Battery life estimation method, device, equipment, medium and product

Abstract

The invention discloses a battery life estimation method and device, equipment, a medium and a product. The method comprises the following steps: acquiring battery data of a to-be-tested battery; the battery data is used for representing a mapping relation between the number of cycles of the to-be-tested battery and the health state of the to-be-tested battery; determining a target recession model suitable for the battery data in the at least two recession models through the target classification model; the target recession model is a recession model corresponding to the fitting curve type of the mapping relation represented by the battery data; and estimating the residual life of the to-be-tested battery based on the target recession model and the battery data. According to the scheme, the accuracy of battery life estimation is improved.

Description

Technical Field

[0001] This application relates to the field of data processing, and includes, but is not limited to, battery life estimation methods, apparatus, equipment, media, and products. Background Technology

[0002] In the context of the new energy strategy, battery life prediction is an important topic.

[0003] For predicting the remaining battery life, relevant technologies generally use regression models for prediction.

[0004] However, when the prediction scenario is complex, regression tasks are prone to overfitting and underfitting. Furthermore, due to the complex electrodynamic systems underlying battery aging, regression models exhibit poor transferability to different battery systems. In addition, regression tasks are highly sensitive to data noise, which can negatively impact both model accuracy and prediction accuracy. Summary of the Invention

[0005] To address the aforementioned issues, this application provides a battery life estimation method, apparatus, device, medium, and product, which improves the accuracy of battery life estimation.

[0006] The technical solution of this application is implemented as follows:

[0007] In a first aspect, this application provides a battery life estimation method, which includes:

[0008] Acquire battery data for the battery under test; the battery data is used to characterize the mapping relationship between the number of cycles of the battery under test and the health status of the battery under test; determine the target degradation model applicable to the battery data from at least two degradation models using a target classification model; the target degradation model is the degradation model corresponding to the type of fitted curve of the mapping relationship characterized by the battery data; estimate the remaining life of the battery under test based on the target degradation model and the battery data.

[0009] First, the target degradation model used for the battery data of the battery under test is determined through a target classification model. Then, the remaining lifespan of the battery under test is predicted using the target degradation model. This allows for the mapping between the target degradation model and the battery data; that is, the type of the fitted curve representing the mapping relationship between the battery cycle count and the battery's health status can be used to determine the corresponding target degradation model. The degradation model corresponding to the fitted curve is then designated as the target degradation model. Because the target degradation model matches the battery data, the accuracy is higher when predicting battery lifespan. Second, since the target classification model can select the degradation mode, it can be applied to various degradation model usage scenarios. Therefore, for battery data corresponding to various types of batteries, the corresponding target degradation model can be determined, exhibiting wide applicability and good transferability. Third, unlike regression models in related technologies, this method combines a target classification model and a degradation model for battery life prediction. It leverages the accuracy of classification models and overcomes the problem of inaccurate degradation model selection in related technologies, simultaneously utilizing the advantages of both and improving the accuracy of lifespan prediction. Finally, since this method differs from the regression models in related technologies, which have poor interpretability compared to classification models, this method improves interpretability.

[0010] In one possible implementation, estimating the remaining lifespan of the battery under test based on a target degradation model and battery data includes: fitting the battery data to the target degradation model to obtain a target degradation curve; determining the health status value at the end of the battery's lifespan; determining the number of cycles at the end of the battery's lifespan based on the target degradation curve and the health status value at the end of the battery's lifespan; and determining the remaining lifespan of the battery based on the number of cycles at the end of the battery's lifespan and the current number of cycles of the battery.

[0011] In this way, the target decay curve is first determined based on the target decay model, and then the remaining lifetime is determined based on the target decay curve, which has the characteristics of being simple and accurate.

[0012] In one possible implementation, acquiring battery data of the battery under test includes: acquiring raw data of the battery under test; the raw data of the battery under test includes a mapping relationship between the number of battery cycles and the health status of the battery under test; cleaning and normalizing the raw data of the battery under test to obtain normalized data of the battery under test; and concatenating the normalized data of the battery under test with the current number of battery cycles to obtain battery data of the battery under test.

[0013] In this way, the battery data not only undergoes data cleaning and normalization to improve prediction accuracy, but also incorporates the mapping relationship between the battery's cycle count and its health status, as well as the battery's current cycle count. Adding the current cycle count allows it to be incorporated into the target classification model, enabling the model to more accurately determine the degradation model by considering the impact of the current cycle count.

[0014] In practice, when classification models determine degradation patterns, the length of the input data cannot be changed; for example, a 200×1 dimensional vector is required. Regardless of the current battery cycle count, it must be processed into a 200×1 vector, thus losing the cycle count information. The more battery cycles a battery has, the more the classification pattern tends to shift towards a linear transformation. To incorporate real-world cycle data into the classification model, the current battery cycle count is treated as one-dimensional information and concatenated with the 200×1 vector. The resulting 201×1 dimensional vector is then input into the classification model. This ensures that the cycle count is not lost, increases its influence, and improves the accuracy of lifespan estimation.

[0015] In one possible implementation, the method further includes: determining the goodness of fit of the battery data to the target degradation model based at least on the remaining lifespan of the battery data; the goodness of fit of the battery data to the target degradation model is used to characterize the degree of matching between the battery data and the target degradation model; if the goodness of fit of the battery data to the target degradation model does not meet the conditions, adjusting the model parameters of the target degradation model until the degradation model applicable to the battery data determined by the adjusted target classification model is the adjusted degradation model; wherein the adjusted degradation model is different from the target degradation model.

[0016] In this way, after lifespan prediction, the target degradation model can be further adjusted and optimized based on the prediction data. The target degradation model can be gradually optimized through continuous prediction and optimization. Moreover, the optimization process is based on the goodness of fit of battery data to the target degradation model, and the method of achieving goodness of fit is more accurate.

[0017] In one possible implementation, determining the battery data under the target degradation model based at least on the remaining lifetime of the battery data includes: determining the lifetime label value of the battery data; determining the residual sum of squares of the battery data based on the lifetime label value and the remaining lifetime of the battery data; determining the total sum of squares of the battery data based on the lifetime label value and the remaining lifetime of the battery data; and determining the goodness of fit of the battery data under the target degradation model based on the residual sum of squares of the battery data and the total sum of squares of the battery data.

[0018] In this way, the goodness of fit is determined based on the estimated battery life and the actual battery life, which is flexible, diverse, and reliable. For example, the goodness of fit can be determined based on the sum of squared residuals or the total sum of squares between the estimated and actual battery life.

[0019] In one possible implementation, the method further includes: adjusting the model parameters of the target classification model when the goodness of fit of the determined battery data under each of the at least two decay models does not meet the conditions, until the decay model applicable to the battery data determined by the adjusted target classification model is the decay model corresponding to the maximum goodness of fit among the at least two decay models.

[0020] This approach also considers cases where the goodness-of-fit condition is not met, ensuring that the applicable degradation model for the determined battery data is the degradation model with the highest goodness of fit among at least two degradation models. It is characterized by its simplicity and reliability.

[0021] In one possible implementation, the method further includes: if the battery data is determined to meet the goodness of fit condition by the target degradation model, then the battery data is determined to be incremental data; and the target classification model is incrementally trained based at least on the incremental data.

[0022] Pre-training classification models requires a large amount of training data, which may not be sufficient in practice. This implementation proposes the concept of incremental training. This allows selecting certain battery data that meets the criteria during the prediction process as incremental data to further train the target classification model. Incremental training addresses the problem of limited data volume during classification model training. Furthermore, since the incremental data consists of battery data that meets the criteria, the model's convergence speed can be improved.

[0023] In one possible implementation, incremental training of the target classification model is performed at least based on incremental data, including: acquiring multiple incremental data for battery lifetime estimation; each of the multiple incremental data is battery data whose fit condition is satisfied by the target degradation model; fusing the multiple incremental data and multiple training data and dividing them into multiple batches of incremental training data; determining the degradation model applicable to the current batch of incremental training data among at least two degradation models using the target classification model; determining the total loss of the current batch based at least on the degradation model applicable to the current batch of incremental training data; adjusting the model parameters of the target classification model based on the total loss of the current batch; and determining the degradation model applicable to the next batch of incremental training data based on the adjusted target classification model, until incremental training is completed.

[0024] In this way, the incremental training process not only integrates the incremental data but also the training data, which can prevent the forgetting of old data during the incremental training process.

[0025] In one possible implementation, determining the total loss of the current batch, at least based on the decay model applicable to the incremental training data of the current batch, includes: dividing multiple incremental data used for battery lifetime estimation into multiple batches of incremental training data; processing the incremental data of the current batch using an old classification model to determine the decay model applicable to the incremental data of the current batch among at least two decay models; the old classification model being the target classification model corresponding to the incremental training data of the previous batch of the current batch; and determining the total loss of the current batch based on the decay model applicable to the incremental training data of the current batch and the decay model applicable to the incremental data of the current batch.

[0026] In this way, when determining the total loss of the current batch, the outputs of both the old classification model and the new classification model (the classification model of the current batch) are considered simultaneously, which ensures that the trained classification model does not deviate too far from the old classification model and has high stability.

