.

5. The algorithm of claim 1, wherein: In step 5, the corresponding relationship curve between cumulative mileage and battery capacity based on linear regression fitting is specifically fitted by a univariate linear model, with cumulative mileage as the independent variable and battery capacity as the dependent variable, and the law of battery capacity change with attenuation is output; and / or some high-quality vehicle samples are used as a backup to perform linear fitting.

Images