A method and system for predicting a current cycle count of a power battery using machine learning

Abstract

This invention discloses a method and system for predicting the current cycle number of a power battery using machine learning. The method includes: acquiring historical data of the target power battery to establish a dataset for training a model; tuning the model using a chaotic sparrow algorithm; training the model using the CatBoost algorithm based on the hyperparameters obtained from the tuning; this model can be used to predict the current cycle number of batteries of the same model as the target power battery with high accuracy. This invention can effectively reduce a series of failures caused by aging of power batteries, and compared to predicting battery life by combining experiments and data, pure data-driven prediction is more convenient and can be widely applied to predicting the lifespan of power batteries of various types and models.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power batteries, and particularly relates to a method and system for predicting the current cycle number of a power battery by using machine learning. BACKGROUND

[0002] The safety of a power battery is generally divided into on-site safety and misuse safety. In long-term use, the discharge capacity of a battery gradually decreases. The main reasons are corrosion of the materials of the two poles, changes in the electrolyte, aging of the separator, etc. These phenomena will cause the degradation of the performance of the battery, which is specifically manifested as a decrease in the discharge capacity. In the IEEE standard, if the SOH (the ratio of the maximum discharge capacity of a battery to its rated capacity) of the battery is lower than 80%, the battery is defined as a faulty battery. In order to avoid safety hazards caused by aging of the battery during use, the current cycle number or the remaining cycle number of the battery needs to be predicted. SUMMARY

[0003] To solve the above technical problems, the purpose of the present application is to provide a method for predicting the current cycle number of a power battery by using machine learning. The method can use machine learning to train the data in the charging and discharging process of the power battery to obtain a model capable of predicting the current cycle number of the power battery, and can have a high prediction accuracy. The method is suitable for various types of power batteries, and can improve the misuse safety of the power battery.

[0004] To achieve the purpose of the present application, the present application provides a method for predicting the current cycle number of a power battery by using machine learning, which comprises:

[0005] Obtaining the historical data of a target power battery, evaluating the importance of the parameters in the historical data to the current cycle number by using a random forest, and establishing a data set for model training;

[0006] Using a chaotic sparrow algorithm to perform parameter tuning to obtain the hyperparameters required by a relatively optimal algorithm;

[0007] Training the training set data by using a CatBoost algorithm to obtain a current cycle number prediction model capable of predicting the current cycle number of a battery of the same type and number as the target power battery;

[0008] Inputting a plurality of sets of measured data of power batteries into the current cycle number prediction model obtained by training, outputting a plurality of predicted values of the current cycle number, analyzing the plurality of predicted values of the current cycle number, and obtaining the value of the current cycle number of the target power battery after eliminating noise data.

[0009] Further, the specific steps of establishing the data set for model training comprise:

[0010] Obtaining the historical data of a target power battery to obtain a historical data set;

[0011] data cleaning is performed on the historical data in the historical data set;

[0012] the importance of the parameters in the historical data to the current cycle number is evaluated using a random forest, and the parameter data is determined according to the importance;

[0013] the number of selected parameters is determined, and an optimal data set for model training is established.

[0014] Further, the evaluation of the importance of the parameters in the historical data to the current cycle number using a random forest and the determination of the parameter data according to the importance include:

[0015] the importance of the parameters to the current cycle number is evaluated using a random forest;

[0016] if n parameters are selected, a direct calculation is performed using a random forest;

[0017] if n-1 parameters are selected:

[0018] n parameters are excluded, and n groups of n-1 parameters are obtained by excluding the parameters each time;

[0019] the n groups of parameter data are calculated using a random forest to obtain the importance of each group of parameters to the current cycle number, and are sorted in the order of 1, 2,..., n-1, with 1 being the most important parameter;

[0020] the serial numbers of each parameter in each group are added to obtain a sequence of n parameters in total, which is sorted from low to high with the parameter name as the representative;

[0021] The parameter with the lowest sequence in the sequence has the highest impact on the current cycle number, and the parameter with the highest sequence has the lowest impact on the current cycle number;

[0022] The parameter with the lowest importance is screened out, and the remaining n-1 parameters are the best parameters under the condition of selecting n-1 parameters. These parameter data are used to establish a data set for model training.

