Electric vehicle decommissioning prediction method based on use requirements

Abstract

The invention provides an electric vehicle decommissioning prediction method based on use requirements, which comprehensively considers possible influence of factors such as static indexes of a vehicle, various states in actual operation, charging and driving behaviors and habits of a user and the like on decommissioning of different types of vehicles. An aging score based on Weibull distribution and a multi-dimensional travel demand quantification means are particularly provided; according to the method, a training set is established through original data collection and preprocessing, feature parameter extraction and feature screening in sequence, then a decommissioning prediction model is established based on a machine learning algorithm, the prediction model is trained through the established training set, and the effect of the prediction model is verified. And after the model is finally trained, the model is applied online, so that the decommissioning state of the vehicle can be predicted more objectively and accurately, advanced planning of later use, decommissioning and recycling of the vehicle and the battery is facilitated, and compared with the prior art, the method has wider practicability.

Description

Technical Field

[0001] The present invention belongs to the technical field of performance and life evaluation prediction of electric vehicles and power batteries, and specifically relates to a method for predicting the retirement of electric vehicles based on usage demand. Background Art

[0002] For new energy vehicles that are about to reach the end of their life, the field currently uses a fixed threshold of power battery capacity or internal resistance decay as the end of life to evaluate and predict its remaining service life. For example, at the capacity level, the battery available capacity decays to 80% of the initial capacity as the end of life, and at the internal resistance level, the battery internal resistance increases to 200% of the initial internal resistance as the end of life. However, due to the single evaluation dimension, this method of determining the end of life using a fixed threshold is not suitable for the complex and changeable operating conditions of real vehicles. It lacks consideration of the differences in usage habits and usage environments of different users. In actual vehicle use, the impact of the vehicle's purpose, type, economy, functionality, etc. on the final retirement decision of the vehicle should not be ignored. On the other hand, with the current development trend of the continuous improvement of the capacity of automotive power batteries, the continued use of a fixed threshold retirement judgment method gradually fails to meet the sustainable utilization requirements of electric vehicles, and cannot provide a more accurate remaining life prediction result based on the characteristics of each vehicle, which may cause insufficient battery use and early retirement of vehicles. Therefore, at this stage, the field urgently needs to develop a method for predicting the retirement status of electric vehicles that can achieve accurate prediction of single vehicles based on real operating data and comprehensive retirement factors from multiple aspects. Summary of the invention

[0003] In view of this, in view of the technical problems existing in the art, the present invention provides an electric vehicle retirement prediction method based on usage demand, which specifically includes the following steps:

[0004] Step 1: Collect static data related to the vehicles and batteries of retired electric vehicles of different uses and types, as well as actual vehicle operation data within a certain period before retirement; perform preprocessing on the collected data, including setting abnormal values ​​to null values, deleting or averaging duplicate values, and filling missing values, and establish corresponding retired vehicle static data sets and actual vehicle operation data sets;

[0005] Step 2: obtain the service life and battery rated energy of each vehicle by processing static data, and distinguish different vehicle types to fit the corresponding vehicle service life Weibull distribution probability density function respectively; calculate the aging scores of different types of vehicles using the probability density function; define the service life and aging score as vehicle dimension characteristic parameters;

[0006] Step 3: Process the actual vehicle operation data to obtain the segmented charging data segments and driving data segments; extract the charging state characteristic parameters including battery capacity, single cell voltage, charging time, cumulative mileage, average temperature and average charging current from the charging data segments; extract the driving state characteristic parameters including full-charge driving range, cumulative mileage, average current of driving segments, average speed and energy consumption rate from the driving data segments; define the charging state characteristic parameters and driving state characteristic parameters together as state dimension characteristic parameters;

[0007] Step 4: Set corresponding quantile ratios for single-day travel, single-trip travel, and travel between two charges in driving behavior, and count the corresponding travel mileages that meet the quantile ratios in different situations; extract charging habit features including daily charging amount and safe SOC retained after each trip for charging behavior, and extract vehicle inactivity features including the number of vehicle inactivity times, total inactivity time, and time interval between two inactivity times within a specific period; define habit features and vehicle inactivity features as usage dimension feature parameters;

