Battery health state analysis method under multi-battery configuration

Abstract

The invention discloses a battery health state analysis method under multi-battery configuration. The method comprises the steps of obtaining target charging and discharging time sequence data of multi-battery task configuration; inputting the target charging and discharging time sequence data into a pre-trained shared feature extraction model to output shared features and task prediction contribution degrees; and inputting the target charging and discharging time sequence data, the sharing characteristics and the task prediction contribution degree into a pre-trained battery health state analysis model to output a battery health state corresponding to each battery task. According to the method, the sharing characteristics and the task prediction contribution degree are extracted, the target charging and discharging time sequence data, the sharing characteristics and the task prediction contribution degree are processed by using the battery health state analysis model, and the final battery health state value is output by combining the information in three aspects, so that the accuracy is greatly improved; therefore, the determination of the high-precision battery health state is also beneficial to the operation safety of equipment.

Description

Technical Field

[0001] This application relates to the field of battery health status analysis technology, specifically to a method for analyzing battery health status in a multi-battery configuration. Background Technology

[0002] In the fields of electric vehicles, energy storage systems, and portable electronic devices, batteries are the core energy supply components. Their health status directly affects the operational safety and maintenance costs of the equipment. Battery state of health (SOH), as a key indicator for measuring the degree of degradation of the battery's current capacity relative to its initial capacity, is the core basis for formulating battery maintenance and replacement strategies. Therefore, battery SOH analysis technology has become a research focus in the field of battery health monitoring. With the diversification of battery application scenarios, batteries in actual working conditions often exhibit the characteristics of multiple types and multiple operating conditions. They not only cover battery types with different chemical compositions, but also need to operate at different charging rates and temperatures. SOH estimation across battery types and operating conditions has become a practical requirement.

[0003] Existing battery health status analysis methods simply train all battery data together. However, in reality, different battery tasks (such as being used for different devices or using different charging and discharging strategies) have different impacts on their health status. Therefore, this analysis method cannot accurately obtain battery health status estimates, which is detrimental to the safety of device operation. Summary of the Invention

[0004] In view of this, this application provides a battery health status analysis method under multi-battery configuration to solve the problem that existing battery health status analysis methods cannot accurately obtain battery health status estimates, which is detrimental to the safety of equipment operation.

[0005] To achieve the above objectives, the following solution is proposed:

[0006] Firstly, a method for analyzing battery health status under a multi-battery configuration includes:

[0007] Obtain the target charge / discharge timing data for multi-battery task configuration;

[0008] The target charge-discharge time series data is input into a pre-trained shared feature extraction model to output shared features and task prediction contribution. The shared feature extraction model is trained using real multi-battery charge-discharge data as training samples and real shared features and task prediction contribution of the multi-battery charge-discharge data as sample labels.

[0009] The target charge-discharge time series data, shared features, and task prediction contribution are input into a pre-trained battery health status analysis model to output the battery health status corresponding to each battery task. The battery health status analysis model is trained using real multi-battery charge-discharge data, shared features, and task prediction contribution as training samples, and using the real battery health status of each battery task in the multi-battery charge-discharge data as sample labels.

[0010] Preferably, acquiring the target charge / discharge timing data for the multi-battery task configuration includes:

[0011] Obtain the raw charge / discharge timing data of the multi-battery task configuration;

[0012] Divide the original charge and discharge timing data of each battery task in the original charge and discharge timing data;

[0013] For each battery task, the original charge-discharge timing data corresponding to each cycle in the original charge-discharge timing data of the battery task is divided into a separate set of first charge-discharge data.

[0014] Data filtering is performed on the first charge-discharge data of each group to obtain the target charge-discharge timing data.

[0015] Preferably, the first charge-discharge data in each group is filtered to obtain target charge-discharge timing data, including:

[0016] For each set of the first charge and discharge data, extract the maximum discharge capacity and the nominal battery capacity of the first set of charge and discharge data;

[0017] Calculate the actual health status of the battery in this cycle based on the maximum discharge capacity and the battery's nominal capacity.

