Battery life prediction method and device, electronic equipment and storage medium

Abstract

The invention provides a battery life prediction method and device, electronic equipment and a storage medium, and relates to the cross technical field of machine learning and new energy batteries. The method comprises the steps of performing feature extraction on charge-discharge cycle data of a target battery to obtain an initial feature set; performing feature screening according to an evaluation value corresponding to each initial feature in the initial feature set to obtain a target feature combination, the evaluation value being used for representing the accuracy of a life prediction result; and determining a life prediction result corresponding to the target battery according to the target feature combination. According to the method, the optimal battery feature combination is determined according to the influence of the multiple battery features on the life prediction result for battery life prediction, and the accuracy of battery life prediction is improved.

Description

Technical Field

[0001] The present invention relates to the cross-technical field of machine learning and new energy batteries, and in particular to a battery life prediction method. Background Art

[0002] Lithium-ion batteries are complex dynamic electrochemical systems with complex aging mechanisms and nonlinear capacity decay processes. At present, data-driven methods for predicting battery life usually require the extraction of more feature data based on cycle data, electrochemical theory, and feature engineering methods to achieve life prediction. Currently, directly using more feature data to predict battery life does not yield high accuracy in life prediction results. Summary of the invention

[0003] The present invention provides a battery life prediction method, device, electronic device and storage medium to solve the defect of low battery life prediction accuracy in the prior art, and to achieve battery life prediction by determining the optimal battery feature combination based on the impact of multiple battery features on the life prediction results, thereby improving the accuracy of battery life prediction.

[0004] The present invention provides a battery life prediction method, comprising the following steps.

[0005] Extract features from the charge and discharge cycle data of the target battery to obtain an initial feature set; Performing feature screening according to the evaluation value corresponding to each initial feature in the initial feature set to obtain a target feature combination, wherein the evaluation value is used to characterize the accuracy of the life prediction result; According to the target feature combination, a life prediction result corresponding to the target battery is determined.

[0006] According to a battery life prediction method provided by the present invention, the feature screening is performed according to the evaluation value corresponding to each initial feature in the initial feature set to obtain a target feature combination, including: Acquire a life prediction model, wherein the life prediction model uses the initial features in the initial feature set as model input and uses the battery life prediction result as model output; Performing model training on the life prediction model according to the initial features to obtain a model prediction error corresponding to the initial features, wherein the model prediction error is the evaluation value; The target feature combination is determined according to the model prediction errors corresponding to the initial features.

[0007] According to a battery life prediction method provided by the present invention, the life prediction model is trained according to the initial features to obtain the model prediction error corresponding to the initial features, including: Determine each evaluation feature set corresponding to each initial feature, wherein each evaluation feature set corresponding to each initial feature is a feature set obtained by gradually deleting each initial feature or gradually adding each initial feature; The life prediction model is trained by each evaluation feature set to obtain a life prediction result, and then the model prediction error corresponding to each evaluation feature set is calculated according to the life prediction result, where the model prediction error corresponding to each evaluation feature set is the model prediction error corresponding to each initial feature; Determining the target feature combination according to the model prediction errors corresponding to the initial features includes: Determining a target feature combination according to the model prediction errors corresponding to each evaluation feature set; Determining the life prediction result corresponding to the target battery according to the target feature combination includes: The model prediction result corresponding to the target feature combination is determined as the life prediction result corresponding to the target battery.

[0008] According to a battery life prediction method provided by the present invention, the step of determining each evaluation feature set corresponding to each initial feature includes: Determining a first feature set, wherein the first feature set includes all initial features in the initial feature set; In each iteration, a first key feature in the first feature set is determined and deleted to obtain an evaluation feature set corresponding to the first key feature, and then the evaluation feature set corresponding to the first key feature is determined as the first feature set for the next iteration, until there is only one initial feature left in the first feature set, wherein the first key feature is an initial feature with the smallest model prediction error corresponding to the feature set obtained after the deletion determined by a cross-validation method; After several iterations, each evaluation feature set corresponding to each initial feature is obtained.

[0009] According to a battery life prediction method provided by the present invention, the step of determining each evaluation feature set corresponding to each initial feature includes: Determine a second feature set and a third feature set, wherein the second feature set includes all initial features in the initial feature set, and the third feature set is an empty set; In each iteration, a second key feature in the second feature set is determined and the second key feature is added to the third feature set and the second key feature is deleted from the second feature set, the added third feature set is the evaluation feature set corresponding to the second key feature, the added third feature set is determined to be the third feature set and the deleted second feature set is determined to be the second feature set for the next iteration until the second feature set is an empty set, wherein the second key feature is an initial feature with the smallest model prediction error corresponding to the feature set obtained after the addition determined by a cross-validation method; After several iterations, each evaluation feature set corresponding to each initial feature is obtained.

[0010] According to a battery life prediction method provided by the present invention, determining a target feature combination according to the model prediction error corresponding to each evaluation feature set includes: Determining a target model prediction error according to the model prediction errors of each evaluation feature set, wherein the target model prediction error is the minimum model prediction error; Determine the evaluation feature set corresponding to the target model prediction error as the target feature combination.

[0011] According to a battery life prediction method provided by the present invention, the feature extraction of the charge and discharge cycle data of the target battery to obtain an initial feature set includes: Extracting features according to the cycle curve differences, capacity increment curves and differential voltage curves between different cycles in the charge and discharge cycle data to obtain battery cycle features, wherein the battery cycle features include at least one of capacity difference, capacity increment difference and differential voltage; Extracting statistical features from the charge-discharge cycle data to obtain battery statistical features; Performing correlation analysis on the battery cycle characteristics and the battery statistical characteristics to obtain battery correlation characteristics; Statistical characteristic calculation is performed on the battery cycle feature, the battery statistical feature and the battery correlation feature to obtain a plurality of initial features, and the plurality of initial features constitute the initial feature set.

[0012] According to a battery life prediction method provided by the present invention, the obtaining of a life prediction model comprises: Determine a Gaussian process regression model, wherein the implicit function of the Gaussian process regression model obeys the Gaussian distribution of the mean function and the covariance function, and the Gaussian process regression model is the life prediction model; Determine the mean function and covariance function corresponding to the Gaussian process regression model; According to the mean function and the covariance function, hyperparameters corresponding to the Gaussian process regression model are determined.

[0013] According to a battery life prediction method provided by the present invention, the determining of the mean function and the covariance function corresponding to the Gaussian process regression model includes: Determining the mean function to be a fixed basis function or a linear function; The covariance function is determined to be a squared exponential kernel function or a Mattern covariance function.

