Capacity attenuation trajectory prediction method of automobile power battery

Abstract

The invention relates to the technical field of capacity prediction of automobile power batteries, in particular to a capacity attenuation trajectory prediction method of an automobile power battery, which comprises the following steps: S1, acquiring historical operation data of in-service or out-of-service batteries of the same model; s2, screening historical operation data and processing the historical operation data to obtain a label value; s3, comparing historical operation data according to the label values, calculating corresponding battery characteristics, and establishing a database; s4, establishing a power battery health state prediction model through the database; and S5, predicting a battery capacity attenuation trajectory by using the prediction model. The invention aims to overcome the defect that the attenuation prediction of the capacity of the power battery is inconvenient in the prior art, and provides the capacity attenuation trajectory prediction method of the automobile power battery, which is simple in operation process and high in universality, can be applied to new energy automobile power battery data under actual working conditions, and can be used for predicting the capacity attenuation trajectory of the automobile power battery. Compared with a traditional battery life prediction method, the method is more practical.

Description

Technical Field

[0001] The present invention relates to the technical field of capacity prediction of automotive power batteries, and more specifically, to a method for predicting a capacity attenuation trajectory of an automotive power battery. Background Art

[0002] Lithium-ion batteries currently have a wide range of applications, including consumer electronics, electric vehicles, and energy storage. During battery development, the cycle life test period is too long to meet the rapid iteration development process of power products. In addition, during battery use, the battery life varies in real time with different environments and methods, and the unpredictability of its capacity decay will bring serious product safety issues.

[0003] Power batteries are an important part of plug-in hybrid electric vehicles. Accurately predicting the remaining life of power batteries can help companies make reasonable decisions on when to replace power batteries, which is conducive to improving the reliability and economic benefits of electric vehicles. The actual service life of the battery is more affected by the conditions of use. The main factors include workload, operating temperature, depth of discharge (DOD) and charging method. In addition, the differences in power system configurations, vehicle driving conditions and climatic conditions of different types of electric vehicles lead to significant differences in the working environment of power batteries during actual use; coupled with the differences in battery types and capacities, conventional battery life test methods cannot meet the needs of battery life characteristics research.

[0004] The existing patent discloses a method, device and storage medium for online rapid prediction of vehicle power battery life. The method obtains the internal parameters of the battery and determines the key electrochemical parameters; constructs a battery electrochemical parameter estimation model based on a neural network, uses the loading current and terminal voltage as features, and uses the key electrochemical parameters as labels for training; constructs a lithium-ion battery life prediction model for online rapid prediction of life: uses the battery electrochemical parameter estimation model to process the data set to obtain the parameter change trend of the electrochemical model; classifies the battery degradation mode according to the key electrochemical parameter change trend and the inflection point and life length, and establishes a battery life prediction model respectively; integrates different battery life prediction models to obtain a comprehensive life online prediction model. Although this method has high prediction accuracy and strong interpretability, it requires an electrochemical model that accurately reflects the internal potential of the battery and the distribution of lithium ion concentration, which is difficult to achieve in actual battery applications. Summary of the invention

[0005] The purpose of the present invention is to overcome the deficiency of the prior art that it is not convenient to predict the attenuation of the capacity of a power battery, and to provide a method for predicting the capacity attenuation trajectory of a vehicle power battery. The method has a simple calculation process and strong universality, and can be applied to the power battery data of new energy vehicles under actual working conditions. The method is more practical than traditional battery life prediction methods.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0007] A method for predicting the capacity attenuation trajectory of a vehicle power battery is provided, which specifically comprises the following steps:

[0008] S1: Obtain historical operating data of in-service or retired batteries of the same model;

[0009] S2: Filter and process historical operation data to obtain label values;

[0010] S3: Compare historical operation data according to tag values, calculate corresponding battery characteristics, and establish a database;

[0011] S4: Establish a power battery health status prediction model through the database;

[0012] S5: Use the prediction model to predict the battery capacity decay trajectory.

[0013] The capacity attenuation trajectory prediction method of the automotive power battery of the present invention can be applied to the capacity attenuation trajectory prediction of the power battery of a plug-in hybrid electric vehicle. First, the features related to the daily working state of the battery are extracted from the historical operation data of the same type of battery, and processed to obtain the label value. The feature engineering technology is used to perform correlation analysis and feature selection on the data set to establish a database; after the power battery health status prediction model is established, the parameters are optimized, and finally the characteristic values ​​of the target vehicle are input to use the prediction model to realize the prediction of the battery capacity of the target vehicle. The ability to accurately predict the remaining life of the power battery can help companies make reasonable decisions on when to replace the power battery, which is conducive to improving the reliability and economic benefits of electric vehicles. The calculation process is simple and has strong universality. It can be applied to the power battery data of new energy vehicles under actual working conditions, which is more practical than traditional battery life prediction methods.