[0008] Step 5: Match the vehicle dimension, state dimension, and usage dimension feature parameters according to the accumulated mileage, and add corresponding retired and non-retired state labels; filter the feature parameters of each dimension based on the state labels and through correlation analysis; and construct a training set using the retained feature parameters;

[0009] Step 6: Establish a vehicle retirement prediction model based on the machine learning algorithm, use the constructed training set for training, and verify the training effect;

[0010] Step 7: Apply the trained model online to the prediction of electric vehicle retirement, so that it can predict and output the corresponding vehicle retirement status based on the real-time data corresponding to the feature parameters of each dimension.

[0011] Furthermore, the static data collected in step 1 at least includes the vehicle scrapping date, production date, vehicle type, battery type, battery rated parameters, etc. The vehicle use includes passenger private cars, special transport vehicles for freight or special transportation, and commercial buses; the actual vehicle operation data at least includes vehicle speed, mileage, total voltage, total current, etc.;

[0012] Data preprocessing includes setting abnormal data items that exceed the reasonable range to null values; deleting all voltage, current, temperature and other data that are repeated in time when the difference is too large, and taking the average value when the difference is small; and filling missing values ​​with linear interpolation or nearest neighbor interpolation according to the data type.

[0013] Furthermore, in step 2, the service life of each vehicle is calculated using scrapping date and production date data; the battery rated energy is obtained by multiplying the rated voltage and rated capacity in the battery rated parameters; and the Weibull distribution probability density function of passenger cars, special-purpose vehicles and buses is obtained in the following form:

[0014]

[0015] Among them, x is the usage time of the vehicle, k is the scale parameter, and λ is the shape parameter;

[0016] The data with service life less than 1 year and more than 10 years are removed to reduce the influence of outliers, and the least squares method is used to fit different types of vehicles to determine the corresponding scale parameters and shape parameters;

[0017] The aging scores SR of different types of vehicles are obtained:

[0018]

[0019] Furthermore, in step three, the original data is segmented into charging data segments and driving data segments based on the vehicle status, charging status, current, vehicle speed and other data; in case there is still data missing after segmentation, a new data frame is formed by linearly interpolating multiple frames of data; data segments whose total duration is greater than a specific value or whose SOC change is less than a specific value are eliminated.

[0020] Furthermore, the specific process of extracting the battery capacity from the charging data fragment in step 3 includes:

[0021] First, remove the SOC data at the left and right boundaries of the charging data segment, and then use the partial SOC range to calculate the battery capacity using the following formula:

[0022]

[0023] Where I is the charging current, t0 is the starting time of the charging segment, t1 is the ending time of the charging segment, SOC0 and SOC1 are the SOC values ​​at the starting and ending times of the charging segment respectively;

[0024] The process of extracting the full-charge range from the driving data snippet includes:

[0025] First, the distance that a fully charged vehicle can travel continuously under the current driving conditions is defined as the equivalent full-charge range R. d , using the accumulated mileage and SOC in the driving data segment to calculate:

[0026]

[0027] Among them, R0 and R1 are the cumulative mileage at the beginning and end of the driving segment, respectively, and SOC0 and SOC1 are the SOC values ​​at the beginning and end of the driving segment, respectively;

[0028] The calculation results corresponding to the driving segments with SOC changes greater than 20 are filtered out.

[0029] Furthermore, in step 4, quantile ratios of 98%, 95%, and 85% are set for single-day travel, single trip, and travel between two charges, respectively, and the corresponding travel mileage that meets the quantile ratio in each case is counted;

[0030] The charging volume is specifically based on the daily charging volume counted for every two months of charging data, and the daily charging volume corresponding to the 80% quantile is selected as the corresponding usage dimension feature;

[0031] The vehicle deactivation feature is identified and extracted based on the cumulative mileage and time difference between adjacent driving segments. When the cumulative mileage difference between adjacent driving segments is small but the time difference is greater than 30 days, it is judged that the vehicle has been deactivated. The number of deactivations of each vehicle, the total deactivation duration and the time interval between two deactivations are counted as the corresponding usage dimension features.