[0023] Further, the determination of the number of selected parameters and the establishment of an optimal data set for model training include:

[0024] a set of hyperparameter values is generated each time the chaotic sparrow algorithm is iterated;

[0025] CatBoost uses the set of hyperparameter values generated in the current iteration each time the iteration is performed;

[0026] The CatBoost algorithm trains the data in the training set and validates the data in the validation set, and outputs the error value of the model prediction for the training set and the validation set each time the training iteration is performed;

[0027] The lowest validation set prediction error of the set number of training iterations is fed back to the chaotic sparrow algorithm;

[0028] The chaotic sparrow algorithm gives the next set of hyperparameter values according to the feedback result;

[0029] After the set number of iterations, the chaotic sparrow algorithm gives the optimal generation of hyperparameter results.

[0030] Further, the hyperparameters required by the relatively optimal algorithm obtained by using the chaotic sparrow algorithm for parameter tuning include:

[0031] Set the number of iterations iter;

[0032] Each iteration gives a set of CatBoost algorithm hyperparameters A i ;

[0033] The CatBoost algorithm uses hyperparameters A1 to train the training set data, avoids overfitting, and after a certain number of training, obtains the relative minimum error value of the test set, denoted as error value B1;

[0034] The chaotic sparrow algorithm gives the next set of hyperparameters A2 according to the error value B1;

[0035] In turn, until the number of iterations iter is reached, stop the loop and obtain a set of error values B1, B2,..., B iter ;

[0036] Select the minimum error value B i in B1, B2,..., B iter ;

[0037] According to the minimum error value B i , a set of relatively optimal CatBoost hyperparameters is obtained.

[0038] Further, the training set data is trained using the CatBoost algorithm to obtain a current cycle number prediction model that can predict the same type of battery as the target power battery, including:

[0039] Set the number of training iterations;

[0040] The CatBoost algorithm trains the training set data and outputs the error of the model prediction of the training set and test set data at each iteration, and uses the mean absolute error MAE to monitor the error;

[0041] Select the model with the lowest error at a given number of iterations as the model for predicting the target power battery.

[0042] Further, the measured data of multiple groups of power batteries are input into the trained current cycle number prediction model to output multiple predicted values of the current cycle number, the multiple predicted values of the current cycle number are analyzed, and a target power battery current cycle number value is obtained after noise data is eliminated, comprising:

[0043] The output of the current cycle number prediction model is the current cycle number of the power battery, and the input parameters of the current cycle number prediction model are determined;

[0044] The target power battery is subjected to m times of charge-discharge cycles, and the specific values of each input parameter in each charge-discharge cycle are recorded;

[0045] The specific values of the m groups of parameters are input into the current cycle number prediction model for prediction to obtain m predicted values of the current cycle number of the power battery;

[0046] The m predicted values are sorted from large to small;

[0047] A confidence space of the predicted values is set, the predicted values not in the confidence space are deleted, the number of the deleted predicted values is defined as h, and h is greater than or equal to 0;

[0048] After the predicted values not in the confidence space are deleted, if the remaining predicted values are not greatly different, the average value or the median of the predicted values is taken as the current cycle number of the target power battery; if the predicted values are greatly different, c times of charge-discharge cycles are supplemented, and the current cycle prediction value of the target power battery in each new supplemented charge-discharge cycle is observed, if the prediction value is not in the original confidence interval, the current cycle prediction value is directly discarded, assuming that a prediction value is discarded, after m-h+(c-a) current cycle prediction values of the target power battery are obtained, it is found that the kurtosis of the m-h+(c-a) data is less than a preset value, and the average value or the median of the values is taken as the current cycle number of the target power battery.

[0049] Further, the kurtosis of the remaining predicted values after the predicted values not in the confidence space are removed is calculated, if the kurtosis is greater than or equal to 3, it is greatly different; if the kurtosis is less than 3, it is not greatly different.

[0050] Further, the method further comprises the step of comparing the obtained current cycle prediction number with the charge-discharge test life of the target power battery to judge the health state of the battery.