[0027] In one possible implementation, determining the total loss of the current batch based on the decay model applicable to the incremental training data of the current batch and the decay model applicable to the incremental data of the current batch includes: determining a first loss of the classification model of the current batch relative to the incremental training data of the current batch based on the decay model applicable to the incremental training data of the current batch; determining a second loss of the old classification model relative to the incremental data of the current batch based on the decay model applicable to the incremental data of the current batch; and determining the total loss of the current batch based on the first loss of the classification model of the current batch relative to the incremental training data of the current batch, the second loss of the old classification model relative to the incremental data of the current batch, and the regularization coefficient.

[0028] In this way, when determining the loss for the current batch, the first loss of the new classification model on the incremental data, the second loss of the old classification model on the incremental data, and the regularization coefficient are combined. This not only provides comprehensive loss information, but also allows for adjustment through the regularization coefficient when there is a large difference between the first and second losses, bringing them closer together. Therefore, incremental training achieves better accuracy and stability.

[0029] Secondly, this application provides a model training apparatus, which includes:

[0030] The first acquisition unit is used to acquire battery data of the battery under test; the battery data is used to characterize the mapping relationship between the number of cycles of the battery under test and the health status of the battery under test.

[0031] The first determining unit is used to determine the target degradation model applicable to the battery data from at least two degradation models through the target classification model; the target degradation model is the degradation model corresponding to the fitted curve type of the mapping relationship represented by the battery data.

[0032] The first estimation unit is used to estimate the remaining life of the battery under test based on the target degradation model and battery data.

[0033] Thirdly, this application provides an electronic device including a memory and a processor, wherein the memory stores a computer program or instructions, which, when executed by the processor, implement any of the methods provided in the first aspect.

[0034] Fourthly, this application also provides a storage medium storing a computer program or instructions that, when executed by a processor, implement any of the methods provided in the first aspect above.

[0035] Fifthly, this application also provides a computer program product comprising a computer program or instructions that, when executed by a processor, implement any of the methods provided in the first aspect above.

[0036] It should be noted that battery life estimation devices, electronic devices, storage media, and computer program products have the same technical effect as battery life estimation methods. For the technical effect of battery life estimation devices, electronic devices, storage media, and computer program products, please refer to the detailed description in the first aspect above, which will not be repeated here. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of a first optional process for a battery life estimation method provided in an embodiment of this application;

[0038] Figure 2 This is a schematic diagram of a second optional process for the battery life estimation method provided in the embodiments of this application;

[0039] Figure 3 A schematic diagram of a third optional process for the battery life estimation method provided in the embodiments of this application;

[0040] Figure 4 A schematic diagram of a fourth optional process for a battery life estimation method provided in an embodiment of this application;

[0041] Figure 5 A schematic diagram of a fifth optional process for a battery life estimation method provided in an embodiment of this application;

[0042] Figure 6A schematic diagram of a sixth optional process for a battery life estimation method provided in an embodiment of this application;

[0043] Figure 7 A schematic diagram of a seventh optional process for a battery life estimation method provided in an embodiment of this application;

[0044] Figure 8 This is a schematic diagram of an eighth optional process for a battery life estimation method provided in an embodiment of this application;

[0045] Figure 9 A schematic diagram of a ninth optional process for a battery life estimation method provided in an embodiment of this application;

[0046] Figure 10 A schematic diagram of an optional flowchart of an algorithm that integrates machine learning and mechanistic models, provided in an embodiment of this application;

[0047] Figure 11 An optional schematic diagram illustrating the processing of the machine learning module provided in the embodiments of this application;

[0048] Figure 12 This is a schematic diagram of an optional structure of the machine learning module provided in an embodiment of this application;

[0049] Figure 13 A schematic diagram of an optional structure of the decay curve of the decay mode provided in an embodiment of this application;

[0050] Figure 14 A schematic diagram of an optional incremental training process provided in an embodiment of this application;

[0051] Figure 15 A schematic diagram of an optional structure of the battery life estimation device provided in an embodiment of this application;

[0052] Figure 16 This is a schematic diagram of an optional structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the specific technical solutions of the application will be further described in detail below with reference to the accompanying drawings of the embodiments of this application. The following embodiments are used to illustrate this application, but are not intended to limit the scope of this application.

[0054] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0055] In the following description, the terms "first," "second," and "third" are used only to distinguish different objects and do not represent a specific order of objects, nor are they constituting a chronological order. It is understood that "first," "second," and "third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0056] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0057] This application provides battery life estimation methods, apparatus, devices, media, and products. In practical applications, the battery life estimation method can be implemented by a battery life estimation apparatus. The functional entities within the battery life estimation apparatus can be collaboratively implemented using the hardware resources of the electronic device, such as processors and other computing and communication resources.

[0058] The following describes various embodiments of the battery life estimation method, apparatus, device, medium, and product provided in this application.

[0059] In a first aspect, embodiments of this application provide a battery life estimation method for estimating the remaining lifespan of a battery. The functionality achieved by this method can be implemented by a processor in an electronic device calling program code. Of course, the program code can be stored in a computer storage medium. Therefore, the electronic device includes at least a processor and a memory chip.

[0060] The implementation process of the battery life estimation method is explained below.

[0061] refer to Figure 1 The process may include, but is not limited to, S101 to S103 described below.

[0062] S101, The electronic device acquires the battery data of the battery under test.

[0063] Battery data is used to characterize the mapping relationship between the number of cycles of the battery under test and the health status of the battery under test.

[0064] This application does not specifically limit the type of battery, and can be configured according to actual needs. For example, the battery can be a lithium battery, lead-acid battery, phosphoric acid battery, or other chemical battery.

[0065] The battery under test can be any type of battery, and this application does not limit the type of battery under test.

[0066] Battery life includes cycle life and calendar life; cycle life refers to the number of cycles a battery completes under a certain charge and discharge regime until the end of its life, while calendar life refers to the time a battery is stored in a certain state until the end of its life.

[0067] Battery data refers to relevant data about the battery used for battery life prediction. Battery data includes the mapping relationship between the battery's cycle count and its state of health.

[0068] The number of battery cycles, also known as the battery cycle count, refers to the number of times a battery is charged and discharged. One cycle refers to one complete charge and discharge cycle of a battery.

[0069] State of health (SOH) refers to the proportion of a battery's total usable capacity under certain conditions to the total usable capacity of a newly manufactured battery.

[0070] The mapping relationship refers to the relationship between the number of cycles and the health status. That is, one cycle corresponds to one health status value.

[0071] It should be noted that battery data may also include other content, which will not be listed here. Exemplary battery data may also include tag data. Tag data is used to characterize the actual degradation model to which the battery data applies.

[0072] The electronic device acquires battery data of the battery under test and reads the mapping relationship between the number of cycles of the battery under test and the health status of the battery under test, which is represented by the battery data.

[0073] S102. The electronic device determines the target degradation model applicable to the battery data among at least two degradation models using a target classification model.

[0074] The target degradation model is the degradation model corresponding to the fitted curve type of the mapping relationship represented by battery data.

[0075] The target classification model is a pre-trained model used to predict battery data degradation.

[0076] This application does not limit the training process of the target classification model, and it can be configured according to actual needs. For example, the classification model can be trained based on the direction transfer function of the loss to obtain the target classification model.

[0077] Each of the at least two recession models corresponds to a fitted curve type for the mapping relationship.

[0078] Degradation models are used to determine battery life based on the number of battery cycles.

[0079] This application does not limit the number or specific type of the at least two decay models, and can be configured according to actual needs. For example, the at least two decay models may include: a sublinear + linear decay model, or a sublinear + linear + superlinear decay model. In practice, the at least two decay models may also include other types of decay models, which will not be listed here.

[0080] In one possible implementation, the decay model can be a mechanistic model; correspondingly, at least two decay models can include: a mechanistic model of solid electrolyte interphase (SEI) growth and a mechanistic model of SEI growth + lithium deposition.

[0081] In another possible implementation, the recession model can be an empirical model; correspondingly, at least two recession models can include: a single-exponential empirical model and a double-exponential empirical model.

[0082] In another possible implementation, the decay model can be a hybrid of mechanistic and empirical models. For example, at least two decay models may include: a mechanistic model of SEI growth and an empirical model of double exponential growth; or, at least two decay models may include: an empirical model of single exponential growth and a mechanistic model of SEI growth + lithium deposition.

[0083] Decline models can also be generalized empirical models, semi-mechanistic models, etc., which will not be listed here.

[0084] It should be noted that at least two recession models cannot simultaneously include two recession models of the same type. For example, at least two recession models cannot simultaneously include two sublinear + linear recession models (a mechanistic model of SEI growth and an empirical model of single exponential growth). Alternatively, at least two recession models cannot simultaneously include two recession models of sublinear + linear + superlinear growth (a mechanistic model of SEI growth and lithium deposition and an empirical model of double exponential growth).

[0085] Each degradation model corresponds to a specific type of fitted curve in the mapping relationship. For example, the type of fitted curve representing battery data in the mapping relationship can include: sublinear + linear, sublinear + linear + superlinear, etc. There is a one-to-one correspondence between the two; that is, one fitted curve type corresponds to one degradation model.

