Battery health degree evaluation method and device based on meta learning

Abstract

The invention discloses a battery health degree evaluation method and device based on meta-learning. According to the method, the real-time battery data is input into the feature extraction network constructed by the neural network and the meta-learning algorithm to extract key features, and then the features are sent into the battery health assessment network based on the meta-learning algorithm and the attention mechanism, so that the health state of the battery is assessed. Meanwhile, parameters of a battery evaluation network are optimized by using a meta-learning algorithm, and an evaluation strategy is adjusted in real time to adapt to the change of battery performance, so that comprehensive evaluation of the battery health degree is realized, and the problem that the battery health degree cannot be comprehensively evaluated in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the field of battery health assessment, and in particular to a battery health assessment method and device based on meta-learning. Background Art

[0002] In the current field of large-scale energy storage resource management, the assessment of battery health is not only a key factor in ensuring the stable operation of the system, but also an important means to accurately predict the battery life and effectively reduce maintenance costs. The accuracy of battery health assessment directly affects the reliability and economic benefits of the energy storage system, so it occupies a core position in energy storage and management.

[0003] However, existing battery health assessment technologies often have certain limitations. They usually rely on monitoring basic parameters such as battery voltage, current and temperature. Although these parameters can provide preliminary information on the battery's operating status, they are not sufficient to fully reflect the complex chemical and physical changes inside the battery. For example, factors such as the battery's internal impedance, charge and discharge cycle efficiency, self-discharge rate, and aging degree are all important indicators that are indispensable when evaluating battery health, but these indicators are often difficult to obtain through simple voltage, current and temperature monitoring. In addition, the diversity of battery operating environments and usage patterns also poses challenges to health assessment. Different environmental conditions, such as temperature changes, charge and discharge frequencies, workloads, etc., will affect the performance and life of the battery. These influences make it impossible for existing technologies to comprehensively evaluate the health of the battery. Summary of the invention

[0004] The present invention provides a battery health assessment method and device based on meta-learning to solve the problem that the battery health cannot be fully assessed in the prior art.

[0005] In a first aspect, the present application provides a battery health assessment method based on meta-learning, comprising:

[0006] Acquire real-time battery data; wherein the real-time battery data includes basic information of the battery, operating data of the battery and environmental data of the battery;

[0007] Inputting the real-time battery data into each trained feature extraction network so that each feature extraction network outputs each first data feature; wherein each feature extraction network is constructed by combining various types of neural networks with a meta-learning algorithm;

[0008] The first data features are input into a first battery health assessment network so that the first battery health assessment network outputs the health of each battery; wherein the battery health assessment network is constructed based on a meta-learning algorithm combined with an attention mechanism of long- and short-cycle features.

[0009] This application first comprehensively considers the basic information, operating data and environmental data of the battery to ensure that the evaluation process covers all key health influencing factors, thereby improving the comprehensiveness and accuracy of the evaluation. Then, a feature extraction network constructed by combining different types of neural networks with a meta-learning algorithm can automatically learn and identify the most sensitive and important features for battery health assessment from multi-dimensional data. Such a feature extraction process not only improves the relevance and accuracy of the features, but also enhances the model's adaptability to new battery data or different battery types. Finally, these selected features are input into the first battery health assessment network built based on a meta-learning algorithm and a long-short cycle feature attention mechanism. It can not only accurately capture the short-term operating status and long-term health trends of the battery, but also adjust the evaluation strategy in real time to adapt to changes in battery performance, thereby achieving a comprehensive assessment of battery health to solve the problem that the battery health cannot be comprehensively assessed in the prior art.

[0010] As a preferred embodiment of the first aspect, the acquiring of real-time battery data is specifically:

[0011] Acquire first real-time battery data;

[0012] Preprocessing the first real-time battery data according to a missing value processing method, a deduplication processing method, an outlier processing method, a noise processing method, and a data normalization processing method to obtain second real-time battery data;

[0013] The health assessment features in the second real-time battery data are extracted to obtain real-time battery data.

[0014] In this preferred embodiment, the present application first obtains the initial first real-time battery data to ensure that the evaluation process has original and comprehensive battery status information as a basis. Subsequently, a series of preprocessing steps are adopted, including missing value processing, deduplication processing, outlier processing, noise processing, and data normalization processing. These methods systematically optimize the data quality, eliminate noise and anomalies that may affect the accuracy of the analysis, and enhance the comparability of the data through normalization processing. Such preprocessing not only improves the consistency and reliability of the data, but also lays a solid foundation for subsequent feature extraction work. Finally, the health assessment features of the preprocessed second real-time battery data are extracted to obtain the preprocessed real-time battery data. The present application ensures that only the most relevant and representative features are used in the evaluation model, thereby improving the accuracy and efficiency of the battery health assessment.

[0015] As a preferred embodiment of the first aspect, the real-time battery data is input into each trained feature extraction network so that each feature extraction network outputs each first data feature, specifically:

[0016] Dividing the real-time battery data according to different types, different capacities and different granularities of batteries to obtain real-time battery data sets of various types;

[0017] According to the various types of real-time data sets and according to the various types of neural networks combined with meta-learning algorithms, feature extraction is performed on the various types of real-time data sets to obtain various first data features; wherein the various types of neural networks include convolutional neural networks, sliding windows and statistical feature extraction networks.

[0018] In this preferred embodiment, the present application forms a targeted real-time battery data set by carefully dividing the real-time battery data according to the different types, capacities and granularities of the batteries. The present application can ensure that the individual differences and specific operating conditions of the batteries are taken into account during the feature extraction process, thereby improving the pertinence and accuracy of the feature extraction. Subsequently, for each type of data set, a neural network architecture suitable for it is selected, such as a convolutional neural network (CNN), a sliding window or a statistical feature extraction network, and a meta-learning algorithm is combined for feature extraction. Such a combination can not only make full use of the advantages of various neural networks in processing different types of data, but also automatically optimize the feature extraction process through the rapid adaptation and learning ability of the meta-learning algorithm to adapt to the characteristics of various battery data. The meta-learning algorithm can learn and extract general rules from previous tasks, so that the model can quickly adjust and extract the most relevant features when facing new or different types of battery data. Therefore, the present application can significantly improve the accuracy and efficiency of battery health assessment, make the assessment results more reliable, and help to achieve accurate monitoring and prediction of battery health status, thereby optimizing battery maintenance and replacement strategies, reducing maintenance costs, and improving the stability and economic benefits of energy storage systems.