[0032] Furthermore, in step 5, the operation data corresponding to the expected cumulative mileage of each retired vehicle greater than 95% of the final mileage and the usage time of the last two months are specifically added with a retired status label 1; and the other data are added with a non-retired status label 0;

[0033] In the correlation analysis of feature parameters, feature screening is completed by calculating the following Pearson correlation coefficient:

[0034]

[0035] Among them, PCC is the characteristic variable x i and retired status y i The Pearson correlation coefficient between and They are x i and i The average value of .

[0036] Furthermore, in step six, any one of the machine learning algorithms selected from random forest, K nearest neighbor, LightGBM, XGBoost, CatBoost, and TabNet is used to establish and train a vehicle retirement prediction model; or two or more models are trained separately, while grid search is used to optimize hyperparameters, and the optimal model is determined through cross-validation; when verifying the training effect, precision, recall, F1-Score, and area under the curve (AUC) are used as evaluation indicators of the model.

[0037] The above-mentioned electric vehicle retirement prediction method based on usage demand provided by the present invention comprehensively considers the possible impact of factors such as the static indicators of the vehicle itself, various states during actual operation, and the charging and driving behaviors and habits of users on the retirement of different types of vehicles, and also specifically proposes an aging score based on Weibull distribution and a multi-dimensional travel demand quantification method; in the method, a training set is established through original data collection and preprocessing, feature parameter extraction, and feature screening, and then a retirement prediction model is established based on a machine learning algorithm, and the prediction model is trained and verified using the established training set. After the model is finally trained, it is applied online, which can more objectively and accurately predict the retirement status of the vehicle, which is conducive to the early planning of the later use, retirement and recycling of vehicles and batteries, and has a wider range of practicality than the existing technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 The overall framework diagram of the method provided by the present invention;

[0039] Figure 2 This is an optional flow chart for data segmentation in the method provided by the present invention. DETAILED DESCRIPTION

[0040] 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.

[0041] The electric vehicle retirement prediction method based on usage demand provided by the present invention is as follows: Figure 1 As shown, the specific steps include:

[0042] Step 1: Collect static data related to the vehicles and batteries of retired electric vehicles of different uses and types, as well as actual vehicle operation data within a certain period before retirement; perform preprocessing on the collected data, including setting abnormal values ​​to null values, deleting or averaging duplicate values, and filling missing values, and establish corresponding retired vehicle static data sets and actual vehicle operation data sets;

[0043] Step 2: obtain the service life and battery rated energy of each vehicle by processing static data, and distinguish different vehicle types to fit the corresponding vehicle service life Weibull distribution probability density function respectively; calculate the aging scores of different types of vehicles using the probability density function; define the obtained service life and aging score as vehicle dimension characteristic parameters;

[0044] Step 3: Process the actual vehicle operation data to obtain the segmented charging data segments and driving data segments; extract the charging state characteristic parameters including battery capacity, single cell voltage, charging time, cumulative mileage, average temperature and average charging current from the charging data segments; extract the driving state characteristic parameters including full-charge driving range, cumulative mileage, average current of driving segments, average speed and energy consumption rate from the driving data segments; define the charging state characteristic parameters and driving state characteristic parameters together as state dimension characteristic parameters;

[0045] Step 4: Set corresponding quantile ratios for single-day travel, single-trip travel, and travel between two charges in driving behavior, and count the corresponding travel mileages that meet the quantile ratios in different situations; extract charging habit features including daily charging amount and safe SOC retained after each trip for charging behavior, and extract vehicle inactivity features including the number of vehicle inactivity times, total inactivity time, and time interval between two inactivity times within a specific period; define charging habit features and vehicle inactivity features as usage dimension feature parameters;

[0046] Step 5: Match the vehicle dimension, state dimension, and usage dimension feature parameters according to the accumulated mileage, and add corresponding retired and non-retired state labels; filter the feature parameters of each dimension based on the state labels and through correlation analysis; and construct a training set using the retained feature parameters;

[0047] Step 6: Establish a vehicle retirement prediction model based on the machine learning algorithm, use the constructed training set for training, and verify the training effect;

[0048] Step 7: Apply the trained model online to the prediction of electric vehicle retirement, so that it can predict and output the corresponding vehicle retirement status based on the feature parameters of each dimension collected in real time.