[0086] A target classification model is used to determine the appropriate decay model for training data among at least two decay models. This application does not limit the specific type of classification model and can be configured according to actual needs. For example, the target classification model may include, but is not limited to, any of the following: logistic regression, support vector machine, decision tree, random forest, Naive Bayes, K-nearest neighbor algorithm, deep neural network model, etc.

[0087] The target classification model is a pre-trained model used to determine the appropriate degradation model for battery data. Therefore, by inputting battery data from an electrical test battery into the target classification model, the target degradation model applicable to the battery data can be determined from at least two degradation models.

[0088] S103. Electronic devices estimate the remaining life of the battery under test based on the target degradation model and battery data.

[0089] Electronic devices fit battery data to a target degradation curve based on a target degradation model; the remaining life of the battery under test is determined based on the target degradation curve.

[0090] The battery life estimation method provided in this embodiment of the application includes: acquiring battery data of the battery under test; using the battery data to characterize the mapping relationship between the number of cycles of the battery under test and the health status of the battery under test; determining the target degradation model applicable to the battery data from at least two degradation models through a target classification model; the target degradation model is the degradation model corresponding to the fitting curve type of the mapping relationship characterized by the battery data; and estimating the remaining life of the battery under test based on the target degradation model and the battery data.

[0091] First, the target degradation model used for the battery data of the battery under test is determined through a target classification model. Then, the remaining lifespan of the battery under test is predicted using the target degradation model. This allows for the mapping between the target degradation model and the battery data; that is, the type of the fitted curve representing the mapping relationship between the battery cycle count and the battery's health status can be used to determine the corresponding target degradation model. The degradation model corresponding to the fitted curve is then designated as the target degradation model. Because the target degradation model matches the battery data, the accuracy is higher when predicting battery lifespan. Second, since the target classification model can select the degradation mode, it can be applied to various degradation model usage scenarios. Therefore, for battery data corresponding to various types of batteries, the corresponding target degradation model can be determined, exhibiting wide applicability and good transferability. Third, unlike regression models in related technologies, this method combines a target classification model and a degradation model for battery life prediction. It leverages the accuracy of classification models and overcomes the problem of inaccurate degradation model selection in related technologies, simultaneously utilizing the advantages of both and improving the accuracy of lifespan prediction. Finally, since this method differs from the regression models in related technologies, which have poor interpretability compared to classification models, this method improves interpretability.

[0092] The following describes the process by which the electronic device in S103 estimates the remaining life of the battery under test based on the target degradation model and battery data.

[0093] refer to Figure 2 The process may include, but is not limited to, S1031 to S1034 below.

[0094] S1031. The electronic device fits the first mapping relationship included in the battery data to the target degradation model to obtain the target degradation curve.

[0095] The battery data fitted here can be either the original data or the normalized data.

[0096] For example, the raw data corresponding to the battery data is the battery data without normalization processing.

[0097] Electronic devices fit battery data to the expression of the target degradation model to obtain the target degradation curve.

[0098] This application does not impose specific limitations on the fitting algorithm, and it can be configured according to actual needs. For example, the least squares method can be used for fitting.

[0099] S1032. The electronic device determines the health status value at the end of the lifespan of the battery under test.

[0100] The health status value at the end of the lifespan of each battery is generally known, and electronic devices directly read the health status value at the end of the lifespan of the battery under test.

[0101] S1033. The electronic device determines the number of cycles at the end of the lifespan of the battery under test based on the target degradation curve and the health status value at the end of the battery's lifespan.

[0102] The electronic device substitutes the health status value of the battery under test at the end of its lifespan into the target degradation curve to obtain the number of cycles corresponding to the end of the battery under test at the end of its lifespan.

[0103] S1034. The electronic device determines the remaining lifespan of the battery under test based on the number of cycles at the end of the battery's lifespan and the current number of cycles of the battery under test.

[0104] The current number of cycles for each battery is generally known, and the electronic device directly reads the current number of cycles for the battery under test. The difference between the number of cycles at the end of the battery's lifespan and the current number of cycles for the battery under test is determined as the remaining lifespan of the battery under test.

[0105] In this way, the target decay curve is first determined based on the target decay model, and then the remaining lifetime is determined based on the target decay curve, which has the characteristics of being simple and accurate.

[0106] The process of acquiring battery data of the battery under test in S101 will be explained below.

[0107] refer to Figure 3 The process may include, but is not limited to, S1011 to S1013 described below.

[0108] S1011. The electronic device acquires the raw data of the battery under test.

[0109] The raw data of the battery under test includes the mapping relationship between the number of cycles and the health status of the battery under test.

[0110] The raw data of the battery under test is used to characterize the mapping relationship between the number of cycles of the battery under test and the state of health (SOH) of the battery under test.

[0111] The raw data of the battery under test is unnormalized battery data.

[0112] If the raw data of the battery under test consists of different charge / discharge currents, charge / discharge ranges, temperatures, etc., it is necessary to process it according to the definition of SOH to obtain the equivalent relationship between the number of cycles and SOH.

[0113] S1012 The electronic device performs data cleaning and normalization processing on the raw data of the battery under test to obtain the normalized data of the battery under test.

[0114] Data cleaning is necessary because data may be missing or contain noise due to sampling accuracy and anomalies. Outlier removal and imputation are required. Furthermore, it can be determined whether data smoothing filtering is needed. If so, appropriate filters can be used to smooth the data.

[0115] Data normalization is performed to reduce the variability in data distribution and improve the stability of model predictions. For example, mean-variance normalization can be used for data normalization.

[0116] S1013. The electronic device concatenates the normalized data of the battery under test with the current cycle number of the battery under test to obtain the battery data of the battery under test.

[0117] When a target classification model determines a degradation mode, the length of the input data cannot be changed; for example, a 200×1 dimensional vector is required. Regardless of the current battery cycle count, it must be processed into a 200×1 vector, thus losing the cycle count information. The more battery cycles, the more the classification mode tends to shift towards a linear transformation. To incorporate real cycle data into the classification model, the current battery cycle count is treated as one-dimensional information and concatenated with the 200×1 vector. The resulting 201×1 dimensional vector is then input into the target classification model. This ensures that the cycle count is not lost, increases its influence, and improves the accuracy of the model's estimation.

[0118] In this way, the battery data not only undergoes data cleaning and normalization to improve prediction accuracy, but also incorporates the mapping relationship between the battery's cycle count and its health status, as well as the battery's current cycle count. Adding the current cycle count allows it to be incorporated into the target classification model, enabling the model to more accurately determine the degradation model by considering the impact of the current cycle count.

[0119] The battery life estimation method provided in this application can further optimize the target classification model based on the predicted data after the battery life prediction is completed.

[0120] refer to Figure 4 The process may include, but is not limited to, S401 and S402 described below.

[0121] S401. Electronic devices, based at least on the remaining lifespan of battery data, determine the goodness of fit of battery data to a target degradation model.

[0122] The fit of battery data to the target degradation model is used to characterize the degree of matching between battery data and the target degradation model.

[0123] In one possible implementation, the electronic device determines the goodness of fit of the battery data to a target degradation model based on the remaining lifespan of the battery data. For example, the goodness of fit can be determined based on factors such as the magnitude of the deviation value.

[0124] In another possible implementation, the electronic device determines the fit of the battery data to the target degradation model based on the remaining lifespan of the battery data and the actual lifespan of the battery data.

[0125] S402. If the battery data does not meet the goodness of fit under the target degradation model, the electronic device adjusts the model parameters of the target degradation model until the degradation model applicable to the battery data determined by the adjusted target classification model is the adjusted degradation model.

[0126] The adjusted recession model differs from the target recession model.

[0127] This application does not specifically limit the content of the conditions for meeting the criteria, and can be configured according to actual needs. For example, the condition for meeting the criteria can be that the goodness of fit is greater than or equal to the goodness of fit threshold.

[0128] This section does not specify the process for adjusting the model parameters of the target classification model, nor the specific types of target model parameters to be adjusted; these can be configured according to actual needs.

[0129] It should be noted that if the battery data does not meet the goodness of fit of the adjusted degradation model, the degradation model will be adjusted again until the battery data meets the goodness of fit of the adjusted degradation model; or, after iterating through each of the at least two degradation models, no degradation model meets the goodness of fit.

[0130] In this way, after lifespan prediction, the target degradation model can be further adjusted and optimized based on the prediction data. The target degradation model can be gradually optimized through continuous prediction and optimization. Moreover, the optimization process is based on the goodness of fit of battery data to the target degradation model, and the method of achieving goodness of fit is more accurate.

[0131] The following explains the process by which electronic devices in S401 determine the fit of battery data to the target degradation model based at least on the remaining lifespan of battery data.

[0132] refer to Figure 5 The process may include, but is not limited to, S4011 to S4014 described below.

[0133] S4011. Electronic devices determine the life tag value of battery data.

[0134] Electronic devices read the life tag value from the battery data.

[0135] S4012. Electronic devices determine the residual variance sum of squares of battery data based on the lifetime tag value of battery data and the remaining lifetime of battery data.