[0019] As a preferred embodiment of the first aspect, the obtaining of each first data feature is specifically:

[0020] The first data features include equal voltage rise time, equal voltage drop time, charging resistance, discharging resistance, equivalent cycle number and ICA curve.

[0021] This application extracts specific data features, including isobaric rise time, isobaric drop time, charging resistance, discharge resistance, equivalent number of cycles and ICA curve. This method can deeply capture the key indicators of battery performance. Isobaric rise time and isobaric drop time reflect the rate of voltage change of the battery during the charging and discharging process, and are important parameters for evaluating the internal chemical reaction rate of the battery and the degree of battery aging. The charging resistance and discharge resistance are directly related to the internal resistance of the battery. The change in internal resistance can indicate the charging and discharging efficiency and thermal management requirements of the battery. The equivalent number of cycles is a key indicator for evaluating the cycle life of the battery, which reflects the degree of degradation of the battery after multiple charge and discharge cycles. The ICA curve (current-capacity curve) can reveal the dynamic characteristics of the battery during the charging and discharging process, including the charging and discharging capacity and energy efficiency of the battery. The comprehensive analysis of these features provides a comprehensive perspective for the health of the battery, making the evaluation results more accurate and comprehensive. Therefore, this application can significantly improve the accuracy and reliability of battery health assessment, help to timely discover the degradation of battery performance, prevent potential failures, thereby extending the service life of the battery, optimizing maintenance strategies, reducing operation and maintenance costs, and improving the overall performance and economic benefits of the energy storage system.

[0022] As a preferred embodiment of the first aspect, the inputting the first data features into a first battery health assessment network so that the first battery health assessment network outputs the health of each battery further includes:

[0023] According to a meta-learning algorithm, the parameters of the first battery health assessment network are updated to obtain an optimal parameter set for each first data feature; wherein the meta-learning algorithm is a gradient-based MAML algorithm;

[0024] The first battery health assessment network is optimized according to the optimal parameter set of each first data feature to obtain a second battery health assessment network, wherein the second battery health assessment network is used to evaluate each second data feature.

[0025] In this preferred embodiment, the present application inputs the extracted battery health-related data features into the first battery health assessment network, and obtains a specific assessment task in combination with the type and operating conditions of the battery to be assessed. This method can ensure that the assessment network is customized for different battery characteristics and usage environments. By training and updating the parameters of the first network using the gradient-based MAML meta-learning algorithm, it is possible to efficiently learn and extract general and effective assessment strategies from various tasks to form an optimal parameter set. This process not only speeds up the convergence of the model, but also improves the model's adaptability to new tasks. Further, the first network is optimized using these optimal parameter sets, and the obtained second battery health assessment network can more accurately assess the new battery health assessment task on the basis of inheriting the previous learning results. Therefore, this method can significantly improve the accuracy and efficiency of battery health assessment, make the assessment results more reliable, and help to achieve accurate monitoring and prediction of battery health status, thereby optimizing battery maintenance and replacement strategies, reducing maintenance costs, and improving the overall performance and economic benefits of energy storage systems.

[0026] In a second aspect, the present application provides a battery health assessment device based on meta-learning. The battery health assessment device based on meta-learning includes an acquisition module, a feature extraction module and an assessment module;

[0027] The acquisition module is used to acquire real-time battery data; wherein the real-time battery data includes basic information of the battery, operating data of the battery and environmental data of the battery;

[0028] The feature extraction module is used to input the real-time battery data into each trained feature extraction network, so that each feature extraction network outputs each first data feature; wherein each feature extraction network is constructed by combining various types of neural networks with a meta-learning algorithm;

[0029] The evaluation module is used to input the various first data features into a first battery health evaluation network so that the first battery health evaluation network outputs various battery healths; wherein the battery health evaluation network is constructed based on a meta-learning algorithm combined with an attention mechanism of long- and short-cycle features.

[0030] This device uses three modules to divide the work and coordinate work to better comprehensively evaluate the health of the battery. This application first comprehensively considers the basic information, operating data and environmental data of the battery to ensure that the evaluation process covers all key health factors, thereby improving the comprehensiveness and accuracy of the evaluation. Then, a feature extraction network constructed by combining different types of neural networks with a meta-learning algorithm can automatically learn and identify the most sensitive and important features for battery health assessment from multi-dimensional data. Such a feature extraction process not only improves the relevance and accuracy of the features, but also enhances the model's adaptability to new battery data or different battery types. Finally, these selected features are input into the first battery health assessment network built based on the meta-learning algorithm and the long-short cycle feature attention mechanism, which can not only accurately capture the short-term operating status and long-term health trend of the battery, but also adjust the evaluation strategy in real time to adapt to changes in battery performance, thereby achieving a comprehensive assessment of battery health, so as to solve the problem that the battery health cannot be fully assessed in the prior art.

[0031] As a preferred embodiment of the second aspect, the acquiring of real-time battery data is specifically:

[0032] Acquire first real-time battery data;

[0033] Preprocessing the first real-time battery data according to a missing value processing method, a deduplication processing method, an outlier processing method, a noise processing method, and a data normalization processing method to obtain second real-time battery data;

[0034] The health assessment features in the second real-time battery data are extracted to obtain real-time battery data.

[0035] In this preferred embodiment, the present application first obtains the initial first real-time battery data to ensure that the evaluation process has original and comprehensive battery status information as a basis. Subsequently, a series of preprocessing steps are adopted, including missing value processing, deduplication processing, outlier processing, noise processing, and data normalization processing. These methods systematically optimize the data quality, eliminate noise and anomalies that may affect the accuracy of the analysis, and enhance the comparability of the data through normalization processing. Such preprocessing not only improves the consistency and reliability of the data, but also lays a solid foundation for subsequent feature extraction work. Finally, the health assessment features of the preprocessed second real-time battery data are extracted to obtain the preprocessed real-time battery data. The present application ensures that only the most relevant and representative features are used in the evaluation model, thereby improving the accuracy and efficiency of the battery health assessment.