[0018] Compare the actual health status of each battery task in each cycle with the preset normal range;

[0019] The first charge-discharge data corresponding to the actual health state that exceeds the normal range is removed from the original charge-discharge timing data, and the remaining first charge-discharge data are used as the second charge-discharge data.

[0020] Feature extraction is performed on the second charge-discharge data of each group to obtain the target charge-discharge time series data.

[0021] Preferably, the step of extracting features from each group of the second charge-discharge data to obtain the target charge-discharge time series data includes:

[0022] For each set of the second charge-discharge data, the second charge-discharge data is divided into equal-length intervals according to a preset number to form a multivariate time series;

[0023] The multivariate time series is interpolated under the conditions of voltage, current, charge capacity, and the differential of charge capacity with respect to voltage, respectively, to obtain the target time series;

[0024] The target time series corresponding to the second charge and discharge data of each group are fused to obtain the target charge and discharge time series data.

[0025] Preferably, the shared feature extraction model includes an initial extraction module, a joint prediction module, a joint identification module, and a first fusion module;

[0026] The first input terminal of the initial extraction module serves as the input terminal of the shared feature extraction model;

[0027] The first output of the initial extraction module is connected to the input of the joint prediction module, the second output of the initial extraction module is connected to the input of the joint identification module, and the first output of the joint prediction module and the output of the joint identification module are both connected to the input of the first fusion module.

[0028] The second output of the joint prediction module and the third output of the initial extraction module both serve as outputs of the shared feature extraction model.

[0029] Preferably, the joint identification module includes a gradient inversion unit and an identification unit;

[0030] The input terminal of the gradient inversion unit serves as the input terminal of the joint authentication module, and the output terminal of the gradient inversion unit is connected to the input terminal of the authentication unit; the output terminal of the authentication unit serves as the output terminal of the joint authentication module.

[0031] The gradient inversion unit inputs the data output by the initial extraction unit to the discrimination unit during the forward propagation process.

[0032] During the backpropagation process, the gradient inversion unit reverses the data returned by the discrimination unit and outputs it to the initial extraction unit.

[0033] Preferably, the battery health status analysis model includes a weighting module, a dedicated feature extraction module, a second fusion module, and a task prediction module;

[0034] The input terminals of the weighting module and the dedicated feature extraction module are both used as input terminals of the battery health status analysis model.

[0035] The output of the weighting module is connected to the input of the dedicated feature extraction module and the input of the task prediction module, respectively. The output of the dedicated feature extraction module is connected to the input of the second fusion module, and the output of the second fusion module is connected to the input of the task prediction module.

[0036] The output of the task prediction module serves as the output of the battery health status analysis model.

[0037] Secondly, a battery health status analysis device for a multi-battery configuration includes:

[0038] The data acquisition module is used to acquire the target charge and discharge timing data of the multi-battery task configuration;

[0039] The shared feature extraction module is used to input the target charge-discharge time series data into a pre-trained shared feature extraction model to output shared features and task prediction contribution. The shared feature extraction model is trained using real multi-battery charge-discharge data as training samples and real shared features and task prediction contribution of the multi-battery charge-discharge data as sample labels.

[0040] The battery health status analysis module is used to input the target charge-discharge time series data, shared features, and task prediction contribution into a pre-trained battery health status analysis model to output the battery health status corresponding to each battery task. The battery health status analysis model is trained using real multi-battery charge-discharge data, shared features, and task prediction contribution as training samples, and using the real battery health status of each battery task in the multi-battery charge-discharge data as sample labels.

[0041] Thirdly, a battery health status analysis device for a multi-battery configuration includes a memory and a processor;

[0042] The memory is used to store programs;

[0043] The processor is configured to execute the program to implement the various steps of the battery health status analysis method under a multi-battery configuration as described in any of the first aspects.

[0044] Fourthly, a storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the battery health status analysis method under a multi-battery configuration as described in any of the first aspects.