[0051] The application also provides a system for predicting the current cycle number of a power battery by machine learning, comprising:

[0052] A data acquisition and evaluation module is used to acquire historical data of a target power battery, evaluate the importance of parameters in the historical data to the current cycle number by using a random forest, and establish a data set for model training;

[0053] The parameter adjusting module is configured to adjust parameters by using the chaotic sparrow algorithm to obtain a set of relatively optimal super parameters required by the algorithm;

[0054] The training module is configured to train the training set data by using the CatBoost algorithm to obtain a current cycle number prediction model capable of predicting the current cycle number of the same type of battery as the target power battery.

[0055] The prediction module is configured to input a plurality of groups of measured data of power batteries into the trained current cycle number prediction model, output a plurality of predicted values of the current cycle number, analyze the plurality of predicted values of the current cycle number, and obtain a current cycle number value of the target power battery after eliminating noise data.

[0056] Compared with the prior art, one or more embodiments of the present application have at least the following beneficial effects:

[0057] The present application comprehensively uses random forest, chaotic sparrow algorithm and CatBoost algorithm, uses random forest to evaluate the importance of historical data parameters, provides an optimal data set for training of the CatBoost algorithm, uses the chaotic sparrow algorithm to provide a set of relatively optimal super parameters for the CatBoost algorithm, to a certain extent, solves the problem of too much experience and tediousness in algorithm parameter adjustment, uses CatBoost to train a relatively accurate prediction model, and adopts multiple prediction analysis results when predicting the target power battery to obtain a relatively accurate current cycle number value of the target power battery. BRIEF DESCRIPTION OF DRAWINGS

[0058] Figure 1 is a flowchart of a method for predicting the current cycle number of a power battery by using machine learning according to an embodiment of the present application;

[0059] Figure 2 is a data processing and selection flowchart according to an embodiment of the present application;

[0060] Figure 3 is a flowchart of finding CatBoost super parameters by using the chaotic sparrow algorithm according to an embodiment of the present application;

[0061] Figure 4 is a flowchart of obtaining an optimal model by training data of CatBoost according to an embodiment of the present application;

[0062] Figure 5 is a flowchart of analyzing the results after model prediction according to an embodiment of the present application. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below with reference to the embodiments and drawings.

[0064] Please refer toFigure 1 The application provides a method for predicting the current cycle number of a power battery by using machine learning, comprising the following steps:

[0065] Step 1: Obtain the historical data of the target power battery, evaluate the importance of the parameters in the historical data to the current cycle number by using a random forest, and establish a data set required for an optimal training model.

[0066] This step processes the historical data of the target power battery, so that the data can better reflect the internal relationship between different parameters and the current cycle number.

[0067] In some embodiments of the application, the data processing comprises deleting abnormal data such as zero values and experience deviations in the historical data.

[0068] This step comprises the following sub-steps:

[0069] Step S11: Obtain the historical data of the target power battery, and obtain a historical data set;

[0070] In some embodiments of the application, the historical data comprises information such as power, internal resistance and temperature after charging and discharging cycles in the charging and discharging process of the target power battery. The specific parameters and the types of data that can be obtained are different, and in other embodiments, other historical data recorded in the charging and discharging process related to the current cycle number of the battery can also be obtained.

[0071] Step S12: Perform data cleaning on the historical data in the historical data set;

[0072] In some embodiments of the application, there are n parameters in the historical data set, which are denoted as X1, X2,..., Xn respectively. n n represents the data of n target power batteries in the charging and discharging process and the current cycle number, and the abnormal values in the historical data set are removed. If an abnormal value appears in a parameter, the corresponding group of data at the current cycle number is deleted.

[0073] Step S13: Evaluate the importance of the parameters in the historical data to the current cycle number by using a random forest;

[0074] In some embodiments of the application, the specific steps of importance evaluation comprise:

[0075] Evaluate the importance of the parameters to the current cycle number by using a random forest;

[0076] If n parameters are selected, the random forest is directly calculated;

[0077] If n-1 parameters are selected:

[0078] Excluding n times of parameters, n times of excluding parameters are different, and n groups of n-1 parameter combinations are obtained;

[0079] The n groups of parameter data are calculated by using the random forest, the importance of each group of parameters to the current cycle number is obtained, and the importance is ranked in 1, 2,..., n-1, and 1 is the highest importance parameter;

[0080] The sequence numbers of each group of parameters are added to obtain a sequence of n parameters in total, and the sequence is sorted from low to high according to the parameter name;

[0081] The parameter at the lowest position in the sequence has the highest influence on the current cycle number, and the parameter at the highest position has the lowest influence on the current cycle number;

[0082] The parameter with the lowest importance is screened out, and the remaining n-1 parameters are the best parameters under the selection of n-1 parameters, and the data set for training the model is established by using these parameter data;

[0083] If n-2, n-3,..., parameters are selected, the processing method is similar to that of selecting n-1 parameters.