[0036] As a preferred embodiment of the second aspect, the real-time battery data is input into each trained feature extraction network so that each feature extraction network outputs each first data feature, specifically:

[0037] Dividing the real-time battery data according to different types, different capacities and different granularities of batteries to obtain real-time battery data sets of various types;

[0038] According to the various types of real-time data sets and according to the various types of neural networks combined with meta-learning algorithms, feature extraction is performed on the various types of real-time data sets to obtain various first data features; wherein the various types of neural networks include convolutional neural networks, sliding windows and statistical feature extraction networks.

[0039] In this preferred embodiment, the present application forms a targeted real-time battery data set by carefully dividing the real-time battery data according to the different types, capacities and granularities of the batteries. The present application can ensure that the individual differences and specific operating conditions of the batteries are taken into account during the feature extraction process, thereby improving the pertinence and accuracy of the feature extraction. Subsequently, for each type of data set, a neural network architecture suitable for it is selected, such as a convolutional neural network (CNN), a sliding window or a statistical feature extraction network, and a meta-learning algorithm is combined for feature extraction. Such a combination can not only make full use of the advantages of various neural networks in processing different types of data, but also automatically optimize the feature extraction process through the rapid adaptation and learning ability of the meta-learning algorithm to adapt to the characteristics of various battery data. The meta-learning algorithm can learn and extract general rules from previous tasks, so that the model can quickly adjust and extract the most relevant features when facing new or different types of battery data. Therefore, the present application can significantly improve the accuracy and efficiency of battery health assessment, make the assessment results more reliable, and help to achieve accurate monitoring and prediction of battery health status, thereby optimizing battery maintenance and replacement strategies, reducing maintenance costs, and improving the stability and economic benefits of energy storage systems.

[0040] As a preferred embodiment of the second aspect, the obtaining of each first data feature is specifically:

[0041] The first data features include equal voltage rise time, equal voltage drop time, charging resistance, discharging resistance, equivalent cycle number and ICA curve.

[0042] This application extracts specific data features, including isobaric rise time, isobaric drop time, charging resistance, discharge resistance, equivalent number of cycles and ICA curve. This method can deeply capture the key indicators of battery performance. Isobaric rise time and isobaric drop time reflect the rate of voltage change of the battery during the charging and discharging process, and are important parameters for evaluating the internal chemical reaction rate of the battery and the degree of battery aging. The charging resistance and discharge resistance are directly related to the internal resistance of the battery. The change in internal resistance can indicate the charging and discharging efficiency and thermal management requirements of the battery. The equivalent number of cycles is a key indicator for evaluating the cycle life of the battery, which reflects the degree of degradation of the battery after multiple charge and discharge cycles. The ICA curve (current-capacity curve) can reveal the dynamic characteristics of the battery during the charging and discharging process, including the charging and discharging capacity and energy efficiency of the battery. The comprehensive analysis of these features provides a comprehensive perspective for the health of the battery, making the evaluation results more accurate and comprehensive. Therefore, this application can significantly improve the accuracy and reliability of battery health assessment, help to timely discover the degradation of battery performance, prevent potential failures, thereby extending the service life of the battery, optimizing maintenance strategies, reducing operation and maintenance costs, and improving the overall performance and economic benefits of the energy storage system.

[0043] As a preferred embodiment of the second aspect, the inputting the first data features into the first battery health assessment network so that the first battery health assessment network outputs the health of each battery further includes:

[0044] According to a meta-learning algorithm, the parameters of the first battery health assessment network are updated to obtain an optimal parameter set for each first data feature; wherein the meta-learning algorithm is a gradient-based MAML algorithm;

[0045] The first battery health assessment network is optimized according to the optimal parameter set of each first data feature to obtain a second battery health assessment network, wherein the second battery health assessment network is used to evaluate each second data feature.

[0046] In this preferred embodiment, the present application inputs the extracted battery health-related data features into the first battery health assessment network, and obtains a specific assessment task in combination with the type and operating conditions of the battery to be assessed. This method can ensure that the assessment network is customized for different battery characteristics and usage environments. By training and updating the parameters of the first network using the gradient-based MAML meta-learning algorithm, it is possible to efficiently learn and extract general and effective assessment strategies from various tasks to form an optimal parameter set. This process not only speeds up the convergence of the model, but also improves the model's adaptability to new tasks. Further, the first network is optimized using these optimal parameter sets, and the obtained second battery health assessment network can more accurately assess the new battery health assessment task on the basis of inheriting the previous learning results. Therefore, this method can significantly improve the accuracy and efficiency of battery health assessment, make the assessment results more reliable, and help to achieve accurate monitoring and prediction of battery health status, thereby optimizing battery maintenance and replacement strategies, reducing maintenance costs, and improving the overall performance and economic benefits of energy storage systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 : A flowchart of an embodiment of a battery health assessment method based on meta-learning provided in the present application;

[0048] Figure 2 : A schematic diagram of the structure of the API interface request parameters provided for this application;

[0049] Figure 3 : A schematic diagram of the structure of the API interface output parameters provided for this application;

[0050] Figure 4 : A structural diagram of an embodiment of a battery health assessment device based on meta-learning provided in this application. DETAILED DESCRIPTION

[0051] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] Embodiment 1

[0053] Please refer to Figure 1 , which is a battery health assessment method based on meta-learning provided in an embodiment of the present invention.

[0054] In this embodiment, the process of the battery health assessment method based on meta-learning in this application is described in detail through steps S01-S03.

[0055] S01: Acquire real-time battery data; wherein the real-time battery data includes basic information of the battery, operating data of the battery and environmental data of the battery.

[0056] As a preferred embodiment of the first embodiment, the real-time battery data is obtained; wherein the real-time battery data includes basic information of the battery, operating data of the battery and environmental data of the battery, specifically:

[0057] The massive amount of raw data related to batteries is collected through big data technology, including but not limited to: basic information of batteries, including battery type, battery rated capacity, model, manufacturer, etc.; real-time operation data of batteries, mainly including battery charging current, discharge current, charging voltage, discharge voltage, charging temperature, discharge temperature, chargeable power, dischargeable power, chargeable amount, dischargeable amount, SOC value, charge and discharge times; environmental data, including factors affecting battery performance such as ambient temperature and humidity. For example, in a large-scale energy storage cluster system, basic information is sorted out through the Excel table of each energy storage device resource model. Real-time operation data and environmental data are collected from each energy storage plant through the edge gateway and the power grid IEC61850 protocol on the energy storage plant side. After the edge gateway collects data, it is sent to the big data Kafka cluster through the standard protocol MQTT. The big data platform obtains resource data from the Internet of Things platform through the Kafka cluster, stores it on the Hadoop cluster, and performs cleaning, conversion, association, analysis, mining and other analyses on it. The time interval between real-time operation data and environmental data is 15 minutes, and the time span is more than 1 year.