[0014] Preferably, in step S2, historical operation data is screened by extracting each parking charging data in non-operation mode, the capacity value is calculated by using the ampere-hour integration method, and the final label value is determined by further smoothing and curve fitting.

[0015] Preferably, the calculation formula of the ampere-hour integration method is as follows:

[0016]

[0017] Among them, t1 is the starting time of the parking charging phase, SOC t1 is the SOC value corresponding to time t1, t2 is the end time of the parking charging phase, SOC t2 is the SOC value corresponding to time t2; I(t) is the current value at time t.

[0018] Preferably, in step S3, historical operation data before the time corresponding to each tag value in step S2 is extracted to calculate the corresponding battery characteristics.

[0019] Preferably, the battery characteristics include distribution characteristics that reflect the distribution of the battery's operating voltage, current, state of charge, and temperature state; cumulative characteristics that gradually increase the battery's cumulative usage time, number of parking charges, cumulative cruising range, cumulative charging capacity, and cumulative discharge capacity as the vehicle is continuously used; and the temperature at the corresponding moment of the battery label capacity.

[0020] Preferably, the distribution feature adopts a data discretization method to convert continuous data into discrete feature values. Specifically, first determine the extreme value of the battery state parameter, divide a plurality of state intervals according to the extreme value, divide the state parameters of all historical operating data before the corresponding moment of the tag capacity into corresponding state intervals, and calculate the proportion of the duration of the historical operating data in the interval to the entire historical operating data cycle.

[0021] Preferably, the battery status historical operation data of the operating voltage, current, state of charge and temperature are divided into multiple state intervals, and the number of intervals n≥3. The physical meaning is to characterize the operating frequency of the battery in a certain interval state. The obtained interval feature can be expressed as: feature(state)=[segment1,segment2……,segment n], where State represents the voltage, current, state of charge and temperature state of the battery, and segment n represents the operating frequency in a certain interval state.

[0022] Preferably, in step S4, the database is split into a training set and a test set, and the model is trained and evaluated by combining feature engineering, cross-validation, and model parameter tuning methods to obtain the optimal machine learning model. The obtained machine learning model can predict the maximum available capacity of the battery at the current moment.

[0023] Preferably, the historical operation data includes data collection time, vehicle status, charging status, operation mode, total voltage, total current, SOC, maximum value of battery cell voltage, minimum value of battery cell voltage, maximum temperature value, and minimum temperature value.

[0024] Preferably, in step S5, when the car owner's car-using habits and the working status of the battery remain unchanged, and the cumulative variables of the vehicle's cruising range, number of parking charges, and cumulative working time gradually increase, the characteristic parameters of the vehicle at different times in the future are preset and input into the model, so as to obtain the battery capacity at different times in the future, thereby realizing the prediction of the battery capacity attenuation trajectory.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] 1. Combining the power battery data of new energy vehicles under actual working conditions is more practical and the prediction process is more convenient.

[0027] 2. Discretizing continuous floating-point data can significantly reduce the amount of calculation and increase the training speed; on the other hand, it can significantly improve the overfitting phenomenon caused by noise.

[0028] 3. A feature set including different operating temperatures, different operating voltages, different operating currents and different operating state of charge (SOC) intervals was established to reflect the differences in working conditions of different batteries.

[0029] 4. A machine learning model of features and battery health status was established, which can predict the battery health status at different stages in the future. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 A flow chart of a method for predicting the capacity attenuation trajectory of a vehicle power battery;

[0031] Figure 2 This is the SOC change curve of different vehicles.

[0032] Figure 3 The voltage variation curves of different vehicles.

[0033] Figure 4 The following are the current variation curves of different vehicles.

[0034] Figure 5 This is a curve diagram of the average operating temperature changes of different vehicles.

[0035] Figure 6 This is a scatter plot of the capacity versus the number of parking charges.

[0036] Figure 7 This is the main attenuation trend line chart derived from the empirical mode decomposition.

[0037] Figure 8 for Figure 7 The plot is interpolated in the corresponding time series.

[0038] Fig. 9This is the normalized graph of the main capacity attenuation trend line of different vehicles.