[0084] Step S14: determining the number of selected parameters, and establishing an optimal data set for model training, specifically including:

[0085] Step S141: using the CatBoost algorithm default hyperparameter, respectively inputting the x parameter data set screened by using the random forest, x-1 parameter data set, x-2 parameter data set,..., parameter data set;

[0086] Step S142: dividing the data set under each number of parameters into a training set and a test set according to a preset ratio, in some embodiments of the present application, the training set and the test set are divided according to a ratio of 9:1;

[0087] Step S143: setting a larger training iteration number of the CatBoost algorithm;

[0088] Step S144: training each training set by using the CatBoost algorithm, and recording the lowest error of the test set under each number of parameters after the iteration is ended without overfitting;

[0089] Step S145: If the lowest error of the test set under x parameters < the lowest error of the test set under x - 1 parameters, then select the data set with x parameters as the test set for model training; if the lowest error of the test set under x parameters > the lowest error of the test set under x - 1 parameters, then continue to compare the lowest error of the test set under x - 1 parameters with the lowest error of the test set under x - 2 parameters, and so on, until the lowest error of the test set under x - n parameters < the lowest error of the test set under x - n - 1 parameters, then select the data set with x - n parameters as the test set for model training.

[0090] Step 2: Use the chaotic sparrow algorithm for parameter tuning to obtain a set of relatively optimal hyperparameters required for the algorithm.

[0091] In this invention, the steps of obtaining a set of relatively optimal hyperparameters required for the algorithm specifically include:

[0092] Set the number of iterations iter;

[0093] Each iteration gives a set of hyperparameters A of the CatBoost algorithm i ;

[0094] The CatBoost algorithm uses hyperparameters A1 to train the training set data. In the case of avoiding overfitting, after a certain number of trainings, the relatively lowest error value of the test set is obtained, denoted as error value B1;

[0095] The chaotic sparrow algorithm gives the next set of hyperparameters A2 according to the error value B1;

[0096] Loop sequentially until the number of iterations iter is reached, stop the loop, and obtain a set of error values B1, B2,..., B iter ;

[0097] Select the minimum error B among B1, B2,..., B iter ; i ;

[0098] According to the minimum error B i Obtain a set of relatively optimal CatBoost hyperparameters.

[0099] Step 3: Use the CatBoost algorithm to train the training set data to obtain a current cycle number prediction model that can predict batteries of the same type as the target power battery.

[0100] In this step, such as Figure 4 , to obtain the optimal model using CatBoost to train the data, it is necessary to find the optimal number of iterations, specifically including:

[0101] Step 31: Set a large number of training iterations so that the error on the test set does not decrease significantly compared to the error on the training set, or even increases.

[0102] Step 32: Observe the error between the training set and the test set during the iteration process. If the test set error does not decrease with the number of iterations and the same value appears in multiple iterations, select the iteration number where the value first appears. If the test set error increases with the number of iterations, select the iteration number of the generation with the lowest test set error. If both the training set error and the test set error decrease, and the rate of decrease of the training set error gradually exceeds the rate of decrease of the test set error, and after multiple iterations, the training set error is significantly lower than the test set error, select the iteration number where the training set error and the test set error are close.

[0103] Step 33: Select the model that has been retrained after the relatively optimal number of iterations and save it to obtain the final prediction model for the current number of iterations.

[0104] Step 4: Input multiple sets of measured power battery data into the current cycle count prediction model, output multiple predicted values ​​for the current cycle count, analyze these predicted values, and after eliminating noise data, obtain a relatively accurate target current cycle count value for the power battery. The data type of the measured data is consistent with the input data type used during training.