[0058] The collected large-scale battery raw data is uniformly preprocessed to provide consistent and reliable data for the subsequent health assessment model, including missing value processing, deduplication processing, outlier check, noise, data normalization, etc. to ensure data quality, so as to better reflect the trends and changes of battery data. For example: in a large-scale energy storage cluster system, data preprocessing uses missing value processing, outlier check, and data normalization to improve data quality.

[0059] Missing value processing: Data values ​​may be missing in the original data, that is, there are data points with no data in the data set. This will affect the results in subsequent processing, so the missing data values ​​need to be filled in and the missing values ​​are replaced by the average value estimation method.

[0060] Outlier check: Use outlier check algorithms to check, identify and delete outliers in raw data to improve data stability and consistency.

[0061] Data normalization: Use linear transformation of the original data to map the result to between 0 and 1. The specific formula is:

[0062]

[0063] , where x is the original data, x min is the minimum value of the original data, x max is the maximum value of the original data.

[0064] In this preferred embodiment, the present application first obtains the initial real-time battery data to ensure that the evaluation process has original and comprehensive battery status information as a basis. Subsequently, a series of preprocessing steps are adopted, including missing value processing, deduplication processing, outlier processing, noise processing, and data normalization processing. These methods systematically optimize the data quality, eliminate noise and anomalies that may affect the accuracy of the analysis, and enhance the comparability of the data through normalization processing. Such preprocessing not only improves the consistency and reliability of the data, but also lays a solid foundation for subsequent feature extraction work. Finally, the health assessment features of the preprocessed second real-time battery data are extracted to obtain the preprocessed real-time battery data. The present application ensures that only the most relevant and representative features are used in the evaluation model, thereby improving the accuracy and efficiency of the battery health assessment.

[0065] S02: Inputting the real-time battery data into each trained feature extraction network so that each feature extraction network outputs each first data feature; wherein each feature extraction network is constructed by combining various types of neural networks with a meta-learning algorithm.

[0066] As a preferred embodiment of the first embodiment, the real-time battery data is input into each trained feature extraction network so that each feature extraction network outputs each first data feature, specifically:

[0067] Dividing the real-time battery data according to different types, different capacities and different granularities of batteries to obtain real-time battery data sets of various types;

[0068] According to the various types of real-time data sets, and according to the various types of neural networks combined with meta-learning algorithms, feature extraction is performed on the various types of real-time data sets to obtain various first data features; wherein the various types of neural networks include convolutional neural networks, sliding windows, and statistical feature extraction networks. The real-time battery data is divided according to different types, different capacities, and different granularities of batteries to obtain various types of real-time battery data sets; wherein the meta-learning algorithm is a gradient-based MAML meta-learning algorithm.

[0069] Specifically, the first data features include but are not limited to: constant voltage rise time, constant voltage drop time, charging resistance, discharge resistance, charging and discharging efficiency, number of charging and discharging times, battery capacity attenuation rate, equivalent cycle number, ICA data, etc.

[0070] This application extracts specific data features, including isobaric rise time, isobaric drop time, charging resistance, discharge resistance, equivalent number of cycles and ICA curve. This method can deeply capture the key indicators of battery performance. Isobaric rise time and isobaric drop time reflect the rate of voltage change of the battery during the charging and discharging process, and are important parameters for evaluating the internal chemical reaction rate of the battery and the degree of battery aging. The charging resistance and discharge resistance are directly related to the internal resistance of the battery. The change in internal resistance can indicate the charging and discharging efficiency and thermal management requirements of the battery. The equivalent number of cycles is a key indicator for evaluating the cycle life of the battery, which reflects the degree of degradation of the battery after multiple charge and discharge cycles. The ICA curve (current-capacity curve) can reveal the dynamic characteristics of the battery during the charging and discharging process, including the charging and discharging capacity and energy efficiency of the battery. The comprehensive analysis of these features provides a comprehensive perspective for the health of the battery, making the evaluation results more accurate and comprehensive. Therefore, this application can significantly improve the accuracy and reliability of battery health assessment, help to timely discover the degradation of battery performance, prevent potential failures, thereby extending the service life of the battery, optimizing maintenance strategies, reducing operation and maintenance costs, and improving the overall performance and economic benefits of the energy storage system.

[0071] In this preferred embodiment, the present application forms a targeted real-time battery data set by carefully dividing the real-time battery data according to the different types, capacities and granularities of the batteries. The present application can ensure that the individual differences and specific operating conditions of the batteries are taken into account during the feature extraction process, thereby improving the pertinence and accuracy of the feature extraction. Subsequently, for each type of data set, a neural network architecture suitable for it is selected, such as a convolutional neural network (CNN), a sliding window or a statistical feature extraction network, and a meta-learning algorithm is combined for feature extraction. Such a combination can not only make full use of the advantages of various neural networks in processing different types of data, but also automatically optimize the feature extraction process through the rapid adaptation and learning ability of the meta-learning algorithm to adapt to the characteristics of various battery data. The meta-learning algorithm can learn and extract general rules from previous tasks, so that the model can quickly adjust and extract the most relevant features when facing new or different types of battery data. Therefore, the present application can significantly improve the accuracy and efficiency of battery health assessment, make the assessment results more reliable, and help to achieve accurate monitoring and prediction of battery health status, thereby optimizing battery maintenance and replacement strategies, reducing maintenance costs, and improving the stability and economic benefits of energy storage systems.

[0072] S03: Input each of the first data features into a first battery health assessment network so that the first battery health assessment network outputs each battery health; wherein the battery health assessment network is constructed based on a meta-learning algorithm combined with an attention mechanism of long- and short-cycle features.

[0073] As a preferred embodiment of the first embodiment, the inputting of each first data feature into the first battery health assessment network so that the first battery health assessment network outputs each battery health is specifically:

[0074] A meta-learning-based LSTM neural network algorithm model that combines long- and short-cycle feature fusion with an attention mechanism is constructed to evaluate battery health. It uses the ability to quickly adapt to different battery types and battery characteristics with a small amount of samples, and can learn how to quickly adjust the parameters of the basic model based on a small amount of training data to adapt to new subtasks, thereby adapting to the health assessment of energy storage batteries of different types and conditions.