[0136] Electronic devices substitute the battery life tag value and the remaining battery life into the residual variance sum of squares formula to obtain the residual variance sum of squares of the battery data.

[0137] S4013. Electronic devices determine the total square of battery data based on the battery data's lifetime tag value and the remaining lifetime of the battery data.

[0138] Electronic devices substitute the battery life tag value and the remaining battery life into the sum of squares formula to obtain the total sum of squares of the battery data.

[0139] S4014. Electronic devices determine the goodness of fit of battery data to a target degradation model based on the residual sum of squares of battery data and the total sum of squares of battery data.

[0140] In one possible implementation, the electronic device first determines the ratio of the residual sum of squares to the total sum of squares, and then determines the result of subtracting that ratio as the goodness of fit of the battery data to the target degradation model.

[0141] Of course, there are other ways to determine the fit of battery data to the target degradation model, which will not be elaborated here.

[0142] In this way, the goodness of fit is determined based on the estimated battery life and the actual battery life, which is flexible, diverse, and reliable. For example, the goodness of fit can be determined based on the sum of squared residuals or the total sum of squares between the estimated and actual battery life.

[0143] It should be noted that if the goodness of fit of the determined battery data under each of the at least two degradation models does not meet the conditions, the electronic device adjusts the model parameters of the target classification model until the degradation model applicable to the battery data determined by the adjusted target classification model is the degradation model corresponding to the maximum goodness of fit among the at least two degradation models.

[0144] This approach also considers cases where the goodness-of-fit condition is not met, ensuring that the applicable degradation model for the determined battery data is the degradation model with the highest goodness of fit among at least two degradation models. It is characterized by its simplicity and reliability.

[0145] The battery life estimation method provided in this application embodiment can be incrementally trained again after battery life estimation to improve the accuracy of the target classification model.

[0146] refer to Figure 6 The process may include, but is not limited to, S601 and S602 described below.

[0147] S601. If the goodness of fit of the battery data is determined to meet the conditions through the target degradation model, the electronic device determines the battery data as incremental data.

[0148] In other words, electronic devices use battery data from the target classification model that meets the goodness-of-fit criteria during the battery life prediction process as incremental data. Thus, as battery life is continuously predicted, and as the target degradation model is continuously used, incremental data can be continuously obtained.

[0149] S602. Electronic devices shall at least perform incremental training on the target classification model based on incremental data.

[0150] In one possible implementation, the electronic device incrementally trains the target classification model based on incremental data.

[0151] In another possible implementation, the electronic device incrementally trains the target classification model based on incremental data and training data.

[0152] The embodiments of this application do not limit the timing of incremental training, and can be configured according to timing requirements.

[0153] In one possible implementation, after the incremental data has accumulated to a certain amount, an incremental training process is initiated to further optimize the target classification model.

[0154] In another possible implementation, each time incremental data is obtained, the incremental data can be fused with the training data to further optimize the target classification model.

[0155] Pre-training classification models requires a large amount of training data, which may not be sufficient in practice. This implementation proposes the concept of incremental training. This allows selecting certain battery data that meets the criteria during the prediction process as incremental data to further train the target classification model. Incremental training addresses the problem of limited data volume during classification model training. Furthermore, since the incremental data consists of battery data that meets the criteria, the model's convergence speed can be improved.

[0156] The following describes the process by which the electronic device in S602 incrementally trains the target classification model based at least on incremental data.

[0157] refer to Figure 7 The process may include, but is not limited to, S6021 to S6026 described below.

[0158] S6021. The electronic device acquires multiple incremental data for battery life estimation.

[0159] Each of the multiple incremental data points is battery data that meets the fit criteria determined by the target decay model.

[0160] The embodiments of this application do not limit the specific amount of incremental data, which can be configured according to actual needs.

[0161] S6022. Electronic devices merge multiple incremental data and multiple training data and divide them into multiple batches of incremental training data.

[0162] The embodiments of this application do not limit the amount of training data, which can be configured according to actual needs.

[0163] This application does not limit the specific fusion method and division method of incremental data and training data in the embodiment, and can be configured according to actual needs.

[0164] In incremental training, incremental data and training data are taken according to certain rules to achieve the fusion of new and old data. Incremental data refers to new data, and training data refers to old data. For example, the batches of incremental and training data can be divided in a 3:1 ratio (the amount of data in each batch may vary). During iterative training, training is performed three times using the new data and once using the old data. For example, there may be three batches of new data and only one batch of old data. Each batch of new data may contain 50 data points, and each batch of old data may contain 200 data points.

[0165] S6023. Electronic devices determine the applicable decay model for the incremental training data of the current batch from at least two decay models using a target classification model.

[0166] The incremental training data for the current batch here can be incremental data, training data, or a mixture of incremental data and training data.

[0167] The incremental training data of the current batch is input into the target classification model, which determines the appropriate decay model for the incremental training data of the current batch from at least two decay models.

[0168] S6024. The electronic device determines the total loss of the current batch based at least on the decay model applicable to the incremental training data of the current batch.

[0169] The embodiments of this application do not limit the method for determining the total loss, and can be determined according to actual needs.

[0170] In one possible implementation, the electronic device determines the total loss of the current batch based on the decay model applicable to the incremental training data of the current batch.

[0171] In another possible implementation, the electronic device determines the total loss of the current batch based on the decay model applicable to the incremental training data of the current batch and the decay model applicable to the incremental data of the current batch.

[0172] S6025. The electronic device adjusts the model parameters of the target classification model based on the total loss of the current batch.

[0173] For example, the model parameters of the classification model can be adjusted using stochastic gradient descent based on the total loss of the current batch.

[0174] S6026, electronic equipment, and based on the adjusted target classification model, determine the decay model applicable to the next batch of incremental training data, until incremental training is completed.

[0175] The incremental training process for each batch of incremental training data is similar and will not be described in detail here. Incremental training is completed when the target classification model converges or the incremental training data training ends.

[0176] In this way, the incremental training process not only integrates the incremental data but also the training data, which can prevent the forgetting of old data during the incremental training process.

[0177] The following describes the process by which the electronic device in S6024 determines the total loss of the current batch based at least on the decay model applicable to the incremental training data of the current batch.

[0178] Referring to the content shown in 8, the process may include, but is not limited to, S60241 to S60243 described below.

[0179] S60241. An electronic device divides multiple incremental data used for battery life estimation into multiple batches of incremental data.

[0180] The specific division method can be configured according to actual needs, and will not be elaborated here.

[0181] S60242. Electronic devices process incremental data of the current batch using an old classification model to determine the applicable decay model for the incremental data of the current batch among at least two decay models.

[0182] The old classification model is the target classification model corresponding to the incremental training data of the previous batch in the current batch.

[0183] Incremental training data is equivalent to new data. The new data is input into the old model for processing and training. That is, the new data is trained by the old model to obtain the classification result of the new data in the old model (the decay model applicable to the current batch of incremental data obtained by the old classification model).

[0184] S60243. The electronic device determines the total loss of the current batch based on the decay model applicable to the incremental training data of the current batch and the decay model applicable to the incremental data of the current batch.

[0185] The total loss for the current batch is determined based on the classification results of the new data in the old model and the classification results of the old data in the new model.

[0186] In this way, when determining the total loss of the current batch, the outputs of both the old classification model and the new classification model (the classification model of the current batch) are considered simultaneously, which ensures that the trained classification model does not deviate too far from the old classification model and has high stability.

[0187] The following describes the process by which the electronic device in S60243 determines the total loss of the current batch based on the decay model applicable to the incremental training data of the current batch and the decay model applicable to the incremental data of the current batch.

[0188] refer to Figure 9 The process may include, but is not limited to, S901 to S903 described below.

[0189] S901, the electronic device determines the first loss of the classification model for the current batch relative to the incremental training data of the current batch based on the decay model applicable to the incremental training data of the current batch.

[0190] Specifically, the first loss can be determined using cross-entropy. The first loss can also be understood as the loss of the new model with respect to incremental training data.

[0191] S902. The electronic device determines the second loss of the old classification model relative to the incremental data of the current batch based on the decay model applicable to the incremental data of the current batch.

[0192] Specifically, the second loss can be determined using cross-entropy. The second loss can also be understood as the loss of the old model for the new data (incremental data).

[0193] S903. The electronic device determines the total loss of the current batch based on the first loss of the classification model of the current batch relative to the incremental training data of the current batch, the second loss of the old classification model relative to the incremental data of the current batch, and the regularization coefficient.

[0194] The regularization coefficient can be configured according to actual needs. For example, if there is a large difference between the first loss of the current batch of classification models relative to the incremental training data of the current batch and the second loss of the old classification models relative to the incremental data of the current batch, the regularization coefficient can be used to reduce the difference between the two.

[0195] In this way, when determining the loss for the current batch, the first loss of the new classification model on the incremental data, the second loss of the old classification model on the incremental data, and the regularization coefficient are combined. This not only provides comprehensive loss information, but also allows for adjustment through the regularization coefficient when there is a large difference between the first and second losses, bringing them closer together. Therefore, incremental training achieves better accuracy and stability.