[0105] In some embodiments of the present invention, the battery data source for the training data to obtain the model is the same as the type and model of the target power battery.

[0106] Step 4 includes the following sub-steps:

[0107] Step S41: Determine the input parameters of the current cycle count prediction model. The output of the current cycle count prediction model is the current cycle count of the power battery.

[0108] Step S42: Perform m charge-discharge cycles on the target power battery and record the specific values ​​of each input parameter in each charge-discharge cycle;

[0109] In some embodiments of the present invention, the input parameters of the model are power, internal resistance, and temperature, and the specific values ​​of the three parameters of power, internal resistance, and temperature are recorded for each charge-discharge cycle.

[0110] Step S43: Input the specific values ​​of m sets of parameters into the current cycle number prediction model for prediction, and obtain the predicted values ​​of the current cycle number of m power batteries;

[0111] In some embodiments of the present invention, the specific values ​​of m sets of three parameters (power, internal resistance, and temperature) are input into the current cycle number prediction model for prediction.

[0112] Step S44: Sort the m predicted values ​​from largest to smallest;

[0113] Step S45: Set the confidence space of the predicted values, delete the predicted values ​​that are not in the confidence space, and define the number of predicted values ​​to be deleted as h, h≥0;

[0114] In some embodiments of the present invention, a 95% confidence level for the mean is set, and a confidence interval at a 95% confidence level for the mean is obtained.

[0115] Step S46: After deleting predicted values ​​that are not within the confidence space, if the remaining predicted values ​​are not significantly different, take the average or median of the predicted values ​​as the current cycle number of the target power battery; if the predicted values ​​differ significantly, perform c additional charge-discharge cycles and observe the current cycle predicted value of the target power battery under each newly added charge-discharge cycle. If the predicted value is not in the original confidence interval, discard the current cycle predicted value directly. Assuming that a predicted values ​​are discarded, and m-h+(ca) current cycle predicted values ​​of the target power battery are obtained, and the kurtosis of the m-h+(ca) data is found to be less than the preset value, take the average or median of the values ​​as the current cycle number of the target power battery.

[0116] The distinction between "not much difference" and "large difference" is made after removing items that are not in this confidence interval from the previous step.

[0117] In some embodiments of the present invention, the kurtosis of the remaining predicted values ​​after removing the predicted values ​​that are not in the confidence space is calculated. That is, if h predicted values ​​that are not in the confidence space are removed, the kurtosis of mh predicted values ​​is calculated here. If the kurtosis is greater than or equal to 3, the difference is large; if the kurtosis is less than 3, the difference is small.

[0118] Step 5: Compare the current cycle prediction number with the charge-discharge test life of the target power battery (usually available in the instruction manual) to determine the health status of the battery.

[0119] In some embodiments of the present invention, it is predicted that the battery has been charged and discharged 1,000 times. If the charge and discharge test (or the instruction manual) states 2,000 times, there is a greater risk of hidden dangers or malfunctions. In other words, the health status of the battery can be judged by comparing the two values.

[0120] The method provided in this embodiment of the invention uses the optimal set of hyperparameters found by the Chaotic Sparrow Algorithm to train the optimal dataset by proportionally dividing it into a training set and a test set. The number of training iterations is selected based on the error between the training set and the test set in each iteration. After setting the number of training iterations again, the training data is trained and a prediction model is obtained. The prediction model can be used to obtain the predicted value of the current cycle number. The health status of the battery and its safety can be evaluated by using the predicted value.

[0121] The present invention also provides a system for implementing the methods provided in the foregoing embodiments.

[0122] Systems that use machine learning to predict the current cycle number of a power battery include:

[0123] The data acquisition and evaluation module is used to acquire historical data of the target power battery, use random forest to evaluate the importance of parameters in the historical data to the current cycle number, and establish a dataset for model training.

[0124] The parameter tuning module is used to perform parameter tuning using the Chaotic Sparrow Algorithm to obtain a set of relatively optimal hyperparameters required for the algorithm.

[0125] The training module is used to train the training set data using the CatBoost algorithm to obtain a prediction model that can predict the current cycle count of batteries of the same type and model as the target power battery.