[0045] As can be seen from the above technical solution, this application first obtains the target charge-discharge time-series data of the multi-battery task configuration; inputs the target charge-discharge time-series data into a pre-trained shared feature extraction model to output shared features and task prediction contributions; the shared feature extraction model is trained using real multi-battery charge-discharge data as training samples and the real shared features and task prediction contributions of the multi-battery charge-discharge data as sample labels; inputs the target charge-discharge time-series data, shared features, and task prediction contributions into a pre-trained battery health status analysis model to output the battery health status corresponding to each battery task; the battery health status analysis model is trained using real multi-battery charge-discharge data, shared features, and task prediction contributions as training samples and the real battery health status of each battery task in the multi-battery charge-discharge data as sample labels. This application obtains target charge-discharge time-series data for multi-battery task configurations, which is a prerequisite for subsequent extraction of shared features. Task configuration means clarifying that the data is collected under different operating conditions or environments. Then, the target charge-discharge time-series data is input into a pre-trained shared feature extraction model to extract shared features. Shared features contain general information that is independent of battery tasks, which can benefit all tasks and minimize task-specific biases. The task prediction contribution can be regarded as a weight value, which can quantify the confidence of the shared feature extraction model in predicting the battery task samples. It can achieve personalized adaptation and enhance interpretability. It can guide the battery health status analysis model to identify which shared features are important and which are secondary. Then, the battery health status analysis model processes the target charge-discharge time-series data, shared features, and task prediction contribution. Combining the information from these three aspects, the final battery health status value is output, which greatly improves accuracy. Therefore, determining a high-precision battery health status will also benefit the safety of equipment operation. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0047] Figure 1 An optional flowchart of a battery health status analysis method under a multi-battery configuration provided in an embodiment of this application;

[0048] Figure 2 This is a schematic diagram of the structure of a shared feature extraction model provided in an embodiment of this application;

[0049] Figure 3This is a schematic diagram of the structure of a battery health status analysis model provided in an embodiment of this application;

[0050] Figure 4 This is a schematic diagram of the structure of a battery health status analysis device in a multi-battery configuration provided in an embodiment of this application;

[0051] Figure 5 This is a schematic diagram of a battery health status analysis device with a multi-battery configuration provided in an embodiment of this application. Detailed Implementation

[0052] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0053] This invention can be used in a wide variety of general-purpose or special-purpose computing environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor devices, distributed computing environments including any of the above devices, etc.

[0054] This invention provides a method for analyzing battery health status in a multi-battery configuration. This method can be applied to various computer terminals or smart terminals, and its execution entity can be the processor or server of the computer terminal or smart terminal. The method flowchart is shown below. Figure 1 As shown, it specifically includes:

[0055] S1: Obtain the target charge / discharge timing data for the multi-battery task configuration.

[0056] Battery State of Health (SOH) is a core indicator that measures the degree of degradation of a battery's current performance relative to its initial state. It is usually defined as the ratio of the battery's current actual discharge capacity to its nominal capacity. It is a key basis for assessing the remaining lifespan of a battery and formulating maintenance and replacement strategies, and is widely used in battery health monitoring tasks in scenarios such as electric vehicles and energy storage systems.

[0057] It is understandable that the differences in battery configurations come from the inherent properties of the batteries themselves, such as different chemical compositions and models, as well as from external operating conditions, such as different charging rates and different temperatures. Therefore, as long as there are two or more scenarios with non-negligible differences in battery configurations (which will lead to different degradation patterns in battery health), they can be divided into multiple independent battery tasks.

[0058] The target charge / discharge timing data of so many battery task configurations demonstrates that the battery health status changes with repeated damage and usage time.

[0059] S2: Input the target charging and discharging time series data into a pre-trained shared feature extraction model to output shared features and task prediction contribution;

[0060] The shared feature extraction model is trained using real multi-battery charging and discharging data as training samples, and using the real shared features and task prediction contribution of the multi-battery charging and discharging data as sample labels.

[0061] Traditional analysis methods typically model data independently for individual batteries. This approach ignores the common patterns and regularities in performance degradation that may exist when multiple batteries operate under the same or similar conditions. In other words, each model starts from scratch and does not utilize the features contained in the data of other batteries. Therefore, this application introduces the concept of shared features because these features contain general and task-independent information that can benefit all tasks while minimizing task-specific biases.

[0062] Shared features are common features across battery tasks that are automatically extracted by the shared feature extraction model after learning the mapping relationship between "multi-battery charge and discharge data → battery health status".