[0039] Fig.10 It is a curve diagram showing the change of the discharge current interval distribution of the first vehicle over time.

[0040] Fig.11 It is a curve diagram showing the change of the charging current interval distribution of the first vehicle over time.

[0041] Fig.12 It is a curve diagram showing the change of the discharge current interval distribution of the second vehicle over time.

[0042] Fig.13 It is a curve diagram showing the change of the charging current interval distribution of the second vehicle over time.

[0043] Fig.14 This is the correlation coefficient distribution diagram.

[0044] Fig.15 This is a decay curve diagram showing the true value of the maximum available capacity of the third vehicle and the value predicted by the machine learning model over time.

[0045] Fig.16 This is a decay curve diagram showing the true value of the maximum available capacity of the fourth vehicle and the value predicted by the machine learning model over time.

[0046] Fig.17 Capacity decay trajectory predicted by the machine learning model for the fifth vehicle.

[0047] Fig.18 Capacity decay trajectory predicted by the machine learning model for the sixth vehicle.

[0048] Fig.19 Flowchart for training a machine learning model. DETAILED DESCRIPTION

[0049] The present invention is further described below in conjunction with specific implementation methods. The accompanying drawings are only used for exemplary descriptions and are only schematic diagrams, not actual drawings, and cannot be understood as limiting this patent; in order to better illustrate the embodiments of the present invention, some parts of the accompanying drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product; for those skilled in the art, it is understandable that some well-known structures and their descriptions in the accompanying drawings may be omitted.

[0050] Embodiment 1

[0051] like Figure 1 As shown, this embodiment is a first embodiment of a method for predicting the capacity decay trajectory of a vehicle power battery, which specifically includes the following steps:

[0052] S1: Obtain historical operating data of in-service or retired batteries of the same model;

[0053] Historical operation data includes data collection time, vehicle status, charging status, operation mode, total voltage, total current, SOC, maximum value of battery cell voltage, minimum value of battery cell voltage, maximum value of temperature, and minimum value of temperature;

[0054] S2: Filter and process historical operation data to obtain label values;

[0055] By extracting each parking charging data in non-operating mode to screen historical operating data, the capacity value is calculated using the ampere-hour integration method, and the final label value is determined through further smoothing and curve fitting; the calculation formula of the ampere-hour integration method is as follows:

[0056]

[0057] Among them, t1 is the starting time of the parking charging phase, SOC t1 is the SOC value corresponding to time t1, t2 is the end time of the parking charging phase, SOC t2 is the SOC value corresponding to time t2; I(t) is the current value at time t.

[0058] The smoothing methods used include empirical mode decomposition (EMD) and K-nearest neighbor algorithm (KNN). The KNN algorithm is mainly used to remove outliers that deviate greatly from the rated capacity. The EMD method is mainly used to obtain the changing trend of battery capacity over time. The EMD method can be implemented through the PyEMD library in the Python computer programming language or the empirical mode decomposition emd function in the matrix laboratory Matlab. The calculated capacity value sequence is input into the function to return the decomposition result of the sequence. The KNN method specifically uses the KNN algorithm in the Pyod library in the Python language to detect outliers of label values.

[0059] S3: Compare historical operation data according to tag values, calculate corresponding battery characteristics, and establish a database;

[0060] Extract the historical operation data before the time corresponding to each tag value in step S2, and calculate the corresponding battery characteristics;

[0061] Battery characteristics include distribution characteristics that reflect the distribution of the battery's operating voltage, current, state of charge (SOC), and temperature status; cumulative characteristics that reflect the gradual increase in the battery's cumulative usage time, number of parking charges, cumulative mileage, cumulative charging capacity, and cumulative discharge capacity as the vehicle is continuously used; and the temperature at the time corresponding to the battery label capacity;

[0062] The distribution feature uses a data discretization method to convert continuous data into discrete feature values. Specifically, the extreme values ​​of battery state parameters such as operating voltage, current, SOC, and temperature are first determined, and multiple state intervals are divided according to the extreme values. The state parameters of all historical operating data before the corresponding moment of the tag capacity are divided into corresponding state intervals, and the proportion of the duration of the historical operating data in the interval to the entire historical operating data cycle is calculated;

[0063] The eigenvalues ​​are obtained as follows:

[0064] The battery status historical operation data of working voltage, current, state of charge, and temperature are divided into multiple state intervals, and the number of intervals n≥3. Its physical meaning is to characterize the working frequency of the battery in a certain interval state. The obtained interval feature can be expressed as: feature(state)=[segment1, segment2……, segment n], where State represents the voltage, current, state of charge, and temperature state of the battery, and segment n represents the working frequency in a certain interval state, that is, the ratio of the data duration in the interval to the historical data duration;

[0065] S4: Establish a power battery health status prediction model through the database;

[0066] The database is split into a training set and a test set. The model is trained and evaluated by combining feature engineering, cross-validation, and model parameter tuning methods to obtain the optimal machine learning model. The obtained machine learning model can predict the capacity of the battery at the current moment. The battery capacity referred to is the maximum available capacity of the battery.