[0196] The following describes the battery life prediction process provided in this application embodiment using a lithium battery as an example.

[0197] For predicting battery life, the relevant technologies mainly include Scheme 1 and Scheme 2.

[0198] Option 1 is an online battery life prediction based on a big data cloud platform using mechanistic information. This mainly includes: collecting battery charge-discharge cycle data under various fast-charging protocols and establishing a database; creating capacity-voltage curves from the discharge cycle data, extracting capacity difference-voltage curves for battery samples at different times (the capacity difference-voltage curves contain aging mechanism information); performing dimensionality reduction on the curves to obtain feature data, using this feature data as input, and the remaining lifespan and capacity degradation inflection point of the battery samples as outputs to train a regression model, resulting in a prediction model; using existing batteries as the batteries to be predicted, reading the discharge cycle data of the batteries to be predicted, predicting the remaining lifespan and capacity degradation inflection point of the batteries using the prediction model, and classifying the lifespan accordingly.

[0199] Option 2 uses data-driven predictive modeling to predict and classify battery cells by lifetime, primarily including: collecting a training dataset by cycling the battery cells between voltages V1 and V2; continuously measuring the battery cell voltage, current, case temperature, and internal resistance during cycling; generating a voltage profile for each cell, where the voltage profile depends on the discharge capacity of a given cycle; calculating the cycle-to-cycle evolution of cell charge using data from the discharge voltage profiles to output a curve Q(V) of cell voltage relative to charge; generating a transformation of ΔQ(V); using the algorithm to generate a transformation of the data stream including capacity, temperature, and internal resistance; applying a machine learning model to determine combinations of subsets of transformations to predict cell operating characteristics; and applying a machine learning model to output the predicted battery operating characteristics.

[0200] An analysis of various battery life prediction methods in related technologies is conducted.

[0201] Lithium-ion batteries are characterized by high energy density, long cycle life, low self-discharge rate, and small size and light weight. They are widely used in mobile communications, electric vehicles, energy storage systems, portable electronic devices, and have significant importance and potential in large-scale energy storage and power grid dispatch.

[0202] Battery life is one of the important indicators for measuring battery electrical performance. State of health (SOH) represents the percentage of a battery's total usable capacity under certain conditions relative to the total usable capacity of a new battery. SOH is an important state parameter of the Battery Management System (BMS) and a key indicator for quantitatively evaluating battery life. For energy-type batteries, it is generally considered that the battery's lifespan ends when the SOH drops to 80%. Battery life includes cycle life and calendar life. The former refers to the number of cycles the battery undergoes under a certain charge-discharge regime until the end of its lifespan, while the latter refers to the time the battery is stored in a certain state until the end of its lifespan.

[0203] During battery use, numerous internal and external factors cause battery capacity degradation, leading to a decrease in the State of Health (SOH), a process known as battery aging. The aging process can be broadly divided into two stages. In the first stage, battery capacity / SOH degradation is linearly related to time or the number of cycles. In the second stage, battery capacity / SOH exhibits a non-linear, accelerated degradation, causing a rapid decline in battery performance; this process is often referred to as a capacity / SOH "plunge." The turning point between these two stages is called the inflection point or the "plunge point." After the inflection point, the rate of capacity / SOH degradation accelerates, battery efficiency and reliability decrease, and safety hazards arise. In practical use, internal and external factors significantly influence the battery degradation process. Internal factors include battery materials, battery design, and manufacturing processes, while external factors include battery usage strategies, the Battery Management System (BMS), and ambient temperature. Different internal and external factors lead to different degradation trends, resulting in the inflection point in capacity / SOH and the second stage of degradation possibly occurring or not, and the timing and number of cycles at which they occur are not uniform.

[0204] As research deepens, related technologies have proposed predictive models to forecast battery life. Currently, the main methods predict the remaining lifespan of lithium-ion batteries by establishing mechanistic models or data-driven approaches. Mechanism-based physical models primarily quantify the factors influencing battery performance, describing the evolution of lithium-ion battery performance degradation based on the chemical reactions occurring during charging and discharging and its specific physical characteristics. The biggest challenge currently is that many factors interact during battery performance evolution, leading to performance decline, making it difficult to reliably simulate the complex dynamics of battery aging. Due to the complexity of existing technologies and actual battery operating conditions, it is difficult to use a unified mechanistic model to predict battery life.

[0205] A data-driven approach offers a method for predicting the remaining lifespan of batteries. This method integrates historical data from the battery itself with cloud-based big data. During implementation, it fully utilizes online historical data from the battery and cloud-based big data containing non-basis information to predict the remaining battery lifespan. It offers numerous advantages, including simple steps, high prediction accuracy, significantly improved utilization of available information sources, and support for any amount of non-basis information as prediction input, making it well-suited for big data applications. However, data-driven methods typically only use capacity degradation information. These methods lack a physical information foundation, have limited practical significance for representing battery lifespan, require extensive training and large datasets, are prone to overfitting and underfitting, and are time-consuming, resulting in low efficiency in predicting remaining battery lifespan.

[0206] The two traditional tasks of machine learning algorithms are regression and classification. Generally, regression is used in machine learning for predicting battery remaining life. When the target task is difficult, regression tasks are prone to overfitting and underfitting. If the model structure is too simple, the model cannot fit the complex output, leading to underfitting. However, when the model structure is too complex, the model will memorize training samples to reduce training error, resulting in small errors on the training set but large errors on the test set, leading to overfitting. Even though machine learning-based remaining useful life (RUL) prediction is widely studied, the complex electrodynamic system behind battery aging makes machine learning less transferable to different battery systems. In addition, regression tasks are sensitive to data noise, which affects the accuracy of predictions. Classification tasks generally have a finite number of class labels, making them easier than regression. Classification does not require high-precision fitting of the target and can tolerate a certain threshold of error. For example, for binary classification problems, determining whether a sample belongs to the first class only requires an output probability value greater than 0.5, without needing an exact probability value. Therefore, it is less sensitive to data noise and has relatively higher accuracy. Classification tasks typically provide good interpretations of the results and are more intuitive and easier to understand. Battery capacity / SOH degradation patterns have only a few fixed modes, each with a different dominant degradation mechanism and a different mechanistic model. Machine learning can effectively distinguish and classify different degradation modes, thereby selecting different mechanistic models based on different classifications.

[0207] Therefore, this embodiment proposes a novel scheme that integrates machine learning algorithms and mechanistic model algorithms for estimating the remaining battery life. This scheme uses machine learning algorithms to classify lithium-ion battery degradation modes, and then selects a mechanistic model based on the classification results to predict the battery's lifespan. This scheme fully leverages the high accuracy of machine learning in classification tasks while overcoming the challenges of mechanistic model selection in traditional schemes. It maximizes the advantages of both models, improves the accuracy of remaining battery life estimation, and provides a reasonable interpretation of the estimation results.

[0208] The technical problem to be solved by this embodiment

[0209] 1. Current methods for estimating battery lifespan can be broadly categorized into machine learning-based and mechanistic model-based methods, each with its own advantages and disadvantages. Machine learning-based methods typically utilize only capacity degradation information, lacking a physical information foundation. This results in limited practical representation of battery lifespan, requires extensive training with large amounts of data, and is prone to overfitting and underfitting. Furthermore, these methods are time-consuming and inefficient in predicting remaining battery lifespan. Mechanism-based physical models primarily quantify the factors influencing battery performance, describing the evolution of lithium-ion battery performance degradation based on the chemical reactions and specific physical characteristics that occur during charging and discharging. The biggest challenge currently is that many factors interact during battery performance evolution, leading to performance degradation, making it difficult to reliably simulate the complex dynamics of battery aging. Due to the complexity of actual battery operating conditions, it is difficult to use a unified mechanistic model to predict battery lifespan.

[0210] 2. Current methods for using machine learning models to predict battery life typically involve training them on large amounts of data, and then using the trained model for prediction. This approach requires accumulating a large amount of data and obtaining labels before training and application. It cannot be directly applied to small batches of data. Furthermore, current machine learning-based battery life predictions are limited to single operating conditions and battery types, exhibiting poor transferability across different battery degradation patterns. In addition, the lack of physical information in machine learning models weakens the interpretability of the output results, leading to unreliable outcomes.

[0211] 3. Traditional mechanistic models are general rules summarized based on a large number of experiments, and then mathematical, physical, and chemical formulas are extracted and summarized based on these general rules. The choice of mechanistic model will have a significant impact on the prediction results for different operating conditions and types of batteries.

[0212] This embodiment of the application overcomes the shortcomings of both machine learning-based models and mechanism-based models for battery life prediction, while making full use of the advantages of both models to develop a fusion model to improve the accuracy and interpretability of battery life prediction.

[0213] This embodiment utilizes a machine learning model, requiring a large amount of data and labels for model training. The size of the training set and the quality of the labels directly impact prediction accuracy. For small batches of data, the application of machine learning models is limited. This embodiment proposes a simple continuous learning scheme, training on small batches of data for prediction. Simultaneously, high-confidence predictions are accumulated for incremental learning in machine learning. Through continuous incremental learning, the overall accuracy of the algorithm is improved.