[0063] The contribution of task prediction can enable subsequent battery health status analysis models to focus more on samples that provide reliable task-independent information, thus avoiding the model from overfitting task-specific features.

[0064] S3: Input the target charge / discharge time series data, shared features, and task prediction contribution into the pre-trained battery health status analysis model to output the battery health status corresponding to each battery task; the battery health status analysis model is trained using real multi-battery charge / discharge data, shared features, and task prediction contribution as training samples, and using the real battery health status of each battery task in the multi-battery charge / discharge data as sample labels.

[0065] For battery health status analysis models, target charge-discharge time-series data ensures the basic fidelity of the model analysis and prevents the loss of original details of the battery task due to feature extraction; shared features represent the universal and transferable physicochemical laws in the battery aging process, so that the model does not have to start from scratch, especially for new batteries with scarce data; task prediction contribution dynamically indicates to the model how much shared features should be utilized for a certain battery task.

[0066] As can be seen from the above technical solution, this application first obtains the target charge-discharge time-series data of the multi-battery task configuration; inputs the target charge-discharge time-series data into a pre-trained shared feature extraction model to output shared features and task prediction contributions; the shared feature extraction model is trained using real multi-battery charge-discharge data as training samples and the real shared features and task prediction contributions of the multi-battery charge-discharge data as sample labels; inputs the target charge-discharge time-series data, shared features, and task prediction contributions into a pre-trained battery health status analysis model to output the battery health status corresponding to each battery task; the battery health status analysis model is trained using real multi-battery charge-discharge data, shared features, and task prediction contributions as training samples and the real battery health status of each battery task in the multi-battery charge-discharge data as sample labels. This application obtains target charge-discharge time-series data for multi-battery task configurations, which is a prerequisite for subsequent extraction of shared features. Task configuration means clarifying that the data is collected under different operating conditions or environments. Then, the target charge-discharge time-series data is input into a pre-trained shared feature extraction model to extract shared features. Shared features contain general information that is independent of battery tasks, which can benefit all tasks and minimize task-specific biases. The task prediction contribution can be regarded as a weight value, which can quantify the confidence of the shared feature extraction model in predicting the battery task samples. It can achieve personalized adaptation and enhance interpretability. It can guide the battery health status analysis model to identify which shared features are important and which are secondary. Then, the battery health status analysis model processes the target charge-discharge time-series data, shared features, and task prediction contribution. Combining the information from these three aspects, the final battery health status value is output, which greatly improves accuracy. Therefore, determining a high-precision battery health status will also benefit the safety of equipment operation.

[0067] This application focuses on the extraction of shared features from multi-battery task configurations, breaking through the limitations of single-task learning in existing technologies. It actively mines and reuses common degradation patterns among multiple battery configurations, improving knowledge utilization and enabling the shared feature extraction and battery health status analysis models to better adapt to the actual operating conditions of multi-battery configurations. Furthermore, this application addresses the problems of large feature differences and difficulty in generalization among multi-battery task configurations, providing a unified and clean shared feature foundation for battery health status analysis.

[0068] The process for obtaining target charge-discharge timing data of multi-battery task configuration in the method provided by this embodiment of the invention is specifically described below:

[0069] Obtain the raw charge / discharge timing data of the multi-battery task configuration;

[0070] Divide the original charge and discharge timing data of each battery task in the original charge and discharge timing data;

[0071] For each battery task, the original charge-discharge timing data corresponding to each cycle in the original charge-discharge timing data of the battery task is divided into a separate set of first charge-discharge data.

[0072] Data filtering is performed on the first charge-discharge data of each group to obtain the target charge-discharge timing data.

[0073] Specifically, it can be understood that the health of a battery gradually deteriorates in cycles. A complete charge-discharge cycle, such as from full charge to empty and then full charge, will cause a cumulative and irreversible micro-damage to the battery's chemical system. Capacity decay and internal resistance increase are the result of hundreds or thousands of cycles. Therefore, dividing the above process according to data cycles is equivalent to establishing a discrete aging trajectory for each battery from brand new to scrap. The data point of each cycle is a sample on this trajectory.