[0067] Feature engineering mainly includes correlation analysis and feature screening steps; correlation analysis mainly uses the Pearson correlation coefficient to calculate the correlation between features and features, and between features and label values.

[0068] Feature screening is mainly done by setting a threshold of the correlation coefficient to screen out features that are highly correlated with the label value and exclude features with strong collinearity between features.

[0069] The capacity of the battery at the current moment is predicted by calculating the characteristic value corresponding to the current moment and inputting it into the prediction model to obtain the capacity prediction value.

[0070] S5: Use the prediction model to predict the battery capacity decay trajectory.

[0071] When the owner's driving habits and the working status of the battery remain unchanged, that is, the distribution of the battery's voltage, current, SOC, temperature and other states are consistent, and the vehicle's cumulative variables gradually increase, that is, the cumulative characteristics, that is, the cumulative variables such as cruising range, number of parking charges, and cumulative working time gradually increase, by pre-setting the characteristic parameters of the vehicle at different times in the future and inputting them into the model, the standard battery capacity at different times in the future can be obtained, thereby realizing the prediction of the battery capacity decay trajectory.

[0072] The cumulative characteristics are calculated by using historical operating data to calculate the unit time growth rate of the cumulative characteristics. Therefore, the cumulative characteristics at different moments in the future are the cumulative characteristic growth rate of the battery multiplied by the battery usage time.

[0073] The temperature at the time when the battery tag capacity corresponds to the temperature, assuming that the temperature values ​​at different times in the future maintain the same change pattern as the temperature in the current vehicle's historical operation data as the seasons change. Perform function fitting on the temperature change over time in the historical operation data, and calculate the temperature values ​​at different time periods in the future as the input feature values.

[0074] Embodiment 2

[0075] like Figures 2 to 18 As shown, this embodiment is the second embodiment of a method for predicting the capacity decay trajectory of a vehicle power battery. This embodiment is similar to the first embodiment, except that this example includes the working data of 25 PHEV plug-in hybrid vehicle power batteries in use, and its original data columns include time, vehicle status, charging status, operating mode, speed, mileage, total voltage, total current, SOC, single cell maximum voltage, single cell minimum voltage, single cell maximum temperature, single cell minimum temperature, etc. Among them, the SOC, voltage, current, temperature change curve Figures 2 to 5 Different vehicles have different actual working conditions during actual use, and therefore show different battery working data information. This embodiment provides a method for predicting the capacity decay trajectory of a plug-in hybrid electric vehicle power battery, the method comprising:

[0076] S1: Obtain historical operating data of 25 vehicle batteries;

[0077] Including time, vehicle status, charging status, operating mode, speed, mileage, total voltage, total current, SOC, single cell maximum voltage, single cell minimum voltage, single cell maximum temperature, single cell minimum temperature and other data;

[0078] S2: Filter and process historical operation data to obtain label values;

[0079] Acquire parking charging data; use the ampere-hour integration method to calculate the maximum available capacity as the tag value.

[0080] Use the unsupervised learning algorithm KNN model in the PyOD library to separate outliers;

[0081] like Figure 6 As shown, most of the calculated capacity values ​​are distributed near the decay trajectory, and outliers can be effectively identified by the KNN algorithm;

[0082] The empirical mode decomposition technique is used to decompose the capacity time series into the main decay trend line and the fluctuation series;

[0083]

[0084] Among them, SOH(t) is the battery health state (State of Health, SOH) at time t, N is the number of decomposed IMF components, IMF (Intrinsitc Mode Function) is the intrinsic mode function, reflecting the main attenuation trend line of battery capacity, and R(t) is the margin. Figure 7 and Figure 8 , Figure 7 is the decomposed main decay trend line, Figure 8 To interpolate the corresponding time series, Fig. 9 It is the normalized main attenuation trend line of battery capacity of different vehicles, which is used as the label value.