[0214] Different degradation mechanisms lead to different battery degradation modes, and these modes vary depending on the type of battery, manufacturing process, and operating conditions. The machine learning model used in this embodiment improves the accuracy of mechanism model selection through learning and training on batch data, thereby enhancing the interpretability of the model's battery performance prediction results.

[0215] This embodiment integrates machine learning and mechanistic models for battery remaining life prediction, avoiding the shortcomings of each model while fully utilizing their advantages to improve the accuracy of battery life prediction. Simultaneously, the mechanistic model enhances the interpretability and reliability of the prediction results from the fused model.

[0216] This embodiment proposes an incremental learning scheme for continuously improving the prediction accuracy of machine learning models. High-confidence data generated by the model predictions are accumulated and used for incremental learning in machine learning. Through continuous incremental learning, the overall accuracy of the algorithm is improved.

[0217] The specific implementation process of this embodiment will be described below.

[0218] This embodiment uses machine learning for battery degradation pattern recognition. Suitable machine learning models include, but are not limited to, logistic regression, support vector machines, decision trees, random forests, Naive Bayes, K-nearest neighbors, and deep neural network models. This example uses a battery degradation curve based on a mechanistic model for illustration. The mechanistic model in this example can be generalized to empirical models and semi-mechanistic models based on mechanistic model generalization. Using different machine learning models for degradation pattern recognition and using different mechanistic / empirical models as input to the fusion model are all within the scope of this embodiment.

[0219] This embodiment of the application proposes an algorithm that integrates machine learning and mechanistic models to improve the accuracy of battery life prediction. (Reference) Figure 10 The overall process of this embodiment may include, but is not limited to, the following S1001 to S1007.

[0220] S1001, The data processing module performs data processing;

[0221] S1002. The machine learning module is trained based on the training set to obtain the mechanism selection module;

[0222] S1003, The mechanism model selection module selects the mechanism model based on the test set;

[0223] S1004, the RUL calculation module performs RUL calculations based on the selected target mechanism model;

[0224] S1005. The feedback module calculates the goodness of fit based on the RUL calculation results and determines the relationship between the goodness of fit and the fitting threshold.

[0225] If the goodness of fit is greater than or equal to the fitting threshold, output the result; if the goodness of fit is less than the fitting threshold, return and re-execute S1003.

[0226] If the goodness of fit is greater than or equal to the fit threshold, then execute S1006 as described below.

[0227] S1006. Use this training data as incremental data for the next training step;

[0228] S1007. Determine whether the number of incremental data is greater than the incremental number threshold.

[0229] If the number of incremental data exceeds the incremental data threshold, the incremental data will be added to the training set to retrain the machine learning module.

[0230] The solution includes a data processing module, a machine learning module, a feedback module, an incremental learning module, and a RUL calculation module. The following section will introduce each module in turn.

[0231] 1. Data Processing Module:

[0232] The data processing module mainly includes: data acquisition, data cleaning, and data normalization.

[0233] Data acquisition mainly includes: acquiring the original measurement data of the cell to be monitored, including cycle capacity, number of cycles, time, current, voltage, temperature, etc. Based on the following formula (1), the cycle number-cycle capacity of the battery is converted into the corresponding relationship of cycle number-SOH. For different charge and discharge currents, charge and discharge ranges, temperatures, and other test data, it is necessary to process them according to the definition of SOH to obtain the equivalent corresponding relationship of cycle number-SOH.

[0234]

[0235] In formula (1), This is the initial capacity of the battery. This is the capacity for looping up to n times.

[0236] For data cleaning, due to factors such as sampling accuracy and sampling anomalies, the data may be missing or contain noise. Outlier removal and imputation are necessary. Depending on the requirements, it should be determined whether data smoothing filtering is needed. If so, appropriate filters should be used to smooth the data.

[0237] For data normalization, in order to reduce the differences in data distribution and improve the stability of model training, the input data is normalized. For example, mean-variance normalization can be adopted based on the following formula (2).

[0238]

[0239] In formula (2), Let x represent the mean of the sequence, σ represent the standard deviation of the sequence, and x represent the mean of the sequence. i This represents the data to be normalized.

[0240] 2. Machine Learning Module:

[0241] For processing of the machine learning module, please refer to... Figure 11 The content shown includes, but is not limited to: raw data 110A, training set 110B, and test set 110C; the processing steps may include, in sequence: label acquisition 1101, data partitioning 1102, model training 1103, model prediction 1104, and output results 1105.

[0242] Label acquisition 1101 is used to process the original data 110A to obtain training set 110B and test set 110C.

[0243] The training set 110B is used as input to the data partition 1102; the test set 110C is used as input to the model prediction 1104.

[0244] Among them, module training 1103 may include, but is not limited to: data selection 11031, model calculation 11032, loss calculation 11033, gradient descent 11034.

[0245] The training process may include: inputting training set 110B into data selection 11031, sequentially performing model calculation 11032, loss calculation 11033, and gradient descent 11034, and then determining whether the number of iterations has been reached. If the number of iterations has been reached, the process ends; if the number of iterations has not been reached, the process returns to re-execute inputting training set 110B into data selection 11031 until the number of iterations has been reached.

[0246] The following sections will explain each process of the machine learning module in turn.

[0247] Label Acquisition: The processed data is manually judged and classified. The number of labels corresponds to the number of mechanistic models. For example, two mechanistic models are selected as inputs to the fusion model, with labels set to 0 and 1 respectively. The standard for setting the labels is based on the shape of the SOH sequence.

[0248] Data partitioning: The data is divided into three parts: a training set, a test set, and a validation set, in a ratio of 7:2:1. The training set is used for model training, the validation set is used to select the best-performing model weights during training, and the test set is used for inference in the prediction phase.

[0249] Model Training: The labeled training set is input into the machine learning model, and the loss is reduced through iteration to improve the model's accuracy. This example uses a multilayer perceptron model (taking a 4-layer model as an example) to illustrate the model training process. The structure of the multilayer perceptron is shown in the figure below. The battery degradation curve x and the current cycle number c are concatenated as the model input, and after passing through a fully connected layer, the intermediate representation h is obtained.

[0250] h = max(W[x, c] + b, 0) Formula (3);

[0251] The fully connected layer consists of a single matrix multiplication and a ReLU activation function max(,). In formula (3), h represents the intermediate features of the fully connected layer output; the max(,) function is used to determine the maximum value between W[x,c]+b and 0; and x represents the input matrix.

[0252] After three fully connected layers, the probability value of a sample belonging to the corresponding category is finally obtained through the sigmoid function. For details, please refer to formula (4).

[0253]

[0254] In formula (4), h represents the intermediate feature output by the fully connected layer, and p(x) is the probability value of a sample belonging to the corresponding category.

[0255] The loss function using binary cross-entropy as the optimization objective can be found in formula (5).

[0256]

[0257] In formula (5), N is the batch size, i is the batch number; p i y represents the probability value determined by the classification model. i This represents the actual value corresponding to the label. The model is iteratively optimized using stochastic gradient descent until the specified number of iterations is reached.

[0258] The structure of the machine learning module can be found by referring to Figure 12 The processing described may include: inputting the input data (decay curve, i.e., training data and current loop count) sequentially into FC layer 1201, RELU layer 1202, FC layer 1201, RELU layer 1202, FC layer 1201, RELU layer 1202 and FC layer 1201 and then outputting the data.

[0259] Model prediction: Assuming there are 2 mechanistic models, after inputting the battery degradation curve and the number of cycles into the model, a probability value of 0-1 will be obtained. When the probability is less than 0.5, the battery degradation curve is classified as the first type, and the first mechanistic model should be used to predict the RUL. Similarly, when the probability is greater than 0.5, the battery degradation curve is classified as the second type, and the second mechanistic model should be used to predict the RUL.

[0260] 3. Mechanism Model Selection and Calculation Module

[0261] Battery SOH (State of Harshness) degradation curves are generally divided into sublinear, linear, and superlinear stages. For batteries operating under different conditions and of different types, the degradation curve may be a combination of one, two, or three of these trends. This example illustrates a scheme using a combination of sublinear + linear and sublinear + linear + superlinear.

[0262] Sublinear and linear decay patterns are generally considered to be caused by SEI growth.

[0263] Superlinearity is generally believed to be caused by lithium deposition.

[0264] The mathematical expressions for the decay modes of the two mechanistic models are shown in Table 1, and the decay curves are as follows: Figure 13 As shown.

[0265] Table 1. Examples of mathematical expressions for the decay modes in the mechanistic model.

[0266]

[0267] Where R represents the gas constant; T represents the temperature; E SEI This represents the skin interaction energy for the formation of SEI, where SEI stands for solid electrolyte membrane; k SEI Q represents the reaction rate at which SEI is formed. SEI() This indicates the capacity loss caused by SEI; tz In this context, z represents a constant coefficient, t represents time (number of cycles), and LP represents lithium deposition.

[0268] After fitting the data, the optimal coefficients of k and E can be obtained; thus, given t (number of cycles), SOH can be calculated using t.