[0049] In a preferred embodiment of the present invention, for 23,741 retired vehicle static data sets and 720 retired electric vehicles three years before retirement real vehicle operation data sets. The collected static data includes at least 11 data items such as vehicle scrapping date, production date, vehicle type, battery type, battery rated parameters, etc. The vehicle uses include passenger cars, special vehicles for freight or special transportation, and buses; the real vehicle operation data includes at least 20 data items such as vehicle speed, mileage, total voltage, total current, etc.

[0050] Due to the errors in sensor collection and network transmission errors, the original operating data obtained contains abnormal values, duplicate values ​​and missing values. In order to avoid the interference of abnormal data on subsequent research, abnormal data should be identified and processed. For abnormal values, according to the nominal range of each data item, the data items that exceed the reasonable data range are set to null values. For time-repeated data, the differences in data items such as voltage, current and temperature between the repeated data are determined. If the difference is too large, all are deleted. If the difference is small, the average value is taken. Finally, fill in the missing values, and select linear interpolation or nearest neighbor interpolation according to the data type. For example, linear interpolation is used for continuous values ​​such as voltage and mileage, and nearest neighbor interpolation is used for discrete values ​​such as vehicle status.

[0051] In step 2, the service life of each vehicle is calculated using scrapping date and production date data; the battery rated energy is obtained by multiplying the rated voltage and rated capacity in the battery rated parameters; passenger cars, special-purpose vehicles and buses are distinguished to obtain the following Weibull distribution probability density functions:

[0052]

[0053] Among them, x is the usage time of the vehicle, k is the scale parameter, and λ is the shape parameter;

[0054] Data with a service life of less than 1 year and more than 10 years are removed to reduce the impact of outliers, and the least squares method is fitted for different types of vehicles to determine the corresponding scale parameters and shape parameters. Based on the results of parameter estimation, the vehicle service life probability distribution curve suitable for different vehicle types can be reconstructed. By integrating the curve area to the right of the current use time, the vehicle aging score can be obtained, which means the percentage of a certain type of vehicle that has not been retired when it is used to a certain time, which is used to evaluate the vehicle reliability at different time points.

[0055] The calculation formula of aging score SR is as follows:

[0056]

[0057] In actual evaluation, the vehicle type and accumulated usage time can be obtained first, and then substituted into the aging score calculation formula of the corresponding vehicle type to obtain the aging score result.

[0058] In step 3, the original data is segmented into charging data segments and driving data segments based on the vehicle status, charging status, current, speed and other data. The process is as follows: Figure 2 As shown in the figure, there are still missing data after segmentation. Generally, the time interval between two frames of data is 10s. For data with a time interval of 20-60s, linear interpolation is used for data interpolation. The interpolated data is all 10s per frame. At the same time, the length of continuous data affects the calculation accuracy of battery capacity and driving range. Therefore, the segments with a total time length greater than 100s are selected and retained, and the segments with SOC changes less than 5 are removed.

[0059] Since the charging adopts a multi-stage constant current charging strategy, the charging data has smaller current fluctuations and more stable conditions compared to the driving data, so it is more reliable to use the charging segment data to calculate the battery capacity in step 3. For each charging segment data after segmentation, considering that the corresponding SOC data at the segment boundary may be incomplete, which affects the calculation result of the battery capacity, the left and right boundary SOC data are removed, and the middle part of the data is used to calculate the battery charging capacity. For example, if the charging segment SOC range after segmentation is 29-71, select the 30-70 part of the data and use the following formula to calculate the battery capacity:

[0060]

[0061] Where I is the charging current, t0 is the starting time of the charging segment, t1 is the ending time of the charging segment, SOC0 and SOC1 are the SOC values ​​at the starting and ending times of the charging segment respectively;

[0062] The process of extracting the full-charge range from the driving data snippet includes:

[0063] First, the distance that a fully charged vehicle can travel continuously under the current driving conditions is defined as the equivalent full-charge range R. d , using the accumulated mileage and SOC in the driving data segment to calculate:

[0064]

[0065] Among them, R0 and R1 are the cumulative mileage at the beginning and end of the driving segment, respectively, and SOC0 and SOC1 are the SOC values ​​at the beginning and end of the driving segment, respectively.