[0014] According to a battery life prediction method provided by the present invention, determining the hyperparameters corresponding to the Gaussian process regression model according to the mean function and the covariance function includes: The negative log marginal likelihood function is minimized for the mean function and the covariance function by using the conjugate gradient descent method to obtain a target hyperparameter, where the target hyperparameter is a parameter that optimizes the performance of the Gaussian process regression model.

[0015] According to a battery life prediction method provided by the present invention, the conjugate gradient descent method is used to minimize the negative logarithmic marginal likelihood function of the mean function and the covariance function to obtain the target hyperparameter, including: The mean function includes a zero mean function, the covariance function includes a square exponential kernel function, the negative log marginal likelihood function is determined according to the zero mean function and the square exponential kernel function, and the negative log marginal likelihood function includes an initial hyperparameter; The negative log marginal likelihood function is solved using a gradient descent algorithm to obtain the target hyperparameter, wherein the negative log marginal likelihood function is solved using a gradient descent algorithm including calculating a partial derivative of the negative log marginal likelihood function with respect to the initial hyperparameter.

[0016] The present invention also provides a battery life prediction device, comprising the following modules: A feature acquisition module is used to extract features from the charge and discharge cycle data of the target battery to obtain an initial feature set; Feature screening module: performing feature screening according to the evaluation value corresponding to each initial feature in the initial feature set to obtain a target feature combination, wherein the evaluation value is used to characterize the accuracy of the life prediction result; The result acquisition module is used to determine the life prediction result corresponding to the target battery according to the target feature combination.

[0017] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, any of the battery life prediction methods described above is implemented.

[0018] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the battery life prediction method as described in any one of the above is implemented.

[0019] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the battery life prediction method as described above is implemented.

[0020] The battery life prediction method, device, electronic device and storage medium provided by the present invention determine the optimal battery feature combination for battery life prediction based on the impact of multiple battery features on the life prediction results, thereby improving the accuracy of battery life prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0022] Figure 1 It is a flow chart of the battery life prediction method provided by the present invention.

[0023] Figure 2 It is a flow chart of the method for obtaining target feature combination provided by the present invention.

[0024] Figure 3 It is a flowchart of the method for obtaining an initial feature set provided by the present invention.

[0025] Figure 4 It is a schematic diagram of cycle data analysis of batteries at different cycle numbers provided by the present invention.

[0026] Figure 5 It is a schematic diagram of the analysis of some battery characteristics provided by the present invention and their correlation with battery life.

[0027] Figure 6 It is the initial feature set provided by the present invention and the correlation coefficient between each feature and battery life.

[0028] Figure 7 It is a flow chart of the method for obtaining a life prediction model provided by the present invention.

[0029] Figure 8 It is a schematic diagram of the framework of the battery life prediction method provided by the present invention.

[0030] Fig. 9It is a comparison diagram and an error diagram between the battery life prediction result provided by the present invention and the measured actual life.

[0031] Fig.10 It is a flow chart of the improved Gaussian process regression method with adaptive feature selection proposed by the present invention.

[0032] Fig.11 It is a structural schematic diagram of the battery life prediction device provided by the present invention.

[0033] Fig.12 It is a schematic diagram of the physical structure of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0035] As my country's energy structure is transforming towards clean, low-carbon and electrified, more and more application scenarios, including electric vehicles and energy storage systems, require the participation of high-energy-density lithium-ion batteries. However, batteries will inevitably age during use, accompanied by battery capacity decay and resistance increase, and the remaining service life will decrease. Accurate battery life prediction is particularly important for ensuring the reliability and safety of in-service batteries, evaluating the battery use value, and accelerating the battery development cycle.

[0036] Lithium-ion batteries are complex dynamic electrochemical systems with complex aging mechanisms and nonlinear capacity decay processes. This poses a severe challenge to accurate life prediction based on early data. At present, data-driven battery life prediction usually extracts more feature data based on feature engineering methods and battery cycle data to achieve life prediction. However, directly using more feature data for battery life prediction results in low prediction accuracy for life prediction. From these extracted features, determining the appropriate feature combination to maximize the utilization of feature information plays an important role in improving the prediction accuracy of battery life. However, there is currently no clear conclusion or method to determine the optimal feature combination.

[0037] In view of this, an embodiment of the present invention provides a battery life prediction method with adaptive feature selection, which extracts features from the charge and discharge cycle data of the target battery to obtain an initial feature set, performs feature screening based on the evaluation values ​​corresponding to each initial feature in the initial feature set, and obtains a target feature combination, wherein the evaluation value is used to characterize the accuracy of the life prediction result, and determines the life prediction result corresponding to the target battery based on the target feature combination. This method can select the optimal battery feature set for battery life prediction based on the impact of multiple battery features on the life prediction result, thereby improving the accuracy of battery life prediction.

[0038] The technical solutions in the embodiments of the present invention will be described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0039] Figure 1 : is a flow chart of the battery life prediction method provided by the present invention. The battery life prediction method can be applied to electronic devices, which can be various types of devices with information processing capabilities during implementation. For example, the electronic device can include a personal computer, a laptop, a PDA or a server, etc.; the electronic device can also be a mobile terminal, for example, the mobile terminal can include a mobile phone, a car computer, a tablet computer or a projector, etc. Figure 1 As shown, the method may include the following steps 101 to 103: Step 101: extracting features from the charge and discharge cycle data of the target battery to obtain an initial feature set.

[0040] It should be noted that the method for extracting features from the charge and discharge cycle data of the target battery can adopt feature engineering methods, statistical methods, etc. based on electrochemical theory. The present invention does not limit the method for extracting features from the charge and discharge cycle data of the target battery to obtain an initial feature set.

[0041] Step 102: Perform feature screening according to the evaluation value corresponding to each initial feature in the initial feature set to obtain a target feature combination, wherein the evaluation value is used to characterize the accuracy of the life prediction result.

[0042] It should be noted that the present invention can perform feature screening according to the evaluation value corresponding to each initial feature, that is, screen the features according to the accuracy of the life prediction results of each initial feature to obtain a target feature combination, which includes multiple target features, and the multiple target features are part of or all of the features screened from multiple initial features.

[0043] Among them, there are many methods for obtaining a target feature combination by performing feature screening according to the evaluation value corresponding to each initial feature in the initial feature set. For example, each initial feature can be evaluated by a feature evaluation method, or a feature verification method can be used. The present invention does not limit the method for obtaining a target feature combination by performing feature screening according to the evaluation value corresponding to each initial feature in the initial feature set.

[0044] Step 103: Determine the life prediction result corresponding to the target battery according to the target feature combination.

[0045] It should be noted that there are many ways to determine the life prediction result corresponding to the target battery based on the target feature combination, such as through algorithm calculation, model generation, etc. The present invention does not limit the method for determining the life prediction result corresponding to the target battery based on the target feature combination.