[0269] Based on the mechanistic model, some simpler empirical models have been proposed for describing battery degradation curves, such as the single-exponential model and the double-exponential model, as shown in Table 2.

[0270] Table 2. Examples of mathematical expressions for the decline modes in empirical models.

[0271]

[0272] Here, a and b represent coefficients, and the optimal coefficients can be obtained through a fitting algorithm.

[0273] The mechanism model selection module selects the appropriate mechanism model based on the classification results of the machine learning module. The selected mechanism model is then used to fit the battery cycling data. The fitting method used is nonlinear least squares, and the optimization algorithm employed is the built-in nonlinear least squares iterative optimization algorithm in Python.

[0274] Figure 13 This is a schematic diagram of the decay curves in a decay pattern. The dashed lines represent the sublinear + linear decay curves, and the solid lines represent the decay curves of a combination of sublinear + linear + superlinear types.

[0275] RUL calculation module:

[0276] The selected mechanistic model and fitting results are used as the final model and parameters. RUL is calculated based on the input data and EOL parameters. The solution method is the conventional solution of nonlinear equations. For example: the input data is [x i y i ], i = 1…n, the model given by the feedback module after judgment is a sublinear + linear pattern, for example, SOH(k) = aexp(bk), and the optimal parameter is a. * and b * Solve for EOL = a * exp(b * k), we can get k * k * -max(x i The value is the RUL value, where max(x) i It should be less than k * Otherwise, output 0.

[0277] 4. Feedback Module

[0278] The feedback model is used to determine the goodness of fit (equivalent to the goodness of fit mentioned above). The goodness of fit for nonlinear fitting can be determined using formula (6).

[0279]

[0280] In formula (6), This represents the error sum of squares. This represents the total sum of squares. For y i The average value.

[0281] Let the goodness-of-fit threshold be I. When the calculated goodness-of-fit is greater than this threshold, i.e., Rfit... 2 If the result is >1, it indicates that the machine learning model accurately identifies the battery degradation mode. Otherwise, the process returns to the mechanism selection module for re-selection and fitting calculation, followed by a judgment from the feedback module. If the goodness of fit of all mechanism models is less than the threshold, the mechanism model with the highest goodness of fit is selected as the final model.

[0282] 5. Incremental Learning Module

[0283] The accuracy of machine learning algorithms depends on both the model design and the amount of data. Large-scale data enables the model to learn more general features, thereby improving generalization and reducing the probability of overfitting. In this embodiment, data continues to accumulate after model training and prediction. Therefore, to utilize the newly added data to improve model accuracy, this embodiment proposes an incremental learning module to continuously incorporate incremental data into the training process.

[0284] Training a model directly using incremental data can lead to catastrophic forgetting, meaning the model forgets previous training results and overfits to new data. Therefore, it's not advisable to directly train a model using new data. To address this forgetting problem, this embodiment merges old and new data to retrain the model. However, the incremental data volume is relatively small compared to the old data, resulting in high training costs for each iteration. Therefore, an efficient method is employed to enable the model to learn from new data without forgetting. This method combines replay and regularization to ensure the stability of incremental learning. The proposed method will be described below.

[0285] refer to Figure 14The incremental training process, as shown, can include: inputting new and old data 1401 into the new model 1402, processing it through the new model 1402, and then feeding it into the output module 1403; processing new data 1404 through the old model 1405 and then feeding it into the output module 1403; and after processing by the classification loss 1406 and gradient descent 1407, the output module 1403 determines whether the number of iterations has been reached. If yes, the process ends; otherwise, it returns to the next round of data processing until the number of iterations is obtained.

[0286] Among them, classification loss 1406 needs to be combined with label 1408 when determining classification loss.

[0287] The replay method involves adding new data to the old data at regular intervals during training. This prevents the model from forgetting the old data during training and balances the additional training cost based on the set intervals. Specifically, new and old data can be combined into a new training set in a certain ratio, and the new and old data can be divided into different batches. During model training, the new and old data sets are selected according to certain rules to achieve the fusion of new and old data. For example, the number of batches of new and old data can be divided into 3:1 (the amount of data in each batch of new and old data can be different), and during iterative training, the model is trained 3 times using the new data and 1 time using the old data.

[0288] Regularization utilizes the concept of knowledge distillation to prevent the model from deviating too far from the original model during optimization. Specifically, this method first obtains the predictions of the old model on new data, introduces the distillation loss of the new model's output into the loss function, and then trains the model on the new data using fine-tuning. This avoids over-adjusting the parameters of the old model during training for the new task, which could lead to a decrease in the performance of the new model on the old task. This method ensures that the distribution of the outputs of the new model and the old model remains unchanged after training, mitigating the new model's forgetting of knowledge from the old data.

[0289] Finally, the loss function for incremental learning consists of two terms: one is the classification loss of the new model on the old data and the incremental data, and the other is the regularization of the new model output and the old model. For details, please refer to formula (7).

[0290]

[0291] In formula (7), This represents the total training loss. This refers to the loss of the new model. Y represents the loss of the old model. new and Y old These represent new data and old data, respectively. and Labels representing new and old data. λ represents the regularization coefficient.

[0292] Secondly, embodiments of this application provide a battery life estimation device, such as... Figure 15 As shown, the battery life estimation device 150 includes: a first acquisition unit 1501, a first determination unit 1502, and a first estimation unit 1503.

[0293] in:

[0294] The first acquisition unit 1501 is used to acquire battery data of the battery under test; the battery data is used to characterize the mapping relationship between the number of cycles of the battery under test and the health status of the battery under test.

[0295] The first determining unit 1502 is used to determine the target degradation model applicable to the battery data from at least two degradation models through the target classification model; the target degradation model is the degradation model corresponding to the fitting curve type of the mapping relationship represented by the battery data.

[0296] The first estimation unit 1503 is used to estimate the remaining life of the battery under test based on the target degradation model and battery data.

[0297] In some embodiments, the first estimation unit 1503 is further configured to: fit the battery data using a target degradation model to obtain a target degradation curve; determine the health status value at the end of the battery's lifespan; determine the number of cycles at the end of the battery's lifespan based on the target degradation curve and the health status value at the end of the battery's lifespan; and determine the remaining lifespan of the battery based on the number of cycles at the end of the battery's lifespan and the current number of cycles of the battery.

[0298] In some embodiments, the first acquisition unit 1501 is further configured to: acquire the raw data of the battery under test; the raw data of the battery under test includes the mapping relationship between the number of cycles of the battery under test and the health status of the battery under test; perform data cleaning and normalization processing on the raw data of the battery under test to obtain the normalized data of the battery under test; and concatenate the normalized data of the battery under test with the current number of cycles of the battery under test to obtain the battery data of the battery under test.

[0299] In some embodiments, the battery life estimation device 150 may further include an adjustment unit, which is configured to: determine the goodness of fit of the battery data under a target degradation model based at least on the remaining life of the battery data; the goodness of fit is used to characterize the degree of matching between the battery data and the target degradation model; if the goodness of fit of the battery data under the target degradation model does not meet the conditions, adjust the model parameters of the target degradation model until the degradation model applicable to the battery data determined by the adjusted target classification model is the adjusted degradation model; wherein the adjusted degradation model is different from the target degradation model.

[0300] In some embodiments, the adjustment unit is further configured to: determine the lifetime label value of the battery data; determine the residual sum of squares of the battery data based on the lifetime label value of the battery data and the remaining lifetime of the battery data; determine the total sum of squares of the battery data based on the lifetime label value of the battery data and the remaining lifetime of the battery data; and determine the goodness of fit of the battery data to the target degradation model based on the residual sum of squares of the battery data and the total sum of squares of the battery data.

[0301] In some embodiments, the adjustment unit is further configured to: adjust the model parameters of the target classification model when the goodness of fit of the determined battery data under each of the at least two degradation models does not meet the conditions, until the degradation model applicable to the battery data determined by the adjusted target classification model is the degradation model corresponding to the maximum goodness of fit among the at least two degradation models.

[0302] In some embodiments, the battery life estimation device 150 may further include an incremental training unit, which is used to: determine the battery data as incremental data if the goodness of fit of the battery data is determined by the target degradation model; and perform incremental training on the target classification model based at least on the incremental data.

[0303] In some embodiments, the incremental training unit is further configured to: acquire multiple incremental data for battery lifetime estimation; each of the multiple incremental data is battery data whose fit condition is satisfied by the target degradation model; fuse the multiple incremental data and multiple training data and divide them into multiple batches of incremental training data; determine the degradation model applicable to the current batch of incremental training data among at least two degradation models using the target classification model; determine the total loss of the current batch based at least on the degradation model applicable to the current batch of incremental training data; adjust the model parameters of the target classification model based on the total loss of the current batch; and determine the degradation model applicable to the next batch of incremental training data based on the adjusted target classification model, until incremental training is completed.

[0304] In some embodiments, the incremental training unit is further configured to: divide multiple incremental data for battery lifetime estimation into multiple batches of incremental data; process the current batch of incremental data using an old classification model to determine the applicable decay model for the current batch of incremental data among at least two decay models; the old classification model is the target classification model corresponding to the previous batch of incremental training data for the current batch; and determine the total loss for the current batch based on the decay model applicable to the current batch of incremental training data and the decay model applicable to the current batch of incremental data.