[0066] The above formula indicates that a driving segment is used as a specific working condition to calculate the mileage that can be driven when the battery is fully charged, so the length of the driving segment has a greater impact on the calculation result of the driving range. The driving segment with a larger SOC change contains more driving information, and the calculated equivalent driving range is relatively more accurate, so the calculation results of the driving segment with an SOC change greater than 20 are selected and retained.

[0067] In step 4, the extraction of vehicle usage dimension features is considered. Starting from the actual retired vehicle operation data, the usage habits of different users are analyzed to understand the time-varying characteristics of user travel needs and the differences in travel needs of vehicles with different purposes, and finally the travel needs are quantified and extracted. Vehicle travel demand mainly reflects the mileage needs of car owners when using the vehicle, which is usually related to the actual environment and work needs. For example, the travel needs of private cars are related to the distance from home to the company, and the travel needs of logistics vehicles are related to the distance between two warehouses. These all indicate that the differences in the usage habits and usage needs of different vehicles should be fully considered before specific problems can be analyzed for vehicle retirement. In the actual analysis process, the quantification of mileage demand is divided into three dimensions: single-day travel, single trip and trip between two charges. The difference is that the driver does not charge and replenish energy in single trips and trips between two charges. The trip between two charges may include multiple single trips, while single-day travel generally has charging and replenishment. In terms of mileage values, single-day travel>travel between two charges>single trip. Considering that travel demand is affected by seasons and temperature and has time-varying characteristics, the mileage data corresponding to each dimension is counted according to a cycle of every two months. Then, the travel demand is quantified. In order to remove the impact of some occasional extreme trips, quantile ratios of 98%, 95%, and 85% are set for single-day travel, single-trip travel, and travel between two charges, respectively, and the corresponding travel mileage that meets the quantile ratio in each case is counted. At the same time, the travel demand for single-day, single-trip, and between two charges under three satisfaction ratios of the full data without dividing the time period is extracted to expand the perspective of looking at the data. Setting different satisfaction ratios can ultimately determine the best indicator for retirement status prediction through subsequent feature selection, which can more comprehensively examine the data characteristics, help discover potential important information, and is also convenient for application to different types of data.

[0068] In addition to travel demand, usage dimension features also include user charging habits and vehicle deactivation characteristics, thereby comprehensively characterizing the use of the vehicle. Charging habits include charging amount and safety SOC. The charging amount is based on the charging data and the daily charging amount is counted every two months. The daily charging amount corresponding to the 80% quantile is selected as the feature. Safety SOC refers to the amount of power that will be retained at the end of each trip. Too low a power level will cause the voltage to drop, and the performance of the lithium battery cannot be controlled, which will also deepen mileage anxiety. Therefore, the generally usable driving range is the corresponding range after the full charge minus the safety SOC part. It can be compared with the single travel demand and the travel demand between two charges. For single-day demand, it is compared with the driving range corresponding to the full charge minus the safety SOC plus the daily charging amount.

[0069] The vehicle disuse feature is identified by analyzing the mileage difference and time difference between different driving segments. When the cumulative mileage difference between adjacent driving segments is small but the time difference is greater than 30 days, it is judged that the vehicle has been disused. The number of disuses of each vehicle, the total disuse duration and the time interval between two disuses are counted as selected features.

[0070] After the feature parameters of the three dimensions are extracted and step five is entered, the nearest neighbor of the cumulative mileage is matched and synthesized into a feature set corresponding to different cumulative mileage data, and then the vehicle retirement status is calibrated. There is a certain delay in the retirement of electric vehicles due to non-fault reasons. There is a certain time difference between the owner's dissatisfaction with the vehicle's performance and status and the final decommissioning. Therefore, the retirement status can be calibrated by the last small part of the mileage dimension and the time dimension. For example, the operating data corresponding to the expected cumulative mileage of each retired vehicle greater than 95% of the final mileage and the use time of the last two months can be added with the retired status label 1; for other data, the non-retired status label 0 is added.