[0075] First, the input data vector is constructed: each first data feature is integrated to construct the input data vector.

[0076] Secondly, divide the data set: divide the input data vector into training data set, validation data set and test data set.

[0077] Then train the model: select the attention mechanism and LSTM neural network algorithm to form the basic model, and train the model based on the meta-learning mechanism. The model input is the feature vector data of the battery, and the output is the battery health (SOH) evaluation value. The meta-learning algorithm summarizes the experience of each task to learn the commonality between tasks, and then applies it to new tasks to improve the efficiency and performance of the model in learning the characteristics of new tasks. Specifically, the training set is first divided into multiple subtasks, each of which contains a part of the data. For example, in a large-scale energy storage cluster system, according to the time dimension, battery type and battery granularity, the extracted battery feature data of different time periods, different battery types (lithium-ion batteries, nickel-hydrogen batteries, etc.) and different granularities (including energy storage power stations, battery stacks, battery clusters, and battery cells) are used to establish different subtasks, and then trained through multiple subtasks to obtain the operating rules of battery feature data under different conditions.

[0078] Finally, predict the battery health: Use the trained model to predict the battery health of the test data set. Input the feature data vector test data set into the model, and obtain the corresponding battery health (SOH) prediction value through forward propagation.

[0079] Among them, this application also needs to evaluate the model, and evaluate the performance of the model through a test data set. The evaluation indicators include RMSE root mean square error (Root Mean Squared Error), MAE mean absolute error (Mean Absolute Error), etc., which are used to measure the performance and generalization ability of the model. For example: in the example of a large-scale energy storage cluster system, RMSE root mean square error is used for model evaluation, specifically:

[0080]

[0081] in is the SOH evaluation value of the energy storage battery, y i is the true value of the health status SOH of the energy storage battery, n is the prediction step, RMSE is the root mean square error, which means the square root of the error between the predicted value and the true value. The smaller it is, the more accurate it is.

[0082] As a preferred embodiment of the first embodiment, the step of inputting each of the first data features into a first battery health assessment network so that the first battery health assessment network outputs each battery health, further includes:

[0083] According to a meta-learning algorithm, the parameters of the first battery health assessment network are updated to obtain an optimal parameter set for each first data feature; wherein the meta-learning algorithm is a gradient-based MAML algorithm;

[0084] The first battery health assessment network is optimized according to the optimal parameter set of each first data feature to obtain a second battery health assessment network, wherein the second battery health assessment network is used to evaluate each second data feature.

[0085] More specifically: suppose the battery health assessment model training has three tasks, marked as T1, T2, and T3. For a general battery health assessment model, the model parameter θ is first randomly initialized, and the model is used to train task T1. During the training process, a relatively optimal parameter can be found for task T1. Similarly, we can find the relatively optimal parameters for tasks T2 and T3. That is, through a set of randomly initialized parameters θ, three relatively optimal parameters can be generated: The relatively optimal parameters calculated at T1, T2, and T3 in the meta-learning MAML algorithm Find a common, generalized optimal parameter θ * , when training the next batch of tasks, replace the randomly generated parameters during initialization to reduce the number of steps of gradient descent and shorten the training time. The formula is:

[0086]

[0087] In the formula, θ * represents the optimization parameter, θ represents the initialization parameter, and β represents a hyperparameter. It means that through the parameters The calculated gradient result for task Ti.

[0088] The MAML algorithm has two main loops, one of which is an inner loop used to determine the optimal parameters for each task in the current task set. The outer loop is used to calculate the optimal parameter θ from the inner loop. * The corresponding gradient is used to update the initial random parameters θ of the task.

[0089] After the model is trained and optimized, it can be applied to actual scenarios to evaluate the health of large-scale energy storage batteries. For example, in a large-scale energy storage cluster system instance, the large-scale energy storage battery health evaluation capability is provided to the outside world through the encapsulated API interface. The business system inputs the latest battery data through the API interface, and the model outputs the corresponding evaluation results, thereby improving the accuracy and efficiency of battery health evaluation, extending battery life, and reducing maintenance costs, thereby enhancing the reliability and economy of the energy storage system. Specific API interface request parameters are as follows: Figure 2 As shown, the interface output parameters are as follows Figure 3 shown.

[0090] In this preferred embodiment, the present application inputs the extracted battery health-related data features into the first battery health assessment network, and obtains a specific assessment task in combination with the type and operating conditions of the battery to be assessed. This method can ensure that the assessment network is customized for different battery characteristics and usage environments. By training and updating the parameters of the first network using the gradient-based MAML meta-learning algorithm, it is possible to efficiently learn and extract general and effective assessment strategies from various tasks to form an optimal parameter set. This process not only speeds up the convergence of the model, but also improves the model's adaptability to new tasks. Further, the first network is optimized using these optimal parameter sets, and the obtained second battery health assessment network can more accurately assess the new battery health assessment task on the basis of inheriting the previous learning results. Therefore, this method can significantly improve the accuracy and efficiency of battery health assessment, make the assessment results more reliable, and help to achieve accurate monitoring and prediction of battery health status, thereby optimizing battery maintenance and replacement strategies, reducing maintenance costs, and improving the overall performance and economic benefits of energy storage systems.

[0091] This application first comprehensively considers the basic information, operating data and environmental data of the battery to ensure that the evaluation process covers all key health influencing factors, thereby improving the comprehensiveness and accuracy of the evaluation. Then, a feature extraction network constructed by combining different types of neural networks with a meta-learning algorithm can automatically learn and identify the most sensitive and important features for battery health assessment from multi-dimensional data. Such a feature extraction process not only improves the relevance and accuracy of the features, but also enhances the model's adaptability to new battery data or different battery types. Finally, these selected features are input into the first battery health assessment network built based on a meta-learning algorithm and a long-short cycle feature attention mechanism. It can not only accurately capture the short-term operating status and long-term health trends of the battery, but also adjust the evaluation strategy in real time to adapt to changes in battery performance, thereby achieving a comprehensive assessment of battery health to solve the problem that the battery health cannot be comprehensively assessed in the prior art.