[0074] In one example, for two NCM batteries, where battery A is cycled 300 times and battery B is cycled 280 times, battery A is divided into 300 groups based on its 1st to 300th cycles (each group contains data from a single cycle of battery A), and battery B is divided into 280 groups based on its 1st to 280th cycles.

[0075] The process of filtering the first charge-discharge data in each group of the above process to obtain the target charge-discharge timing data is explained in detail below:

[0076] For each set of the first charge and discharge data, extract the maximum discharge capacity and the nominal battery capacity of the first set of charge and discharge data;

[0077] Calculate the actual health status of the battery in this cycle based on the maximum discharge capacity and the battery's nominal capacity.

[0078] Compare the actual health status of each battery task in each cycle with the preset normal range;

[0079] The first charge-discharge data corresponding to the actual health state that exceeds the normal range is removed from the original charge-discharge timing data, and the remaining first charge-discharge data are used as the second charge-discharge data.

[0080] Feature extraction is performed on the second charge-discharge data of each group to obtain the target charge-discharge time series data.

[0081] Specifically, the formula for calculating the true health status of the battery based on the maximum discharge capacity and nominal capacity for each cycle is as follows:

[0082] ;

[0083] in, Indicates true health status. Indicates the maximum discharge capacity. This indicates the battery's nominal capacity.

[0084] The normal range can be set to [0.5, 2], which ensures that the remaining first charge and discharge data of each group are within the normal range, thereby ensuring the rationality of the data distribution.

[0085] Specifically, the step of extracting features from each group of the second charge-discharge data to obtain the target charge-discharge time series data in the above process may include:

[0086] For each set of the second charge-discharge data, the second charge-discharge data is divided into equal-length intervals according to a preset number to form a multivariate time series;

[0087] The multivariate time series is interpolated under the conditions of voltage, current, charge capacity, and the differential of charge capacity with respect to voltage, respectively, to obtain the target time series;

[0088] The target time series corresponding to the second charge and discharge data of each group are fused to obtain the target charge and discharge time series data.

[0089] Specifically, to characterize the performance features of the battery during the charging process, key parameters, including voltage, current, capacity, and timestamp, can be extracted from the constant current constant voltage (CCCV) stage. Since the length of CCCV data in different cycles fluctuates due to differences in sampling rate and operating conditions, linear interpolation is used to standardize the data of each cycle.

[0090] To ensure data consistency, a preset number of 1000 is set. This means that the time series of each cycle is divided into 1000 equal-length intervals, and finally, the second charge / discharge data of each cycle is resampled into 1000 points of fixed length.

[0091] ;

[0092] Where k = 0, 1, ..., 999, This represents the equally spaced timestamps generated after linear interpolation of the second charge / discharge data in the CCCV phase of a certain cycle. This indicates the start timestamp of the constant current constant voltage (CCCV) phase in a single charge-discharge cycle of a battery task. This indicates the end timestamp of the constant current constant voltage (CCCV) phase in a single charge-discharge cycle for a specific battery task. The two timestamps correspond to the same battery task.

[0093] The process of handling the differences in voltage, current, capacitance, and charge capacity as a function of voltage is as follows:

[0094] ;

[0095] ;

[0096] ;

[0097] ;

[0098] in, Indicates linear interpolation. This represents the voltage after linear interpolation. This represents the current after linear interpolation. This represents the capacity after linear interpolation. This represents the original timestamp of the constant current constant voltage (CCCV) phase during a single charge-discharge cycle of the battery.

[0099] Therefore, after the above processing, each second charge-discharge data is modeled as a multivariate time series containing the voltage, current, capacity, and the differential of charge capacity with voltage variation, i.e., the target time series, and then fused to obtain the target charge-discharge time series data.

[0100] Optionally, to effectively achieve cross-task shared representation learning, this application proposes two complementary learning methods: adversarial learning and supervised learning. These two methods work together to extract clean and information-rich shared features while effectively mitigating task interference and noise issues. However, in practical applications, the shared feature space often retains specific details, leading to interference effects and poor performance. To address this problem, adversarial learning can be used. First, a joint discrimination module is introduced, which is a multi-classifier used to identify the task source of the shared features, i.e., predicting the battery task to which a given input feature belongs.