[0085] S3: Compare historical operation data according to tag values, calculate corresponding battery characteristics, and establish a database;

[0086] Feature engineering, in which the state distribution characteristics of the battery are to bin the series of current, temperature, SOC, etc., and calculate the time proportion of different series in the divided intervals. The cumulative characteristics of the battery state include the number of parking charging times, the number of discharge times, the number of hybrid charging times, the cumulative SOC change of parking charging, the cumulative SOC change of the discharge stage, the cumulative SOC change of hybrid charging, the cumulative number of working days, the cumulative number of months, etc.

[0087] Among them, the current series intervals are divided as follows: (-inf, -1C], (-1C, -0.5C], (-0.5C, -0.1C], (-0.1C, 0), 0, (0, 0.1C], (0.1C, 0.5C], (0.5, 1C], (1C, inf), 9 intervals, respectively recorded as DCR-Ps4, DCR-Ps3, DCR-Ps2, DCR-Ps1, DCR-P0, DCR-Pd1, DCR-Pd2, DCR-Pd3, DCR-Pd4, calculate the frequency distribution of each interval, 1C = 30A.

[0088] Among them, the temperature series intervals are divided into four intervals: [0℃, 15℃), [15℃, 30℃), [30℃, 45℃), [45℃, inf), respectively recorded as Dtep-Pt1, Dtep-Pt2, Dtep-Pt3, Dtep-Pt4, and the frequency distribution of each interval is calculated.

[0089] The SOC sequence intervals are divided into five intervals: [0, 20%), [20%, 40%), [40%, 60%), [60%, 80%), and [80%, 100%), respectively denoted as SOCUSE-D1, SOCUSE-D2, SOCUSE-D3, SOCUSE-D4, and SOCUSE-D5. The frequency distribution of each interval is calculated.

[0090] like Figures 10 to 13 Curves showing the change of different current interval distributions over time: Fig.10 and Fig.11 is the curve of the current interval distribution corresponding to the first vehicle over time, Fig.12 and Fig.13 is a curve showing the change of the current interval distribution corresponding to the second vehicle over time.

[0091] S4: Establish a power battery health status prediction model through the database;

[0092] The database is split into a training set and a test set. The model is trained and evaluated by combining feature engineering, cross-validation, and model parameter tuning methods to obtain the optimal machine learning model. The obtained machine learning model can predict the capacity of the battery at the current moment. The battery capacity referred to is the maximum available capacity of the battery.

[0093] Calculate the correlation between feature values ​​and label values, perform feature screening, and then train the model.

[0094] Among them, the correlation is evaluated by the Pearson correlation coefficient, and the formula is as follows:

[0095]

[0096] Where x is the eigenvalue, is the average of the feature values, y is the label value, Take the average of the label values.

[0097] Fig.14 This is the correlation coefficient distribution diagram.

[0098] In this example, the elastic net model is used for training and model evaluation. The model evaluation indicators are mae and RMSE. The formula is as follows:

[0099]

[0100]

[0101] Among them, mae (relative) is the average percentage error, y is the true value, is the predicted value and rmse(relative) is the root mean square percentage error.

[0102] The relative values ​​of MAE and RMSE are both within 2%, and the relative values ​​of more than half of the vehicles are less than 1%. The current model has a high prediction accuracy and can accurately predict the maximum available capacity of the battery.

[0103] S5: Use the prediction model to predict the battery capacity decay trajectory;

[0104] Construct the state characteristics of the target vehicle in the future and input them into the prediction model to obtain the battery capacity values ​​of the target vehicle at different times in the future.

[0105] Among them, the subsequent state feature estimation method is: distribution-type features, which are considered to keep the value of the last month unchanged. Cumulative-type features, linear fitting of cumulative-type features-accumulated months, and then use the fitting results for extension.

[0106] like Figures 15 to 18 As shown in the figure, the cumulative number of months is extended to 120 months, and the third, fourth, fifth, and sixth vehicles are taken as examples to estimate their subsequent capacity decay. If the user's behavior habits do not change, then according to the current estimated value, the capacity decay rate will not exceed 20% after 120 months.