[0305] In some embodiments, the incremental training unit is further configured to: determine a first loss of the classification model for the current batch relative to the incremental training data of the current batch based on the decay model applicable to the incremental training data of the current batch; determine a second loss of the old classification model relative to the incremental data of the current batch based on the decay model applicable to the incremental data of the current batch; and determine the total loss of the current batch based on the first loss of the classification model for the current batch relative to the incremental training data of the current batch, the second loss of the old classification model relative to the incremental data of the current batch, and the regularization coefficient.

[0306] It should be noted that the device provided in this application embodiment includes all the units included, which can be implemented by a processor in an electronic device; of course, it can also be implemented by specific logic circuits; in the implementation process, the processor can be a central processing unit (CPU), a microprocessor (MPU), a digital signal processor (DSP), or a field-programmable gate array (FPGA), etc.

[0307] The descriptions of the above device embodiments are similar to those of the above method embodiments, and have similar beneficial effects. For technical details not disclosed in the device embodiments of this application, please refer to the descriptions of the method embodiments of this application for understanding.

[0308] It should be noted that, in the embodiments of this application, if the above methods are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, or the parts that contribute to related technologies, can be embodied in the form of software products. These computer software products are stored in a storage medium and include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), magnetic disks, or optical disks. Thus, the embodiments of this application are not limited to any specific hardware and software combination.

[0309] Thirdly, embodiments of this application provide an electronic device including a memory and a processor. The memory stores computer programs or instructions, which, when executed by the processor, implement the method described in the first aspect.

[0310] In one example, reference Figure 16 As shown, the electronic device 160 includes: a processor 1601, at least one communication bus 1602, a user interface 1603, at least one external communication interface 1604, and a memory 1605. The communication bus 1602 is configured to enable communication between these components. The user interface 1603 may include a display screen, and the external communication interface 1604 may include standard wired and wireless interfaces.

[0311] The memory 1605 is configured to store instructions and applications executable by the processor 1601, and can also cache data to be processed or already processed by the processor 1601 and various modules in the electronic device (e.g., image data, audio data, voice communication data and video communication data), which can be implemented by flash memory or random access memory (RAM).

[0312] Fourthly, embodiments of this application provide a storage medium, namely a computer-readable storage medium, on which a computer program or instructions are stored, which, when executed by a processor, implement any of the methods provided in the first aspect of the above embodiments.

[0313] Fifthly, embodiments of this application provide a computer program product, which includes a computer program or instructions that, when executed by a processor, implement any of the methods provided in the first aspect of the above embodiments.

[0314] It should be noted that the descriptions of the above embodiments of storage media, devices, apparatuses, and program products are similar to the descriptions of the above method embodiments and have similar beneficial effects. For technical details not disclosed in the embodiments of storage media, devices, apparatuses, and program products of this application, please refer to the descriptions of the method embodiments of this application for understanding.

[0315] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment" or "in some embodiments" appearing throughout the specification do not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this application, the sequence numbers of the above-described processes do not imply a sequential 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 embodiments of this application. The sequence numbers of the above-described embodiments are merely descriptive and do not represent the superiority or inferiority of the embodiments.

[0316] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0317] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0318] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0319] In addition, each functional unit in the various embodiments of this application can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0320] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as mobile storage devices, read-only memory (ROM), magnetic disks, or optical disks.

[0321] Alternatively, if the integrated units described above are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, or the parts that contribute to related technologies, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0322] The above are merely embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A battery life estimation method, characterized by, The method comprises: obtaining battery data of a battery to be tested; the battery data is used to represent a mapping relationship between a cycle number of the battery to be tested and a state of health of the battery to be tested; determining, by a target classification model, a target degradation model to which the battery data is applicable among at least two degradation models; the target degradation model is a degradation model corresponding to a fitting curve type of the mapping relationship represented by the battery data; estimating a remaining life of the battery to be tested based on the target degradation model and the battery data.

2. The method of claim 1, wherein, The estimation of the remaining life of the battery to be tested based on the target degradation model and the battery data comprises: fitting the battery data by the target degradation model to obtain a target degradation curve; determining a state of health value of a life end of the battery to be tested; determining a cycle number of the life end of the battery to be tested based on the target degradation curve and the state of health value of the life end of the battery to be tested; determining the remaining life of the battery to be tested based on the cycle number of the life end of the battery to be tested and a current cycle number of the battery to be tested.

3. The method of claim 1, wherein, The obtaining of the battery data of the battery to be tested comprises: obtaining original data of the battery to be tested; the original data comprises a mapping relationship between a cycle number of the battery to be tested and a state of health of the battery to be tested; performing data cleaning and normalization processing on the original data of the battery to be tested respectively to obtain normalized data of the battery to be tested; splicing the normalized data of the battery to be tested and a current cycle number of the battery to be tested to obtain the battery data of the battery to be tested.

4. The method of claim 1, wherein, The method further comprises: determining a fitting degree of the battery data under the target degradation model based at least on a remaining life of the battery data; the fitting degree is used to represent a matching degree between the battery data and the target degradation model; if the fitting degree of the battery data under the target degradation model does not satisfy a condition, adjusting model parameters of the target degradation model until a degradation model to which the battery data is applicable determined by the target classification model after the adjustment is an adjusted degradation model; wherein the adjusted degradation model is different from the target degradation model.

5. The method of claim 4, wherein, The determination of the fitting degree of the battery data under the target degradation model based at least on the remaining life of the battery data comprises: determining a life label value of the battery data; determining a residual sum of squares of the battery data based on the life label value of the battery data and the remaining life of the battery data; determining a total sum of squares of the battery data based on the life label value of the battery data and the remaining life of the battery data; determining the fitting degree of the battery data under the target degradation model based on the residual sum of squares of the battery data and the total sum of squares of the battery data.

6. The method of claim 4, wherein, The method further comprises: in a case where the fitting degrees of the battery data under each of the at least two degradation models determined do not satisfy the condition, adjusting model parameters of the target classification model until an adjusted target classification model determines that the battery data is applicable to a degradation model corresponding to the maximum fitting degree among the at least two degradation models.

7. The method of claim 4, wherein, The method further comprises: if the fitting degree of the battery data determined by the target degradation model meets a condition, determining that the battery data is incremental data; incrementally training the target classification model based at least on the incremental data.

8. The method of claim 7, wherein, The incrementally training the target classification model based at least on the incremental data comprises: obtaining a plurality of incremental data for battery life estimation; each of the plurality of incremental data is battery data whose fitting degree determined by the target degradation model meets a condition; fusing the plurality of incremental data and a plurality of training data to divide the plurality of incremental data into a plurality of batches of incremental training data; determining, by the target classification model, a degradation model applicable to a current batch of incremental training data among the at least two degradation models; determining a total loss of the current batch based at least on the degradation model applicable to the current batch of incremental training data; adjusting model parameters of the target classification model based on the total loss of the current batch; and determining, based on the adjusted target classification model, a degradation model applicable to a next batch of incremental training data until the incrementally training is completed.

9. The method of claim 8, wherein, The determining the total loss of the current batch based at least on the degradation model applicable to the current batch of incremental training data comprises: dividing a plurality of incremental data for battery life estimation into a plurality of batches of incremental data; processing the current batch of incremental data by an old classification model to determine a degradation model applicable to the current batch of incremental data among the at least two degradation models; the old classification model is a target classification model corresponding to the incremental training data of a previous batch of the current batch; determining the total loss of the current batch based on the degradation model applicable to the current batch of incremental training data and the degradation model applicable to the current batch of incremental data.

10. The method of claim 9, wherein, The determining the total loss of the current batch based on the degradation model applicable to the current batch of incremental training data and the degradation model applicable to the current batch of incremental data comprises: determining a first loss of the classification model of the current batch with respect to the incremental training data of the current batch based on the degradation model applicable to the incremental training data of the current batch; determining a second loss of the old classification model with respect to the incremental data of the current batch based on the degradation model applicable to the incremental data of the current batch; determining the total loss of the current batch based on the first loss, the second loss, and a regularization coefficient.

11. A battery life estimation apparatus characterized by comprising: The device comprises: a first obtaining unit configured to obtain battery data of a battery to be tested; the battery data is used to represent a mapping relationship between a cycle number of the battery to be tested and a state of health of the battery to be tested; The first determination unit is configured to determine a target degradation model to which the battery data is applicable from at least two degradation models by using a target classification model; the target degradation model is a degradation model corresponding to a fitting curve type of the mapping relationship represented by the battery data. The first estimation unit is configured to estimate the remaining life of the battery under test based on the target degradation model and the battery data.

12. An electronic device, comprising: The electronic device includes a memory and a processor, and the memory stores a computer program, and the computer program is executed by the processor to implement the method in any one of claims 1-10.

13. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program or instructions, and the computer program or instructions are run by a computer to implement the method in any one of claims 1-10.

14. A computer program product, characterised in that, The computer program product includes a computer program or instructions, and the computer program or instructions are run by a computer to implement the method in any one of claims 1-10.

Images