[0046] Existing data-driven battery life prediction methods extract a large number of features to predict the life, resulting in low accuracy in battery life prediction. Therefore, there is currently no extensive research or definite conclusion on how to select the optimal feature combination to achieve the highest accuracy in predicting the remaining battery life. The method provided by the present invention selects features based on the accuracy of the life prediction results according to each initial feature, and can select the optimal feature combination based on only a small amount of early battery cycle data to achieve the highest possible accuracy in early prediction of battery life.

[0047] Figure 2 FIG. 1 is a flow chart of the method for obtaining target feature combination provided by the present invention. Figure 2 As shown, the feature screening according to the evaluation value corresponding to each initial feature in the initial feature set to obtain the target feature combination may include: Step 201: Acquire a life prediction model, wherein the life prediction model uses initial features in the initial feature set as model input and uses battery life prediction results as model output.

[0048] It should be noted that the type of the life prediction model can be a linear regression model, a polynomial regression model, an exponential decay model, etc. The present invention does not limit the type of the life prediction model.

[0049] Step 202: Perform model training on the life prediction model according to the initial features to obtain the model prediction error corresponding to the initial features, where the model prediction error is the evaluation value.

[0050] It should be noted that the method of training the battery life prediction model according to the initial features can be to combine the initial features into various feature combinations and then train the model, etc. The present invention does not limit the method of training the life prediction model according to the initial features.

[0051] Among them, the method for calculating the model prediction error according to the life prediction results can be to calculate the mean absolute error, mean absolute percentage error, mean square error, etc. according to the model prediction results. The present invention does not limit the method for calculating the model prediction error.

[0052] Step 203: Determine a target feature combination according to the model prediction errors corresponding to the initial features.

[0053] It should be noted that the method for determining the target feature combination based on the model prediction errors corresponding to the initial features may be to select the initial features corresponding to the minimum model prediction error, or to select the initial features that meet the error requirements, etc. The present invention does not limit the method for determining the target feature combination based on the model prediction errors corresponding to the initial features.

[0054] It can be understood that the method provided by the present invention obtains a life prediction model, calculates the life prediction error in the process of training the life prediction model according to various initial features, and thus screens out the optimal feature combination, which can improve the accuracy of the life prediction results.

[0055] In some embodiments, model training is performed on the life prediction model according to the initial features to obtain model prediction errors corresponding to the initial features, which may include: determining each evaluation feature set corresponding to the initial features, where each evaluation feature set corresponding to the initial features is a feature set obtained by gradually deleting the initial features or gradually adding the initial features; model training is performed on the life prediction model according to the evaluation feature sets to obtain a life prediction result, and then the model prediction errors corresponding to the evaluation feature sets are calculated according to the life prediction result, where the model prediction errors corresponding to the evaluation feature sets are the model prediction errors corresponding to the initial features.

[0056] It should be noted that the model prediction error corresponding to each initial feature can be obtained based on the evaluation feature set corresponding to each initial feature. For example, the model prediction error of the feature combination A including the initial feature A is obtained, and then the model prediction error of the feature combination A with the initial feature A deleted is obtained. In this way, the influence of A on the accuracy of the model prediction result can be obtained. The model prediction error of the feature combination with the initial feature A deleted can be the model prediction error corresponding to A.

[0057] Further, determining the target feature combination according to the model prediction errors corresponding to the initial features may include: determining the target feature combination according to the model prediction errors corresponding to the evaluation feature sets. Determining the life prediction result corresponding to the target battery according to the target feature combination may include: determining the model prediction result corresponding to the target feature combination as the life prediction result corresponding to the target battery.

[0058] It can be understood that determining each evaluation feature set corresponding to each initial feature, then training the life prediction model through each evaluation feature set to obtain a life prediction result, and then calculating the model prediction error corresponding to each evaluation feature set based on the life prediction result to determine the model prediction error corresponding to each initial feature can improve the accuracy of the life prediction result.

[0059] In some embodiments, determining the evaluation feature sets corresponding to the initial features may include: determining a first feature set, the first feature set including all the initial features in the initial feature set; in each iteration, determining the first key feature in the first feature set and deleting it to obtain the evaluation feature set corresponding to the first key feature, and then determining the evaluation feature set corresponding to the first key feature as the first feature set for the next iteration until only one initial feature remains in the first feature set, wherein the first key feature is the initial feature with the smallest model prediction error corresponding to the feature set obtained after deletion determined by a cross-validation method; after several iterations, obtaining the evaluation feature sets corresponding to the initial features.

[0060] It should be noted that the present invention designs an adaptive feature selection algorithm based on the sequential backward method. From the feature pool of the initial feature set, according to the evaluation results of each feature combination on life prediction, the features that affect the model prediction performance are gradually deleted. After several iterations, the combination that remains is the feature combination that makes the model prediction result most accurate.

[0061] Exemplarily, for a specific iteration cycle, based on the Cross-Validation (CV) method, the average prediction error is calculated after deleting a feature in each iteration. After traversing all features, the life prediction errors are recorded and compared. Then the feature with the smallest model prediction error after deletion is deleted. After keeping the feature deleted, the next iteration cycle is started in the same way to delete another feature. And so on. Obviously, one feature will be removed from the feature pool after each cycle. In this way, different feature combinations will be formed by deleting features in turn in each cycle. Assuming that the number of features in the feature pool is N, after (N-1) cycles, (N-1) feature combinations will be obtained according to the number of features removed. Finally, the optimal feature combination that minimizes the model error is selected from the (N-1) feature combinations. The specific algorithm process is shown in the algorithm flow shown below.

[0062] Table 1 Adaptive feature selection algorithm for battery life prediction Specifically, the present invention provides an adaptive feature selection method, which gradually deletes a feature through a sequential back-off algorithm so that the model life prediction error based on the remaining features after removing the feature is minimized. After traversing all the features in the feature pool, the optimal feature combination can be adaptively selected from the feature pool, combined with the cross-validation method, so that the error of early battery life prediction is minimized and has good generalization ability. In this way, the problem that the prior art cannot determine the optimal feature combination can be solved, thereby improving the accuracy of early battery life prediction.

[0063] Among them, the adaptive feature selection algorithm deletes features from the feature pool in sequence based on the minimum model prediction average error. Moreover, in addition to the mapping relationship between the generated features and the battery life, the algorithm also considers the relationship between these features. More importantly, through cross-validation and the adaptive feature selection algorithm, some repeated and redundant feature information can be effectively deleted to avoid the problems of overfitting and poor generalization that may be caused by information redundancy. Therefore, when the initially generated feature pool contains enough information, the adaptive feature selection algorithm can obtain the optimal feature combination, so that the model can achieve the best prediction performance and the most accurate battery life prediction accuracy.