[0126] The prediction module is used to input multiple sets of measured power battery data into the trained current cycle count prediction model, output multiple predicted values ​​of the current cycle count, analyze the multiple predicted values ​​of the current cycle count, and obtain the target power battery's current cycle count value after eliminating noise data.

[0127] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. Regarding the system disclosed in the embodiments for predicting the current cycle number of a power battery using machine learning, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be found in the method section.

[0128] While the embodiments disclosed in this invention are as described above, the content is merely for the purpose of facilitating understanding of the invention and is not intended to limit the invention. Any person skilled in the art to which this invention pertains may make any modifications and variations in form and detail of the implementation without departing from the spirit and scope disclosed herein; however, the scope of patent protection for this invention shall still be determined by the scope defined in the appended claims.

Claims

1. A method for predicting the current cycle number of a power battery using machine learning, characterized in that, The method includes the following steps: Historical data of the target power battery is obtained, and random forest is used to evaluate the importance of parameters in the historical data to the current cycle number, and a dataset for model training is established. The hyperparameters required for the algorithm were obtained by using the Chaotic Sparrow Algorithm for parameter tuning; The CatBoost algorithm is used to train the training set data to obtain a prediction model that can predict the current cycle count of batteries of the same type and model as the target power battery. Multiple sets of measured power battery data are input into the trained current cycle count prediction model, which outputs multiple predicted values ​​of the current cycle count. After analyzing the multiple predicted values ​​of the current cycle count and eliminating noise data, the current cycle count value of the target power battery is obtained. The process involves inputting multiple sets of measured power battery data into a trained current cycle count prediction model, outputting multiple predicted current cycle counts, analyzing these predicted values, and eliminating noise data to obtain the target power battery's current cycle count value. This includes: The output of the current cycle count prediction model is the current cycle count of the power battery, and the input parameters of the current cycle count prediction model are determined. Perform m charge-discharge cycles on the target power battery and record the specific values ​​of each input parameter in each charge-discharge cycle. The specific values ​​of m sets of parameters are input into the current cycle number prediction model for prediction, and the predicted values ​​of the current cycle number of m power batteries are obtained. Sort the m predicted values ​​from largest to smallest; Define a confidence space for the predicted values, delete predicted values ​​that are not in the confidence space, and define the number of predicted values ​​to be deleted as h, where h≥0; After deleting predicted values ​​that are not in the confidence space, if the remaining predicted values ​​are not significantly different, the average or median of the predicted values ​​is taken as the current cycle number of the target power battery. If the predicted values ​​are significantly different, c additional charge-discharge cycles are performed, and the current cycle predicted value of the target power battery under each newly added charge-discharge cycle is observed. If the predicted value is not in the original confidence interval, the current cycle predicted value is directly discarded. Assuming a predicted values ​​are discarded, after obtaining m-h+(ca) current cycle predicted values ​​of the target power battery, it is found that the kurtosis of the m-h+(ca) data is less than the preset value, and the average or median of the values ​​is taken as the current cycle number of the target power battery. The method for judging whether the predicted values ​​are not significantly different or significantly different is to calculate the kurtosis of the remaining predicted values ​​after removing the predicted values ​​that are not in the confidence space. If the kurtosis is greater than or equal to the preset kurtosis value, it is considered significantly different; if the kurtosis is less than the preset kurtosis value, it is considered not significantly different.

2. The method for predicting the current cycle number of a power battery using machine learning according to claim 1, characterized in that, The specific steps for establishing the dataset for model training include: Obtain historical data of the target power battery to obtain a historical dataset; Perform data cleaning on historical data in the historical dataset; Random forests are used to evaluate the importance of parameters in historical data to the current iteration number, and parameter data is determined based on importance. Determine the number of parameters to use and establish the optimal dataset for model training.