[0092] Embodiment 2

[0093] Please refer to Figure 4 , which is a battery health assessment device based on meta-learning provided in an embodiment of the present application.

[0094] In this embodiment, the battery health assessment device based on meta-learning includes an acquisition module 10 , a feature extraction module 20 and an assessment module 30 .

[0095] The acquisition module 10 is used to acquire real-time battery data; wherein the real-time battery data includes basic information of the battery, operating data of the battery and environmental data of the battery.

[0096] As a preferred embodiment of the second embodiment, the real-time battery data is obtained; wherein the real-time battery data includes basic information of the battery, operating data of the battery and environmental data of the battery, specifically:

[0097] The massive amount of raw data related to batteries is collected through big data technology, including but not limited to: basic information of batteries, including battery type, battery rated capacity, model, manufacturer, etc.; real-time operation data of batteries, mainly including battery charging current, discharge current, charging voltage, discharge voltage, charging temperature, discharge temperature, chargeable power, dischargeable power, chargeable amount, dischargeable amount, SOC value, charge and discharge times; environmental data, including factors affecting battery performance such as ambient temperature and humidity. For example, in a large-scale energy storage cluster system, basic information is sorted out through the Excel table of each energy storage device resource model. Real-time operation data and environmental data are collected from each energy storage plant through the edge gateway and the power grid IEC61850 protocol on the energy storage plant side. After the edge gateway collects data, it is sent to the big data Kafka cluster through the standard protocol MQTT. The big data platform obtains resource data from the Internet of Things platform through the Kafka cluster, stores it on the Hadoop cluster, and performs cleaning, conversion, association, analysis, mining and other analyses on it. The time interval between real-time operation data and environmental data is 15 minutes, and the time span is more than 1 year.

[0098] The collected large-scale battery raw data is uniformly preprocessed to provide consistent and reliable data for the subsequent health assessment model, including missing value processing, deduplication processing, outlier check, noise, data normalization, etc. to ensure data quality, so as to better reflect the trends and changes of battery data. For example: in a large-scale energy storage cluster system, data preprocessing uses missing value processing, outlier check, and data normalization to improve data quality.

[0099] Missing value processing: Data values ​​may be missing in the original data, that is, there are data points with no data in the data set. This will affect the results in subsequent processing, so the missing data values ​​need to be filled in and the missing values ​​are replaced by the average value estimation method.

[0100] Outlier check: Use outlier check algorithms to check, identify and delete outliers in raw data to improve data stability and consistency.

[0101] Data normalization: Use linear transformation of the original data to map the result to between 0 and 1. The specific formula is:

[0102]

[0103] , where x is the original data, x min is the minimum value of the original data, x max is the maximum value of the original data.

[0104] In this preferred embodiment, the present application first obtains the initial real-time battery data to ensure that the evaluation process has original and comprehensive battery status information as a basis. Subsequently, a series of preprocessing steps are adopted, including missing value processing, deduplication processing, outlier processing, noise processing, and data normalization processing. These methods systematically optimize the data quality, eliminate noise and anomalies that may affect the accuracy of the analysis, and enhance the comparability of the data through normalization processing. Such preprocessing not only improves the consistency and reliability of the data, but also lays a solid foundation for subsequent feature extraction work. Finally, the health assessment features of the preprocessed second real-time battery data are extracted to obtain the preprocessed real-time battery data. The present application ensures that only the most relevant and representative features are used in the evaluation model, thereby improving the accuracy and efficiency of the battery health assessment.

[0105] The feature extraction module 20 is used to input the real-time battery data into each trained feature extraction network so that each feature extraction network outputs each first data feature; wherein each feature extraction network is constructed by combining various types of neural networks with a meta-learning algorithm.

[0106] As a preferred embodiment of the second embodiment, the real-time battery data is input into each trained feature extraction network so that each feature extraction network outputs each first data feature, specifically:

[0107] Dividing the real-time battery data according to different types, different capacities and different granularities of batteries to obtain real-time battery data sets of various types;

[0108] According to the various types of real-time data sets, and according to the various types of neural networks combined with meta-learning algorithms, feature extraction is performed on the various types of real-time data sets to obtain various first data features; wherein the various types of neural networks include convolutional neural networks, sliding windows, and statistical feature extraction networks. The real-time battery data is divided according to different types, different capacities, and different granularities of batteries to obtain various types of real-time battery data sets; wherein the meta-learning algorithm is a gradient-based MAML meta-learning algorithm.

[0109] Specifically, the first data features include but are not limited to: constant voltage rise time, constant voltage drop time, charging resistance, discharge resistance, charging and discharging efficiency, number of charging and discharging times, battery capacity attenuation rate, equivalent cycle number, ICA data, etc.

[0110] This application extracts specific data features, including isobaric rise time, isobaric drop time, charging resistance, discharge resistance, equivalent number of cycles and ICA curve. This method can deeply capture the key indicators of battery performance. Isobaric rise time and isobaric drop time reflect the rate of voltage change of the battery during the charging and discharging process, and are important parameters for evaluating the internal chemical reaction rate of the battery and the degree of battery aging. The charging resistance and discharge resistance are directly related to the internal resistance of the battery. The change in internal resistance can indicate the charging and discharging efficiency and thermal management requirements of the battery. The equivalent number of cycles is a key indicator for evaluating the cycle life of the battery, which reflects the degree of degradation of the battery after multiple charge and discharge cycles. The ICA curve (current-capacity curve) can reveal the dynamic characteristics of the battery during the charging and discharging process, including the charging and discharging capacity and energy efficiency of the battery. The comprehensive analysis of these features provides a comprehensive perspective for the health of the battery, making the evaluation results more accurate and comprehensive. Therefore, this application can significantly improve the accuracy and reliability of battery health assessment, help to timely discover the degradation of battery performance, prevent potential failures, thereby extending the service life of the battery, optimizing maintenance strategies, reducing operation and maintenance costs, and improving the overall performance and economic benefits of the energy storage system.