[0064] In some embodiments, determining each evaluation feature set corresponding to each initial feature may include: determining a second feature set and a third feature set, wherein the second feature set includes all initial features in the initial feature set, and the third feature set is an empty set; in each iteration, determining the second key feature in the second feature set and adding the second key feature to the third feature set and deleting the second key feature from the second feature set, the added third feature set being the evaluation feature set corresponding to the second key feature, determining the added third feature set as the third feature set and determining the deleted second feature set as the second feature set for the next iteration until the second feature set is an empty set, wherein the second key feature is an initial feature with the smallest model prediction error corresponding to the feature set obtained after the addition, which is determined by the cross-validation method; after several iterations, the evaluation feature sets corresponding to each initial feature are obtained.

[0065] It should be noted that the present invention designs an adaptive feature selection algorithm based on the sequential forward method. From the feature pool of the second feature set, according to the evaluation results of each feature combination on life prediction, the features that positively affect the model prediction performance are gradually added to the third feature set. After several iterations, the combination that remains is the feature combination that makes the model prediction result most accurate.

[0066] Specifically, for a specific iteration cycle, based on the cross-validation method, one feature from the second feature set is selected in each iteration to be added to the third feature set, and then the average prediction error of the third feature set is calculated. After traversing all features, the life prediction errors are recorded and compared. Then the feature with the smallest model prediction error after the addition is deleted from the second feature set. After keeping the feature added, the next iteration cycle is started in the same way to add another feature. And so on. Obviously, a feature will be added to the third feature set after each cycle. In this way, different feature combinations will be formed by adding features in each cycle. Assuming that the number of features in the feature pool is N, after (N-1) cycles, (N-1) feature combinations will be obtained according to the number of features added. Finally, the optimal feature combination that minimizes the model error is selected from the (N-1) feature combinations.

[0067] It can be understood that when the initially generated feature pool contains sufficient information, the adaptive feature selection algorithm can obtain the optimal feature combination, so that the model can achieve the best prediction performance and the most accurate battery life prediction accuracy.

[0068] Furthermore, determining the target feature combination based on the model prediction errors corresponding to the evaluation feature sets may include: determining a target model prediction error based on the model prediction errors of the evaluation feature sets, the target model prediction error being the minimum model prediction error; and determining the evaluation feature set corresponding to the target model prediction error as the target feature combination.

[0069] It can be understood that determining the evaluation feature set with the smallest model prediction error as the target feature combination represents the selection of the corresponding evaluation feature combination with the highest prediction accuracy, which can enable the model to achieve the optimal prediction performance and the most accurate battery life prediction accuracy.

[0070] Figure 3 FIG. 1 is a flow chart of the method for obtaining an initial feature set provided by the present invention. Figure 3 As shown, the feature extraction of the charge and discharge cycle data of the target battery to obtain the initial feature set may include: Step 301: extracting features according to the cycle curve differences, capacity increment curve differences and differential voltage curve differences between different cycles in the charge and discharge cycle data to obtain battery cycle features, wherein the battery cycle features include at least one of capacity difference, capacity increment difference and differential voltage; Step 302: extracting statistical features from the charge and discharge cycle data to obtain battery statistical features; Step 303: performing correlation analysis on the battery cycle characteristics and the battery statistical characteristics to obtain battery correlation characteristics; Step 304: Calculate the statistical characteristics of the battery cycle characteristics, the battery statistical characteristics and the battery correlation characteristics to obtain a plurality of initial characteristics, and the plurality of initial characteristics constitute the initial characteristic set.

[0071] Figure 4 Schematic diagram of the cycle data analysis of the battery under different cycle times provided by the present invention. Figure 4 As shown, firstly, the differences in cycle curves, capacity increment curves and differential voltage curves between different cycles of the battery are analyzed, as shown in FIG. Figure 4 As shown. Among them, Figure 4 (a) shows the discharge capacity-voltage curve of the battery during discharge at different cycle numbers. It can be seen that as the battery cycle number increases, the battery discharge capacity decreases at the same voltage. Figure 4 (b) is the battery voltage-discharge capacity increment (IC) curve. As the number of cycles increases, the absolute value of its extreme value decreases. Figure 4 (c) is the battery capacity-differential voltage (DV) curve, the curvature radius of the curve decreases with the increase of cycle number.

[0072] Based on the above battery cycle data analysis, we extract features such as capacity difference (∆Q) and capacity increment difference (∆dQ / dV), combine the corresponding statistical features, and conduct a preliminary correlation analysis between each feature and battery life based on the Pearson product moment Pearson correlation coefficient. The expression of the Pearson correlation coefficient is as follows: ,in, and They are the characteristics and corresponding battery life. , and , are the mean and standard deviation of the two respectively, is the covariance function, is the expected function. The closer the absolute value of the correlation coefficient is to 1, the higher the correlation between the two data.

[0073] Figure 5 FIG. 1 is a schematic diagram of the correlation analysis between some battery characteristics and battery life provided by the present invention. Figure 5 As shown, the voltage-capacity difference curve between the 100th cycle and the 10th cycle of battery discharge extracted from batteries with different lifespans is as follows Figure 5 As shown in (a), the lighter colored ones are longer-life batteries, and the darker colored ones are shorter-life batteries. Obviously, as the battery life increases, the battery capacity difference between the 100th cycle and the 10th cycle decreases. In order to further demonstrate the correlation between capacity difference and battery life, the minimum capacity difference of each battery (i.e. Figure 5 (a) The scatter plot of the valley value of the curve) and the logarithmic battery life value, such as Figure 5 (b) It is obvious that there is a clear linear relationship between the two. Figure 5 (c) and Figure 5 (d) Shows the correlation between the capacity increment difference between the 100th cycle and the 10th cycle and the battery life. Figure 5 (e) and Figure 5 (f) shows the correlation between the capacity difference corresponding to the differential voltage between the 100th cycle and the 10th cycle and the battery life.

[0074] Figure 6 is the initial feature set provided by the present invention and the correlation coefficient between each feature and battery life. Figure 6 As shown, according to the adaptive feature extraction algorithm proposed in the present invention, according to the above feature extraction method and the steps described in Table 1, based on Figure 6The initial features in the feature pool shown in the figure are used for model training and adaptive feature selection. When the adaptively extracted feature combination is obtained as listed in Table 2, the model has the smallest life prediction error for the validation set. As can be seen from Table 2, the selected optimal feature combination has some features with low correlation with battery life in addition to those with high absolute values ​​of correlation coefficients. Moreover, it is not Figure 6 All features with high absolute values ​​of correlation coefficients are included in the optimal feature combination selected in Table 2.