[0071] In the correlation analysis of feature parameters, feature screening is completed by calculating the following Pearson correlation coefficient:

[0072]

[0073] Among them, PCC is the characteristic variable x i and retired status y i The Pearson correlation coefficient between and They are x i and i The average value of . According to the calculation results of the correlation coefficient, the feature set with high correlation with the retired status label and low autocorrelation between different features is retained. Through feature screening, a total of 17 features in three dimensions are retained, as shown in Table 1.

[0074] Table 1 The 17 features retained after screening and their descriptions

[0075]

[0076]

[0077] In step six, any machine learning algorithm from among random forest, K nearest neighbor, LightGBM, XGBoost, CatBoost, and TabNet can be used to establish and train a vehicle retirement prediction model. Alternatively, two or more models can be trained separately, while grid search is used to optimize hyperparameters, and the optimal model is determined through cross-validation. To verify the accuracy and generalization of the model, 10% of vehicles of each type can be randomly reserved as test data, which are not input into model training, and the remaining data can be used to train the model using the ten-fold cross-validation method.

[0078] When verifying the training effect, the following formulas are used to calculate the precision, recall, and F1-Score:

[0079]

[0080] Among them, TP means that the predicted category is positive and the actual category is also positive, FP means that the predicted category is positive but the actual category is negative, TN means that the predicted category is negative and the actual category is also negative, and FN means that the predicted category is negative but the actual category is positive.

[0081] At the same time, the area under the curve (AUC) is used as the evaluation indicator of the model.

[0082] In this example, TabNet was finally confirmed as the optimal machine learning model, with an F1 score of 0.836 and an AUC index of 0.977 on the reserved test set, proving that the accuracy and generalization of its model prediction have reached a high level.

[0083] 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.

[0084] 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 method for predicting the retirement of electric vehicles based on usage demand, characterized in that: The specific steps include: Step 1: Collect static data related to the vehicles and batteries of retired electric vehicles of different uses and types, as well as actual vehicle operation data within a certain period before retirement; perform preprocessing on the collected data, including setting abnormal values ​​to null values, deleting or averaging duplicate values, and filling missing values, and establish corresponding retired vehicle static data sets and actual vehicle operation data sets; Step 2: obtain the service life and battery rated energy of each vehicle by processing static data, and distinguish different vehicle types to fit the corresponding vehicle service life Weibull distribution probability density function respectively; calculate the aging scores of different types of vehicles using the probability density function; define the service life and aging score as vehicle dimension characteristic parameters; Step 3: Process the actual vehicle operation data to obtain the segmented charging data segments and driving data segments; extract charging state characteristic parameters including battery capacity, single cell voltage, charging time, cumulative driving mileage, average temperature and average charging current from the charging data segments; Extract driving state characteristic parameters including full-charge driving range, cumulative driving range, average current of driving section, average speed and energy consumption rate from driving data segments; define charging state characteristic parameters and driving state characteristic parameters together as state dimension characteristic parameters; Step 4: Set corresponding quantile ratios for single-day travel, single-trip travel, and travel between two charges in driving behavior, and count the corresponding travel mileages that meet the quantile ratios in different situations; extract charging habit features including daily charging amount and safe SOC retained after each trip for charging behavior, and extract vehicle inactivity features including the number of vehicle inactivity times, total inactivity time, and time interval between two inactivity times within a specific period; define habit features and vehicle inactivity features as usage dimension feature parameters; Step 5: Match the vehicle dimension, state dimension, and usage dimension feature parameters according to the accumulated mileage, and add corresponding retired and non-retired state labels; filter the feature parameters of each dimension based on the state labels and through correlation analysis; and construct a training set using the retained feature parameters; Step 6: Establish a vehicle retirement prediction model based on the machine learning algorithm, use the constructed training set for training, and verify the training effect; Step 7: Apply the trained model online to the prediction of electric vehicle retirement, so that it can predict and output the corresponding vehicle retirement status based on the real-time data corresponding to the feature parameters of each dimension.