[0111] In this preferred embodiment, the present application forms a targeted real-time battery data set by carefully dividing the real-time battery data according to the different types, capacities and granularities of the batteries. The present application can ensure that the individual differences and specific operating conditions of the batteries are taken into account during the feature extraction process, thereby improving the pertinence and accuracy of the feature extraction. Subsequently, for each type of data set, a neural network architecture suitable for it is selected, such as a convolutional neural network (CNN), a sliding window or a statistical feature extraction network, and a meta-learning algorithm is combined for feature extraction. Such a combination can not only make full use of the advantages of various neural networks in processing different types of data, but also automatically optimize the feature extraction process through the rapid adaptation and learning ability of the meta-learning algorithm to adapt to the characteristics of various battery data. The meta-learning algorithm can learn and extract general rules from previous tasks, so that the model can quickly adjust and extract the most relevant features when facing new or different types of battery data. Therefore, the present application can significantly improve the accuracy and efficiency of battery health assessment, make the assessment results more reliable, and help to achieve accurate monitoring and prediction of battery health status, thereby optimizing battery maintenance and replacement strategies, reducing maintenance costs, and improving the stability and economic benefits of energy storage systems.

[0112] The evaluation module 30 is used to input the various first data features into the first battery health evaluation network so that the first battery health evaluation network outputs the health of each battery; wherein the battery health evaluation network is constructed based on a meta-learning algorithm combined with an attention mechanism of long- and short-cycle features.

[0113] As a preferred embodiment of the second embodiment, the inputting of each first data feature into the first battery health assessment network so that the first battery health assessment network outputs each battery health is specifically:

[0114] A meta-learning-based LSTM neural network algorithm model that combines long- and short-cycle feature fusion with an attention mechanism is constructed to evaluate battery health. It uses the ability to quickly adapt to different battery types and battery characteristics with a small amount of samples, and can learn how to quickly adjust the parameters of the basic model based on a small amount of training data to adapt to new subtasks, thereby adapting to the health assessment of energy storage batteries of different types and conditions.

[0115] First, the input data vector is constructed: each first data feature is integrated to construct the input data vector.

[0116] Secondly, divide the data set: divide the input data vector into training data set, validation data set and test data set.

[0117] Then train the model: select the attention mechanism and LSTM neural network algorithm to form the basic model, and train the model based on the meta-learning mechanism. The model input is the feature vector data of the battery, and the output is the battery health (SOH) evaluation value. The meta-learning algorithm summarizes the experience of each task to learn the commonality between tasks, and then applies it to new tasks to improve the efficiency and performance of the model in learning the characteristics of new tasks. Specifically, the training set is first divided into multiple subtasks, each of which contains a part of the data. For example, in a large-scale energy storage cluster system, according to the time dimension, battery type and battery granularity, the extracted battery feature data of different time periods, different battery types (lithium-ion batteries, nickel-hydrogen batteries, etc.) and different granularities (including energy storage power stations, battery stacks, battery clusters, and battery cells) are used to establish different subtasks, and then trained through multiple subtasks to obtain the operating rules of battery feature data under different conditions.

[0118] Finally, predict the battery health: Use the trained model to predict the battery health of the test data set. Input the feature data vector test data set into the model, and obtain the corresponding battery health (SOH) prediction value through forward propagation.

[0119] Among them, this application also needs to evaluate the model, and evaluate the performance of the model through a test data set. The evaluation indicators include RMSE root mean square error (Root Mean Squared Error), MAE mean absolute error (Mean Absolute Error), etc., which are used to measure the performance and generalization ability of the model. For example: in the example of a large-scale energy storage cluster system, RMSE root mean square error is used for model evaluation, specifically:

[0120]

[0121] in is the SOH evaluation value of the energy storage battery, y i is the true value of the health status SOH of the energy storage battery, n is the prediction step, RMSE is the root mean square error, which means the square root of the error between the predicted value and the true value. The smaller it is, the more accurate it is.

[0122] As a preferred embodiment of the second embodiment, the step of inputting the first data features into a first battery health assessment network so that the first battery health assessment network outputs the health of each battery further includes:

[0123] According to a meta-learning algorithm, the parameters of the first battery health assessment network are updated to obtain an optimal parameter set for each first data feature; wherein the meta-learning algorithm is a gradient-based MAML algorithm;

[0124] The first battery health assessment network is optimized according to the optimal parameter set of each first data feature to obtain a second battery health assessment network, wherein the second battery health assessment network is used to evaluate each second data feature.

[0125] More specifically: suppose the battery health assessment model training has three tasks, marked as T1, T2, and T3. For a general battery health assessment model, the model parameter θ is first randomly initialized, and the model is used to train task T1. During the training process, a relatively optimal parameter can be found for task T1. Similarly, we can find the relatively optimal parameters for tasks T2 and T3. That is, through a set of randomly initialized parameters θ, three relatively optimal parameters can be generated: The relatively optimal parameters calculated at T1, T2, and T3 in the meta-learning MAML algorithm Find a common, generalized optimal parameter θ * , when training the next batch of tasks, replace the randomly generated parameters during initialization to reduce the number of steps of gradient descent and shorten the training time. The formula is:

[0126]

[0127] In the formula, θ * represents the optimization parameter, θ represents the initialization parameter, and β represents a hyperparameter. It means that through the parameters The calculated gradient result for task Ti.

[0128] The MAML algorithm has two main loops, one of which is an inner loop used to determine the optimal parameters for each task in the current task set. The outer loop is used to calculate the optimal parameter θ from the inner loop. * The corresponding gradient is used to update the initial random parameters θ of the task.

[0129] After the model is trained and optimized, it can be applied to actual scenarios to evaluate the health of large-scale energy storage batteries. For example, in a large-scale energy storage cluster system instance, the large-scale energy storage battery health evaluation capability is provided to the outside world through the encapsulated API interface. The business system inputs the latest battery data through the API interface, and the model outputs the corresponding evaluation results, thereby improving the accuracy and efficiency of battery health evaluation, extending battery life, and reducing maintenance costs, thereby enhancing the reliability and economy of the energy storage system. Specific API interface request parameters are as follows: Figure 2 As shown, the interface output parameters are as follows Figure 3 shown.