[0075] Based on the above three features, the statistical characteristics are calculated respectively, including the minimum value, mean, variance, skewness, peak state, median, difference between the upper quartile and the lower quartile, and then combined with the slope and intercept of the capacity difference and battery life fitting curve, the resistance difference between the two cycles of the battery, the temperature difference, the charging time difference, etc., to form the corresponding full feature pool, and calculate the absolute value of the correlation coefficient between each feature in the feature pool and the logarithm of the battery life, such as Figure 6 shown.

[0076] Table 2 Optimal feature combinations for adaptive feature selection This is because the adaptive feature combination algorithm proposed in the present invention can effectively identify and remove repeated and redundant information in the feature pool, and obtain the most complete information in the feature pool as much as possible. For example, although the average value of the battery charging time has a low correlation with the battery life, it reflects the battery charging strategy and battery aging information to a certain extent. When the charging rate is fast and the battery aging degree is high, the battery charging time is shorter. Therefore, it can be seen that the adaptive feature selection method proposed in the present invention can effectively select as complete useful information as possible from the feature pool, and remove redundant features to ensure accurate prediction of the early battery life.

[0077] Figure 7 FIG. 1 is a flow chart of the method for obtaining a life prediction model provided by the present invention. Figure 7 As shown, the acquisition of the life prediction model may include: Step 401: Determine a Gaussian process regression model, the implicit function of which obeys the Gaussian distribution of the mean function and the covariance function, and the Gaussian process regression model is the life prediction model.

[0078] It should be noted that the initial features in the initial feature set can be used as model input , the battery life prediction result is used as the model output The Gaussian process regression model can be expressed as: , where the implicit function The mean is , the covariance is Gaussian distribution, that is: , , .

[0079] When the mean function is 0, the prior distribution of the model output is expressed as: ,in, represents the noise covariance matrix, is a symmetric positive definite matrix, The simplified notation for Represents two variables and The larger the value, the higher the similarity between the two.

[0080] There are many choices for covariance functions, the most common one is the squared exponential (SE) covariance function (also known as radial basis function RBF), as shown below: ,in, Represents the signal variance, which is used to adjust the value of the covariance function. express and The distance between. The characteristic length is A significant advantage of the square exponential kernel is that it can be infinitely differentiated and is always continuous.

[0081] Step 402: Determine the mean function and covariance function corresponding to the Gaussian process regression model.

[0082] Among them, determining the mean function and covariance function corresponding to the Gaussian process regression model may include: determining the mean function to be a fixed basis function or a linear function; determining the covariance function to be a square exponential kernel function or a Mattern covariance function.

[0083] It should be noted that in actual application scenarios, the zero mean function mentioned above may not be applicable. Moreover, it is difficult to determine a suitable mean function for a given prior function. One solution is to determine a fixed basis function, such as a linear mean function. When the mean function is linear, Gaussian process regression can be expressed as: ,in, is a Gaussian process regression model with a zero mean function. is a series of fixed basis functions. Optimization parameters Normal distribution . Then the Gaussian process is expressed as: , Afterwards, the mean and variance of the battery life predicted by the model are expressed as: , , in, All the training sets The collection of All the training sets A collection of . , ,in, , .

[0084] For the covariance function, in addition to the above-mentioned squared exponential (SE) kernel, the Matern covariance function is another commonly used kernel function for Gaussian process regression, which can be expressed as: ,in, is the modified Bessel function. is the distance scale parameter. is a smoothing hyperparameter, The larger the value, the smoother the function. and The cases are represented as Matern5 and Matern3 respectively.

[0085] Step 403: Determine the hyperparameters corresponding to the Gaussian process regression model according to the mean function and the covariance function.

[0086] It should be noted that determining the hyperparameters corresponding to the Gaussian process regression model based on the mean function and the covariance function may include: performing minimization on the negative log marginal likelihood function of the mean function and the covariance function by the conjugate gradient descent method to obtain target hyperparameters, wherein the target hyperparameters are parameters that optimize the performance of the Gaussian process regression model.

[0087] It should be noted that the optimal hyperparameters of the model are obtained by minimizing the negative log marginal likelihood (NLML) using the conjugate gradient descent method. The model output is calculated based on the obtained optimal hyperparameters to predict the battery life.

[0088] Furthermore, the method of minimizing the negative log marginal likelihood function of the mean function and the covariance function by the conjugate gradient descent method to obtain the target hyperparameters may include: the mean function includes a zero mean function, the covariance function includes a square exponential kernel function, the negative log marginal likelihood function is determined according to the zero mean function and the square exponential kernel function, and the negative log marginal likelihood function includes initial hyperparameters; the negative log marginal likelihood function is solved by a gradient descent algorithm to obtain the target hyperparameters, and the method of solving the negative log marginal likelihood function by a gradient descent algorithm includes calculating the partial derivatives of the negative log marginal likelihood function with respect to the initial hyperparameters.

[0089] For example, the above mean function and covariance function both contain a series of hyperparameters: In order to make the model learn the optimal hyperparameters during training, the maximum likelihood method is used to optimize the hyperparameters. For the convenience of calculation, the optimal hyperparameters are obtained by minimizing the negative logarithmic marginal likelihood (NLML) function, that is: ,in, is the negative log marginal likelihood function as shown below. and are the hyperparameters and optimal hyperparameters respectively.

[0090] Taking the zero mean function and SE kernel as an example, its negative log marginal likelihood function is expressed as: . Use the gradient descent (CG) algorithm to solve the above formula to obtain the optimal hyperparameters. To do this, it is necessary to calculate the partial derivative of the marginal likelihood function with respect to the hyperparameters as follows: ,in, , Represents the trace of A. After determining the optimal hyperparameters, the Gaussian process regression model is used for prediction. For new input , and its model output is expressed as , observation results and The joint distribution between is expressed as: ,in, , , Then, we can get the mean of the model prediction results and variance As shown below: , Based on the mean and variance of the prediction results, a 95% confidence interval was calculated to assess the uncertainty of the model prediction results.

[0091] The improved Gaussian process regression method with adaptive feature selection proposed in the present invention combines the sequential backward feature selection algorithm with the Gaussian process regression algorithm, performs adaptive optimal feature combination selection based on the feature pool generated by battery cycle data analysis, and predicts the early life of the battery based on the selected optimal feature combination.

[0092] Among them, the accuracy of battery life prediction is improved through feature engineering, Gaussian process regression and adaptive feature selection algorithm. That is, the method proposed in the present invention can achieve accurate early prediction of battery life based on the optimal feature combination.

[0093] The following describes an exemplary application of an embodiment of the present invention in a practical application scenario.

[0094] The present invention extracts a number of battery features to form a feature pool based on the early charge and discharge cycle data of the battery based on electrochemical theory, feature engineering methods and statistical methods, and automatically obtains the optimal feature combination based on the improved Gaussian process regression method combined with adaptive feature selection, thereby achieving accurate early prediction of battery life.