[0107] Embodiment 3

[0108] like Fig.19 As shown, this embodiment is a third embodiment of a method for predicting the capacity attenuation trajectory of a vehicle power battery. This embodiment is similar to the second embodiment, except that step S4 also includes evaluating the performance of the prediction model, which is evaluated through machine learning model training. The specific process is as follows:

[0109] S41: The data of one vehicle is used as the validation set, and the data of the remaining vehicles are used as the training set;

[0110] S42: Calculating a label (capacity retention rate) and a feature from historical operation data of the original vehicle battery;

[0111] S43: Input training set labels and features to train the model;

[0112] S44: Input the validation set features into the trained model to obtain the predicted value, compare the predicted value with the validation set label, and evaluate the performance of the prediction model.

[0113] During the whole process, the validation set and the predicted value are cross-validated to improve and correct the accuracy of the prediction model.

[0114] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A method for predicting the capacity attenuation trajectory of a vehicle power battery, characterized in that: The specific steps include: S1: Obtain historical operating data of in-service or retired batteries of the same model; S2: Filter and process historical operation data to obtain label values; S3: Compare historical operation data according to tag values, calculate corresponding battery characteristics, and establish a database; S4: Establish a power battery health status prediction model through the database; S5: Use the prediction model to predict the battery capacity decay trajectory.

2. The method for predicting the capacity attenuation trajectory of a vehicle power battery according to claim 1, characterized in that: In step S2, historical operation data is screened by extracting each parking charging data in non-operation mode, the capacity value is calculated by using the ampere-hour integration method, and the final label value is determined by further smoothing and curve fitting.

3. The method for predicting the capacity attenuation trajectory of a vehicle power battery according to claim 2, characterized in that: The calculation formula of the ampere-hour integration method is as follows: Among them, C is the battery capacity, t1 is the start time of the parking charging phase, SOC t1 is the SOC value corresponding to time t1, t2 is the end time of the parking charging phase, SOC t2 is the SOC value corresponding to time t2; I(t) is the current value at time t.

4. The method for predicting the capacity attenuation trajectory of a vehicle power battery according to claim 1, characterized in that: In step S3, the historical operation data before the time corresponding to each tag value in step S2 is extracted to calculate the corresponding battery characteristics.

5. The method for predicting the capacity attenuation trajectory of a vehicle power battery according to any one of claims 1 to 4, characterized in that: The battery characteristics include distribution characteristics reflecting the distribution of the battery's operating voltage, current, state of charge (SOC), and temperature state; As the vehicle is continuously used, the cumulative battery usage time, number of charging stops, cumulative mileage, cumulative charging capacity, and cumulative discharge capacity gradually increase; as well as the temperature at the corresponding moment of the battery label capacity.

6. The method for predicting the capacity attenuation trajectory of a vehicle power battery according to claim 5, characterized in that: The distribution feature adopts a data discretization method to convert continuous data into discrete feature values. Specifically, the extreme value of the battery state parameter is first determined, and multiple state intervals are divided according to the extreme value. The state parameters of all historical operating data before the corresponding moment of the tag capacity are divided into corresponding state intervals, and the proportion of the duration of the historical operating data in the interval to the entire historical operating data cycle is calculated.

7. The method for predicting the capacity attenuation trajectory of a vehicle power battery according to claim 6, characterized in that: The battery status historical operation data of working voltage, current, state of charge and temperature are divided into multiple state intervals, and the number of intervals n≥3. Its physical meaning is to characterize the working frequency of the battery in a certain interval state. The obtained interval feature can be expressed as: feature(state)=[segment1,segment2……,segment n], where State represents the voltage, current, state of charge and temperature state of the battery, and segment n represents the working frequency in a certain interval state.

8. The method for predicting the capacity attenuation trajectory of a vehicle power battery according to claim 1, characterized in that: In step S4, the database is split into a training set and a test set, and the model is trained and evaluated by combining feature engineering, cross-validation, and model parameter tuning methods to obtain the optimal machine learning model. The obtained machine learning model can predict the battery capacity at the current moment.

9. The method for predicting the capacity attenuation trajectory of a vehicle power battery according to claim 1, characterized in that: The historical operation data includes data collection time, vehicle status, charging status, operation mode, total voltage, total current, charge status, maximum value of battery cell voltage, minimum value of battery cell voltage, maximum temperature value, and minimum temperature value.

10. The method for predicting the capacity attenuation trajectory of a vehicle power battery according to claim 1, characterized in that: In step S5, when the car owner's car-using habits and the working status of the battery remain unchanged, and the accumulated variables of the vehicle's cruising range, the number of parking charges, and the accumulated working time gradually increase, the characteristic parameters of the vehicle at different times in the future are preset and input into the prediction model, so as to obtain the battery capacity at different times in the future and realize the prediction of the battery capacity attenuation trajectory.

Images