[0130] In this preferred embodiment, the present application inputs the extracted battery health-related data features into the first battery health assessment network, and obtains a specific assessment task in combination with the type and operating conditions of the battery to be assessed. This method can ensure that the assessment network is customized for different battery characteristics and usage environments. By training and updating the parameters of the first network using the gradient-based MAML meta-learning algorithm, it is possible to efficiently learn and extract general and effective assessment strategies from various tasks to form an optimal parameter set. This process not only speeds up the convergence of the model, but also improves the model's adaptability to new tasks. Further, the first network is optimized using these optimal parameter sets, and the obtained second battery health assessment network can more accurately assess the new battery health assessment task on the basis of inheriting the previous learning results. Therefore, this method can significantly improve the accuracy and efficiency of battery health assessment, make the assessment results more reliable, and help to achieve accurate monitoring and prediction of battery health status, thereby optimizing battery maintenance and replacement strategies, reducing maintenance costs, and improving the overall performance and economic benefits of energy storage systems.

[0131] This device uses three modules to divide the work and coordinate work to better comprehensively evaluate the health of the battery. This application first comprehensively considers the basic information, operating data and environmental data of the battery to ensure that the evaluation process covers all key health factors, thereby improving the comprehensiveness and accuracy of the evaluation. Then, a feature extraction network constructed by combining different types of neural networks with a meta-learning algorithm can automatically learn and identify the most sensitive and important features for battery health assessment from multi-dimensional data. Such a feature extraction process not only improves the relevance and accuracy of the features, but also enhances the model's adaptability to new battery data or different battery types. Finally, these selected features are input into the first battery health assessment network built based on the meta-learning algorithm and the long-short cycle feature attention mechanism, which can not only accurately capture the short-term operating status and long-term health trend of the battery, but also adjust the evaluation strategy in real time to adapt to changes in battery performance, thereby achieving a comprehensive assessment of battery health, so as to solve the problem that the battery health cannot be fully assessed in the prior art.

[0132] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A battery health assessment method based on meta-learning, It is characterized in that include: Acquire real-time battery data; wherein the real-time battery data includes basic information of the battery, operating data of the battery and environmental data of the battery; Inputting the real-time battery data into each trained feature extraction network so that each feature extraction network outputs each first data feature; wherein each feature extraction network is constructed by combining various types of neural networks with a meta-learning algorithm; The first data features are input into a first battery health assessment network so that the first battery health assessment network outputs the health of each battery; wherein the battery health assessment network is constructed based on a meta-learning algorithm combined with an attention mechanism of long- and short-cycle features.

2. According to the meta-learning-based battery health assessment method of claim 1, It is characterized in that The real-time battery data acquisition is specifically as follows: Acquire first real-time battery data; Preprocessing the first real-time battery data according to a missing value processing method, a deduplication processing method, an outlier processing method, a noise processing method, and a data normalization processing method to obtain second real-time battery data; The health assessment features in the second real-time battery data are extracted to obtain real-time battery data.

3. The battery health assessment method based on meta-learning according to claim 1, It is characterized in that The step of inputting the real-time battery data into each trained feature extraction network so that each feature extraction network outputs each first data feature is specifically: Dividing the real-time battery data according to different types, different capacities and different granularities of batteries to obtain real-time battery data sets of various types; According to the various types of real-time data sets and according to the various types of neural networks combined with meta-learning algorithms, feature extraction is performed on the various types of real-time data sets to obtain various first data features; wherein the various types of neural networks include convolutional neural networks, sliding windows and statistical feature extraction networks.

4. The battery health assessment method based on meta-learning according to claim 3, It is characterized in that The obtaining of each first data feature is specifically: The first data features include equal voltage rise time, equal voltage drop time, charging resistance, discharging resistance, equivalent cycle number and ICA curve.

5. The battery health assessment method based on meta-learning according to claim 1, It is characterized in that The step of inputting the first data features into a first battery health assessment network so that the first battery health assessment network outputs the health of the respective batteries further includes: According to a meta-learning algorithm, the parameters of the first battery health assessment network are updated to obtain an optimal parameter set for each first data feature; wherein the meta-learning algorithm is a gradient-based MAML algorithm; The first battery health assessment network is optimized according to the optimal parameter set of each first data feature to obtain a second battery health assessment network, wherein the second battery health assessment network is used to evaluate each second data feature.

6. A battery health assessment device based on meta-learning, It is characterized in that It includes an acquisition module, a feature extraction module and an evaluation module; The acquisition module is used to acquire real-time battery data; wherein the real-time battery data includes basic information of the battery, operating data of the battery and environmental data of the battery; The feature extraction module is used to input the real-time battery data into each trained feature extraction network, so that each feature extraction network outputs each first data feature; wherein each feature extraction network is constructed by combining various types of neural networks with a meta-learning algorithm; The evaluation module is used to input the various first data features into a first battery health evaluation network so that the first battery health evaluation network outputs various battery healths; wherein the battery health evaluation network is constructed based on a meta-learning algorithm combined with an attention mechanism of long- and short-cycle features.

7. The battery health assessment device based on meta-learning according to claim 6, It is characterized in that The real-time battery data acquisition is specifically as follows: Acquire first real-time battery data; Preprocessing the first real-time battery data according to a missing value processing method, a deduplication processing method, an outlier processing method, a noise processing method, and a data normalization processing method to obtain second real-time battery data; The health assessment features in the second real-time battery data are extracted to obtain real-time battery data.

8. The battery health assessment device based on meta-learning according to claim 6, It is characterized in that The step of inputting the real-time battery data into each trained feature extraction network so that each feature extraction network outputs each first data feature is specifically: Dividing the real-time battery data according to different types, different capacities and different granularities of batteries to obtain real-time battery data sets of various types; According to the various types of real-time data sets and according to the various types of neural networks combined with meta-learning algorithms, feature extraction is performed on the various types of real-time data sets to obtain various first data features; wherein the various types of neural networks include convolutional neural networks, sliding windows and statistical feature extraction networks.

9. The battery health assessment device based on meta-learning according to claim 8, It is characterized in that The obtaining of each first data feature is specifically: The first data features include equal voltage rise time, equal voltage drop time, charging resistance, discharging resistance, equivalent cycle number and ICA curve.

10. The battery health assessment device based on meta-learning according to claim 6, It is characterized in that The step of inputting the first data features into a first battery health assessment network so that the first battery health assessment network outputs the health of the respective batteries further includes: According to a meta-learning algorithm, the parameters of the first battery health assessment network are updated to obtain an optimal parameter set for each first data feature; wherein the meta-learning algorithm is a gradient-based MAML algorithm; The first battery health assessment network is optimized according to the optimal parameter set of each first data feature to obtain a second battery health assessment network, wherein the second battery health assessment network is used to evaluate each second data feature.

Images