2. The method according to claim 1, characterized in that: The static data collected in step 1 at least include the vehicle scrapping date, production date, vehicle type, battery type, and battery rated parameters. The vehicle use includes passenger private cars, special transport vehicles, and commercial buses; the actual vehicle operation data at least includes vehicle speed, mileage, total voltage, and total current parameters; Data preprocessing includes setting abnormal data items that exceed the reasonable range to null values; deleting all voltage, current and temperature data that are repeated in time when the difference is too large, and taking the average value when the difference is small; and filling missing values ​​with linear interpolation or nearest neighbor interpolation according to the data type.

3. The method according to claim 1, characterized in that: In step 2, the service life of each vehicle is calculated using scrapping date and production date data; the battery rated energy is obtained by multiplying the rated voltage and rated capacity in the battery rated parameters; and the Weibull distribution probability density function of each type of vehicle is obtained in the following form: Among them, x is the usage time of the vehicle, k is the scale parameter, and λ is the shape parameter; The data with service life less than 1 year and more than 10 years are removed to reduce the influence of outliers, and the least squares method is used to fit different types of vehicles to determine the corresponding scale parameters and shape parameters; The aging scores SR of different types of vehicles are obtained:

4. The method according to claim 1, characterized in that: In step three, the original data is segmented into charging data segments and driving data segments based on the vehicle status, charging status, current, and vehicle speed data. In the case where there is still missing data after segmentation, a new data frame is formed by linearly interpolating multiple frames of data. Data segments whose total duration is greater than a specific value or whose SOC change is less than a specific value are eliminated.

5. The method according to claim 1, characterized in that: The specific process of extracting the battery capacity from the charging data fragment in step 3 includes: First, remove the SOC data at the left and right boundaries of the charging data segment, and then use the partial SOC range to calculate the battery capacity using the following formula: Where I is the charging current, t0 is the starting time of the charging segment, t1 is the ending time of the charging segment, SOC0 and SOC1 are the SOC values ​​at the starting and ending times of the charging segment respectively; The process of extracting the full-charge range from the driving data snippet includes: First, the distance that a fully charged vehicle can travel continuously under the current driving conditions is defined as the equivalent full-charge range R. d , using the accumulated mileage and SOC in the driving data segment to calculate: Among them, R0 and R1 are the cumulative mileage at the beginning and end of the driving segment, respectively, and SOC0 and SOC1 are the SOC values ​​at the beginning and end of the driving segment, respectively; The calculation results corresponding to the driving segments with SOC changes greater than 20 are filtered out.

6. The method according to claim 1, characterized in that: In step 4, quantile ratios of 98%, 95%, and 85% are set for single-day travel, single trip, and trip between two charges, respectively, and the corresponding travel mileage that meets the quantile ratio in each case is counted; The charging volume is specifically based on the daily charging volume counted for every two months of charging data, and the daily charging volume corresponding to the 80% quantile is selected as the corresponding usage dimension feature; The vehicle deactivation feature is identified and extracted based on the cumulative mileage and time difference between adjacent driving segments. When the cumulative mileage difference between adjacent driving segments is small but the time difference is greater than 30 days, it is judged that the vehicle has been deactivated. The number of deactivations of each vehicle, the total deactivation duration and the time interval between two deactivations are counted as the corresponding usage dimension features.

7. The method according to claim 1, characterized in that: In step 5, the retired vehicle's expected cumulative mileage is greater than 95% of the final mileage and the running data corresponding to the last two months of use are added with a retired status label of 1; the other data are added with a non-retired status label of 0; In the correlation analysis of feature parameters, feature screening is completed by calculating the following Pearson correlation coefficient: Among them, PCC is the characteristic variable x i and retired status y i The Pearson correlation coefficient between and They are x i and i The average value of .

8. The method according to claim 1, characterized in that: In step six, any one of the machine learning algorithms selected from random forest, K nearest neighbor, LightGBM, XGBoost, CatBoost, and TabNet is used to establish and train a vehicle retirement prediction model; or two or more models are used for training respectively, while grid search is used for hyperparameter optimization, and the optimal model is determined through cross-validation; when verifying the training effect, precision, recall, F1 value, and area under the curve AUC are used as evaluation indicators of the model.

Images