[0095] Figure 8 Schematic diagram of the battery life prediction method provided by the present invention. Figure 8 As shown, the battery life prediction method includes the following steps.

[0096] (1) Based on electrochemical theory, the battery charge and discharge cycle data is subjected to capacity increment analysis (dQ / dV, ICA) and differential voltage analysis (dV / dQ, DVA). Combined with the capacity difference between different cycles, charging time and temperature, several characteristic indicators are calculated using statistical analysis methods to generate the corresponding characteristic pool.

[0097] Specifically, according to the capacity difference-voltage curve between the 100th cycle and the 10th cycle ( )、Capacity increment difference-voltage curve( ) and capacity difference-differential voltage curve ( ). Analyze the statistical characteristics of these features such as minimum value, mean, variance, kurtosis, skewness, median, difference between upper quartile and lower quartile, etc. to form a feature pool containing a series of battery features.

[0098] In this embodiment, based on part of the charge and discharge data in the MIT-Stanford public data set, the improved Gaussian process regression method with adaptive feature selection is proposed to perform early prediction of battery life.

[0099] Implementation data: This data is publicly available on the Experimental Data Platform (matr.io). The data includes charge and discharge cycle data of 124 lithium iron phosphate (LFP) 18650 batteries with a rated capacity of 1.1Ah. These batteries are rapidly charged at different rates in the 0%-80% SOC range. In the 80%-100% SOC range, the battery is in a constant voltage charging stage after reaching the maximum cut-off voltage until the current drops below the charge cut-off current. During discharge, all batteries are discharged at a rate of 4C. The battery cycle depth is 0%-100%, and the cycle continues until the battery capacity decays to 80% of the rated capacity. The number of cycles from the beginning to 80% of the rated capacity is defined as the battery life.

[0100] (2): Establish a Gaussian process regression (GPR) model based on the corresponding mean function and covariance function.

[0101] (3): Different mean functions and different covariance functions choose.

[0102] (4): The optimal hyperparameters of the model are obtained by minimizing the negative log marginal likelihood (NLML) using the conjugate gradient descent method. The model output is calculated based on the obtained optimal hyperparameters to predict the battery life.

[0103] (5): Design an adaptive feature selection algorithm based on the sequential backward method. From the feature pool in step 501, based on the evaluation results of each feature combination on life prediction, gradually delete the features that affect the model prediction performance. After several iterations, the combination that remains is the feature combination that makes the model prediction result most accurate.

[0104] Fig. 9 The figure is a comparison chart and error chart between the battery life prediction results and the measured actual life provided by the present invention. In general, the battery life results predicted by this method are consistent with the actual life results measured, and most of the actual life results are within the 95% confidence interval ( Fig. 9 The gray shaded area of ​​(a)). The confidence intervals of life prediction for both the training set and the test set are in a small range, indicating that the variance of the prediction results is small.

[0105] To more clearly show the difference between the battery life prediction and the measured values, we plot Fig. 9 (b) is a scatter plot. The black solid line with a slope of 1 in the figure indicates the ideal state where the predicted result is equal to the measured value. The closer the scatter distribution is to the solid line, the better the predicted result. Fig. 9As can be seen in (b), the scattered points representing the training set and the test set are all distributed near the ideal solid line and are relatively close to the solid line, which indicates that the method of the invention can achieve early prediction of battery life with high accuracy.

[0106] Fig. 9 (c) is the distribution of the residual error of life prediction obtained by the invention method. It can be seen that the residual error of life prediction of most batteries is within the range of ±50 cycles. The residual error of life prediction of all batteries is within the range of [-250,150] cycles. The overall distribution curve of relative error (dark color) and residual (light color) of battery life prediction results of training set and test set is shown in Fig. 9 (d) As shown. It can be seen that the relative errors of battery life prediction for both data sets are mainly distributed within the range of ±10%. However, the violin plot of the test set is narrower than that of the training set, indicating that the error distribution range of the test set is wider, while the error of the training set is more concentrated in the range of ±5%. The distribution of the predicted life residual is also similar to the distribution of the relative error.

[0107] Fig.10 Flowchart of the improved Gaussian process regression method with adaptive feature selection proposed by the present invention. Fig.10 As shown, the present invention generates a feature pool through battery data analysis and feature extraction, obtains the optimal model parameters by selecting the mean and covariance functions and minimizing the negative log marginal likelihood, performs adaptive feature selection based on the initial features and models in the feature pool, and the adaptive feature selection is obtained through the training process of the Gaussian process regression model, and performs battery early life prediction based on the selected optimal feature combination.

[0108] Compared with the prior art, the present invention has a wider range of applications. On the one hand, the present invention provides a general method that can adaptively select the optimal feature combination from a feature pool containing a series of features to minimize the error in early prediction of battery life. The inventive method solves the problem of how to determine the optimal feature combination. On the other hand, the present invention can achieve accurate life prediction in the early stage of battery use, when the capacity has not yet begun to decay rapidly, to ensure the safety and reliability of the battery in service.

[0109] Based on the foregoing embodiments, an embodiment of the present invention provides a battery life prediction device, and each module included in the device, and each unit included in each module, can be implemented by a processor; of course, it can also be implemented by a specific logic circuit; 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.

[0110] The battery life prediction device provided by the present invention is described below. The battery life prediction device described below and the battery life prediction method described above can be referenced to each other.

[0111] Fig.11 Schematic diagram of the structure of the battery life prediction device provided by the present invention. Fig.11 As shown, the device 500 includes a feature acquisition module 501, a feature screening module 502 and a result acquisition module 503, wherein: A feature acquisition module 501 is used to extract features from the charge and discharge cycle data of the target battery to obtain an initial feature set; Feature screening module 502: performing feature screening according to the evaluation value corresponding to each initial feature in the initial feature set to obtain a target feature combination, wherein the evaluation value is used to characterize the accuracy of the life prediction result; The result acquisition module 503 is used to determine the life prediction result corresponding to the target battery according to the target feature combination.

[0112] In some embodiments, the feature screening module 502 includes a model acquisition unit, an error acquisition unit and a combination determination unit, wherein: A model acquisition unit, used to acquire a life prediction model, wherein the life prediction model uses the initial features in the initial feature set as model input and uses the battery life prediction result as model output; an error acquisition unit, configured to perform model training on the life prediction model according to the initial features to obtain a model prediction error corresponding to the initial features, wherein the model prediction error is the evaluation value; The combination determination unit is used to determine the target feature combination according to the model prediction errors corresponding to the initial features.