3. The method for predicting the current cycle number of a power battery using machine learning according to claim 2, characterized in that, The method of using random forests to evaluate the importance of parameters in historical data to the current iteration number, and determining parameter data based on importance, includes: Random forests are used to evaluate the importance of parameters to the current iteration number. If n parameters are selected, random forest can be used for direct calculation. If n-1 parameters are selected: By eliminating parameters n times, and each of the n eliminations is different, we obtain n sets of n-1 combinations of parameters; Random forest is used to calculate the importance of each parameter to the current iteration number using n sets of parameter data. The parameters are then sorted into 1, 2, ..., n-1, with 1 being the parameter with the highest importance. Add the sequence numbers of each parameter in each group to get a sequence of n parameters in total, and sort them from low to high according to the parameter name; The parameter with the lowest ranking in the sequence has the greatest impact on the current loop count, while the parameter with the highest ranking has the least impact on the current loop count. After removing the parameters with the lowest importance, the remaining n-1 parameters are the best parameters among the n-1 parameters. The dataset for training the model is built using the data of these parameters.

4. The method for predicting the current cycle number of a power battery using machine learning according to claim 2, characterized in that, The process of determining the number of parameters to be selected and establishing the optimal dataset for model training includes: The Chaotic Sparrow Algorithm generates a set of hyperparameter values ​​in each iteration; CatBoost uses the set of hyperparameter values ​​generated for the current iteration in each iteration. The CatBoost algorithm is trained on the training set and validated on the validation set. Each training iteration outputs the error value of the model's predictions on the training and validation sets; The prediction error of the validation set with the lowest number of training iterations is fed back to the Chaotic Sparrow Algorithm. The Chaotic Sparrow Algorithm provides the next set of hyperparameter values ​​based on the feedback results; After the set number of iterations, the algorithm stops and gives the optimal first-generation hyperparameter result.

5. The method for predicting the current cycle number of a power battery using machine learning according to claim 1, characterized in that, The process of using the chaotic sparrow algorithm to tune parameters yields a set of relatively optimal hyperparameters required for the algorithm, including: Set the iteration number iter; Each iteration provides a set of hyperparameters A for the CatBoost algorithm. i ; The CatBoost algorithm uses hyperparameter A1 to train on the training set data. In order to avoid overfitting, it obtains the lowest relative error value on the test set after a certain number of training iterations, which is denoted as error value B1. The Chaotic Sparrow Algorithm provides the next set of hyperparameters A2 based on the error value B1; This process is repeated until the iteration number iter is reached, at which point the loop stops, and a set of error values ​​B1, B2, ..., B are obtained. iter ; Select B1, B2, ..., B iter Minimum error B in i ; Based on the minimum error value B i We obtained a set of relatively optimal CatBoost hyperparameters.

6. The method for predicting the current cycle number of a power battery using machine learning according to claim 1, characterized in that, The method of training the training set data using the CatBoost algorithm to obtain a prediction model capable of predicting the current cycle count of batteries of the same type and model as the target power battery includes: Set the number of training iterations; The CatBoost algorithm trains on the training set data and outputs the error of the model's prediction on the training set and test set data in each iteration, using the mean absolute error (MAE) to monitor the error. The model with the lowest error under a given number of iterations is selected as the model for predicting the target power battery.

7. The method for predicting the current cycle number of a power battery using machine learning according to claim 1, characterized in that, In the method of judging whether the predicted values ​​are not significantly different or significantly different, if the kurtosis is greater than or equal to 3, then the difference is significant; if the kurtosis is less than 3, then the difference is not significant.

8. The method for predicting the current cycle number of a power battery using machine learning according to claim 1, characterized in that, The method further includes the step of comparing the obtained current cycle prediction number with the charge-discharge test life of the target power battery to determine the health status of the battery.

9. A system for predicting the current cycle number of a power battery using machine learning, characterized in that, The system for implementing the method of any one of claims 1-8 comprises: The data acquisition and evaluation module is used to acquire historical data of the target power battery, use random forest to evaluate the importance of parameters in the historical data to the current cycle number, and establish a dataset for model training. The parameter tuning module is used to perform parameter tuning using the Chaotic Sparrow Algorithm to obtain a set of relatively optimal hyperparameters required for the algorithm. The training module is used to train the training set data using the CatBoost algorithm to obtain a prediction model that can predict the current cycle count of batteries of the same type and model as the target power battery. The prediction module is used to input multiple sets of measured power battery data into the trained current cycle count prediction model, output multiple predicted values ​​of the current cycle count, analyze the multiple predicted values ​​of the current cycle count, and obtain the target power battery's current cycle count value after eliminating noise data.

Images