[0113] In some embodiments, the error acquisition unit is specifically used to: determine each evaluation feature set corresponding to each initial feature, each evaluation feature set corresponding to each initial feature is a feature set obtained by gradually deleting each initial feature or gradually adding each initial feature; The life prediction model is trained by each evaluation feature set to obtain a life prediction result, and then the model prediction error corresponding to each evaluation feature set is calculated according to the life prediction result, where the model prediction error corresponding to each evaluation feature set is the model prediction error corresponding to each initial feature; The combination determination unit is specifically used to: determine the target feature combination according to the model prediction errors corresponding to the evaluation feature sets; The result acquisition module 503 is specifically used to: determine the model prediction result corresponding to the target feature combination as the life prediction result corresponding to the target battery.

[0114] In some embodiments, the error acquisition unit is specifically used to: determine a first feature set, wherein the first feature set includes all initial features in the initial feature set; In each iteration, a first key feature in the first feature set is determined and deleted to obtain an evaluation feature set corresponding to the first key feature, and then the evaluation feature set corresponding to the first key feature is determined as the first feature set for the next iteration, until there is only one initial feature left in the first feature set, wherein the first key feature is an initial feature with the smallest model prediction error corresponding to the feature set obtained after the deletion determined by a cross-validation method; After several iterations, each evaluation feature set corresponding to each initial feature is obtained.

[0115] In some embodiments, the error acquisition unit is specifically used to: determine a second feature set and a third feature set, the second feature set includes all initial features in the initial feature set, and the third feature set is an empty set; In each iteration, a second key feature in the second feature set is determined and the second key feature is added to the third feature set and the second key feature is deleted from the second feature set, the added third feature set is the evaluation feature set corresponding to the second key feature, the added third feature set is determined to be the third feature set and the deleted second feature set is determined to be the second feature set for the next iteration until the second feature set is an empty set, wherein the second key feature is an initial feature with the smallest model prediction error corresponding to the feature set obtained after the addition determined by a cross-validation method; After several iterations, each evaluation feature set corresponding to each initial feature is obtained.

[0116] In some embodiments, the combination determination unit is specifically used to: determine the target model prediction error based on the model prediction errors of each evaluation feature set, and the target model prediction error is the minimum model prediction error; determine the evaluation feature set corresponding to the target model prediction error as the target feature combination.

[0117] In some embodiments, the feature acquisition module 501 is specifically used to: extract features according to the cycle curve differences, capacity increment curve differences and differential voltage curve differences between different cycles in the charge and discharge cycle data to obtain battery cycle features, wherein the battery cycle features include at least one of capacity difference, capacity increment difference and differential voltage difference; Extracting statistical features from the charge-discharge cycle data to obtain battery statistical features; Performing correlation analysis on the battery cycle characteristics and the battery statistical characteristics to obtain battery correlation characteristics; Statistical characteristic calculation is performed on the battery cycle feature, the battery statistical feature and the battery correlation feature to obtain a plurality of initial features, and the plurality of initial features constitute the initial feature set.

[0118] In some embodiments, the model acquisition unit includes a model determination component, a function determination component and a hyperparameter determination component, wherein: The model determination component is used to determine a Gaussian process regression model, the implicit function of the Gaussian process regression model obeys the Gaussian distribution of the mean function and the covariance function, and the Gaussian process regression model is the life prediction model; The function determination component is used to determine the mean function and covariance function corresponding to the Gaussian process regression model; The hyperparameter determination component is used to determine the hyperparameters corresponding to the Gaussian process regression model according to the mean function and the covariance function.

[0119] In some embodiments, the function determination component is specifically used to: determine the mean function to be a fixed basis function or a linear function; determine the covariance function to be a square exponential kernel function or a Mattern covariance function.

[0120] In some embodiments, the hyperparameter determination component is specifically used to: minimize the negative log marginal likelihood function of the mean function and the covariance function by the conjugate gradient descent method to obtain the target hyperparameter, and the target hyperparameter is the parameter that optimizes the performance of the Gaussian process regression model.

[0121] In some embodiments, the hyperparameter determination component is further specifically used for: the mean function includes a zero mean function, the covariance function includes a square exponential kernel function, the negative log marginal likelihood function is determined according to the zero mean function and the square exponential kernel function, and the negative log marginal likelihood function includes initial hyperparameters; the negative log marginal likelihood function is solved using a gradient descent algorithm to obtain the target hyperparameters, and the using the gradient descent algorithm to solve the negative log marginal likelihood function includes calculating the partial derivatives of the negative log marginal likelihood function with respect to the initial hyperparameters.

[0122] In the embodiment of the present invention, the optimal battery feature combination can be determined according to the influence of multiple battery features on the life prediction result to perform battery life prediction, thereby improving the accuracy of battery life prediction.

[0123] Fig.12Schematic diagram of the physical structure of the electronic device provided by the present invention. Fig.12 As shown, the electronic device may include: a processor 601, a communication interface 602, a memory 603 and a communication bus 604, wherein the processor 601, the communication interface 602 and the memory 603 communicate with each other through the communication bus 604. The processor 601 may call the logic instructions in the memory 603 to execute the battery life prediction method, which includes: extracting features from the charge and discharge cycle data of the target battery to obtain an initial feature set; performing feature screening according to the evaluation values ​​corresponding to each initial feature in the initial feature set to obtain a target feature combination, wherein the evaluation value is used to characterize the accuracy of the life prediction result; and determining the life prediction result corresponding to the target battery according to the target feature combination.

[0124] In addition, the logic instructions in the above-mentioned memory 603 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk.

[0125] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the battery life prediction method provided by the above methods, which method includes: extracting features from the charge and discharge cycle data of the target battery to obtain an initial feature set; performing feature screening according to the evaluation values ​​corresponding to each initial feature in the initial feature set to obtain a target feature combination, and the evaluation values ​​are used to characterize the accuracy of the life prediction results; and determining the life prediction result corresponding to the target battery according to the target feature combination.

[0126] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a server or data center that includes one or more available media integrated therein. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive Solid State Disk (SSD)), etc.

[0127] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the battery life prediction method provided by the above-mentioned methods, the method comprising: extracting features from the charge and discharge cycle data of the target battery to obtain an initial feature set; performing feature screening according to the evaluation values ​​corresponding to each initial feature in the initial feature set to obtain a target feature combination, the evaluation values ​​being used to characterize the accuracy of the life prediction result; and determining the life prediction result corresponding to the target battery according to the target feature combination.

[0128] The above-mentioned computer-readable storage medium can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, - but not limited to - an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM) or flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0129] Computer-readable signal media may include a data signal propagated in baseband or as part of a carrier wave, which carries a computer-readable program code. Such propagated data signals may take a variety of forms, including, but not limited to, electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0130] The program code embodied on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wire, optical cable, radio frequency (RF), etc., or any suitable combination of the foregoing.

[0131] Computer program code for performing the operations of the present specification may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0132] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0133] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0134] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A battery life prediction method, characterized in that: include: Extract features from the charge and discharge cycle data of the target battery to obtain an initial feature set; Performing feature screening according to the evaluation value corresponding to each initial feature in the initial feature set to obtain a target feature combination, wherein the evaluation value is used to characterize the accuracy of the life prediction result; According to the target feature combination, a life prediction result corresponding to the target battery is determined.

2. The life prediction method according to claim 1 is characterized in that the feature screening is performed according to the evaluation value corresponding to each initial feature in the initial feature set to obtain the target feature combination, including: Acquire a life prediction model, wherein the life prediction model uses the initial features in the initial feature set as model input and uses the battery life prediction result as model output; Performing model training on the life prediction model according to the initial features to obtain a model prediction error corresponding to the initial features, wherein the model prediction error is the evaluation value; The target feature combination is determined according to the model prediction errors corresponding to the initial features.

3. The life prediction method according to claim 2, characterized in that: The life prediction model is trained according to the initial features to obtain the model prediction errors corresponding to the initial features, including: Determine each evaluation feature set corresponding to each initial feature, where each evaluation feature set corresponding to each initial feature is a feature set obtained by gradually deleting each initial feature or gradually adding each initial feature; The life prediction model is trained by each evaluation feature set to obtain a life prediction result, and then the model prediction error corresponding to each evaluation feature set is calculated according to the life prediction result, where the model prediction error corresponding to each evaluation feature set is the model prediction error corresponding to each initial feature; Determining the target feature combination according to the model prediction errors corresponding to the initial features includes: Determining a target feature combination according to the model prediction errors corresponding to each evaluation feature set; Determining the life prediction result corresponding to the target battery according to the target feature combination includes: The model prediction result corresponding to the target feature combination is determined as the life prediction result corresponding to the target battery.

4. The battery life prediction method according to claim 3, characterized in that: The determining of each evaluation feature set corresponding to each initial feature includes: Determining a first feature set, wherein the first feature set includes all initial features in the initial feature set; In each iteration, a first key feature in the first feature set is determined and deleted to obtain an evaluation feature set corresponding to the first key feature, and then the evaluation feature set corresponding to the first key feature is determined as the first feature set for the next iteration, until there is only one initial feature left in the first feature set, wherein the first key feature is an initial feature with the smallest model prediction error corresponding to the feature set obtained after the deletion determined by a cross-validation method; After several iterations, each evaluation feature set corresponding to each initial feature is obtained.

5. The battery life prediction method according to claim 3, characterized in that: The determining of each evaluation feature set corresponding to each initial feature includes: Determine a second feature set and a third feature set, wherein the second feature set includes all initial features in the initial feature set, and the third feature set is an empty set; In each iteration, a second key feature in the second feature set is determined and the second key feature is added to the third feature set and the second key feature is deleted from the second feature set, the added third feature set is the evaluation feature set corresponding to the second key feature, the added third feature set is determined to be the third feature set and the deleted second feature set is determined to be the second feature set for the next iteration until the second feature set is an empty set, wherein the second key feature is an initial feature with the smallest model prediction error corresponding to the feature set obtained after the addition determined by a cross-validation method; After several iterations, each evaluation feature set corresponding to each initial feature is obtained.

6. The battery life prediction method according to claim 3, characterized in that: Determining the target feature combination according to the model prediction errors corresponding to the evaluation feature sets includes: Determining a target model prediction error according to the model prediction errors of each evaluation feature set, wherein the target model prediction error is the minimum model prediction error; Determine the evaluation feature set corresponding to the target model prediction error as the target feature combination.

7. The battery life prediction method according to claim 1, characterized in that: The feature extraction of the charge and discharge cycle data of the target battery to obtain an initial feature set includes: Extracting features according to the cycle curve differences, capacity increment curves and differential voltage curves between different cycles in the charge and discharge cycle data to obtain battery cycle features, wherein the battery cycle features include at least one of capacity difference, capacity increment difference and differential voltage; Extracting statistical features from the charge-discharge cycle data to obtain battery statistical features; Performing correlation analysis on the battery cycle characteristics and the battery statistical characteristics to obtain battery correlation characteristics; Statistical characteristic calculation is performed on the battery cycle feature, the battery statistical feature and the battery correlation feature to obtain a plurality of initial features, and the plurality of initial features constitute the initial feature set.

8. The battery life prediction method according to claim 2, characterized in that: The obtaining of the life prediction model comprises: Determine a Gaussian process regression model, wherein the implicit function of the Gaussian process regression model obeys the Gaussian distribution of the mean function and the covariance function, and the Gaussian process regression model is the life prediction model; Determine the mean function and covariance function corresponding to the Gaussian process regression model; According to the mean function and the covariance function, hyperparameters corresponding to the Gaussian process regression model are determined.

9. The battery life prediction method according to claim 8, characterized in that: The determining of the mean function and covariance function corresponding to the Gaussian process regression model includes: Determining the mean function to be a fixed basis function or a linear function; The covariance function is determined to be a squared exponential kernel function or a Mattern covariance function.

10. The battery life prediction method according to claim 8, characterized in that: Determining the hyperparameters corresponding to the Gaussian process regression model according to the mean function and the covariance function includes: The negative log marginal likelihood function is minimized for the mean function and the covariance function by using the conjugate gradient descent method to obtain a target hyperparameter, where the target hyperparameter is a parameter that optimizes the performance of the Gaussian process regression model.

11. The battery life prediction method according to claim 10, characterized in that: The mean function includes a zero mean function, the covariance function includes a square exponential kernel function, and the conjugate gradient descent method is used to minimize the negative logarithmic marginal likelihood function of the mean function and the covariance function to obtain the target hyperparameter, including: Determining the negative log marginal likelihood function according to the zero mean function and the square exponential kernel function, wherein the negative log marginal likelihood function includes initial hyperparameters; The negative log marginal likelihood function is solved using a gradient descent algorithm to obtain the target hyperparameter, wherein the negative log marginal likelihood function is solved using a gradient descent algorithm including calculating a partial derivative of the negative log marginal likelihood function with respect to the initial hyperparameter.

12. A battery life prediction device, characterized in that: include: A feature acquisition module is used to extract features from the charge and discharge cycle data of the target battery to obtain an initial feature set; Feature screening module: performing feature screening according to the evaluation value corresponding to each initial feature in the initial feature set to obtain a target feature combination, wherein the evaluation value is used to characterize the accuracy of the life prediction result; The result acquisition module is used to determine the life prediction result corresponding to the target battery according to the target feature combination.

13. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the battery life prediction method according to any one of claims 1 to 11 is implemented.

14. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the battery life prediction method according to any one of claims 1 to 11 is implemented.

15. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the battery life prediction method according to any one of claims 1 to 11 is implemented.

Images