[0101] However, while adversarial learning methods can eliminate biases in specific tasks, they may also lose relevant information useful for the task, leading to performance degradation. To address this issue, supervised learning methods can be combined to ensure that shared representations retain informational value for all tasks. Therefore, a joint predictor is introduced, which combines the joint discriminator module and the joint predictor to jointly extract shared features. While extracting shared features, task interference and noise problems are effectively mitigated, balancing the utilization of shared features across battery tasks with the utilization of battery-specific features. This enables accurate and generalized estimation and analysis of battery health status under multi-battery tasks.

[0102] Specifically, this application pre-trains a shared feature extraction model, such as Figure 2 As shown, the shared feature extraction model includes an initial extraction module, a joint prediction module, a joint identification module, and a first fusion module;

[0103] The first input terminal A1 of the initial extraction module serves as the input terminal of the shared feature extraction model;

[0104] The first output terminal a1 of the initial extraction module is connected to the input terminal of the joint prediction module, the second output terminal a2 of the initial extraction module is connected to the input terminal of the joint identification module, and the first output terminal b1 of the joint prediction module and the output terminal of the joint identification module are both connected to the input terminal of the first fusion module.

[0105] The second output terminal b2 of the joint prediction module and the third output terminal a3 of the initial extraction module both serve as output terminals of the shared feature extraction model.

[0106] Specifically, set This represents the initial shared features extracted by the initial extraction module. The initial extraction module is configured with a neural network structure (such as 1-DCNN, LSTM, etc., which are not limited in this implementation). The joint discrimination module takes the initial shared features as input and predicts the index of battery task m (where m∈{1,2,...,M}, and the target charge / discharge time-series data contains cyclic data corresponding to M battery tasks). For m, the classification loss function of the joint discrimination module is... Defined as:

[0107] ;

[0108] in, This represents the raw charge / discharge timing data for battery task m. This represents a sample that follows a data distribution D. Find the expected value, where 1[m] is an indicator function defined as follows:

[0109] ;

[0110] By minimizing The joint identification module can accurately classify the initial shared features input based on the source of the battery task. If the initial shared features contain task-specific information, the joint identification module can easily distinguish them; if they do not contain such information, that is, if the initial shared features are completely unrelated to the battery task, the joint identification module will have difficulty classifying them correctly, indicating that they do not possess task-specific features.

[0111] Therefore, in order to achieve this invariance, this application configures a gradient flipping unit and a discrimination unit in the joint discrimination module;

[0112] The input terminal of the gradient inversion unit serves as the input terminal of the joint authentication module, and the output terminal of the gradient inversion unit is connected to the input terminal of the authentication unit; the output terminal of the authentication unit serves as the output terminal of the joint authentication module.

[0113] The gradient inversion unit inputs the data output by the initial extraction unit to the discrimination unit during the forward propagation process.

[0114] During the backpropagation process, the gradient inversion unit reverses the data returned by the discrimination unit and outputs it to the initial extraction unit.

[0115] Specifically, the gradient inversion unit multiplies the gradient returned by the discrimination unit by only calculating the gradient during backpropagation. The initial shared features are then passed to the initial extraction module. Through this adversarial mechanism, the initial extraction module attempts to learn to prevent the discrimination unit from distinguishing the shared features of the battery task to which the initial shared features belong, while the discrimination unit attempts to accurately distinguish the battery task to which the initial shared features belong. Ultimately, the initial shared features retain only the common information across tasks and eliminate task-specific interference.

[0116] Specifically, for the initial shared features The gradient reversal unit acts as the identity function during forward propagation:

[0117] ;

[0118] During backpropagation, the gradient is multiplied by a negative number, thus changing its direction:

[0119] ;

[0120] in It is the identity matrix. It is a hyperparameter that controls the strength of gradient reversal, which is gradually increased. This approach allows the shared feature extraction model to initially focus on maintaining supervised performance, and then gradually shift its attention to minimizing task-specific differences. This adversarial training achieves simultaneous optimization of the initial shared features and the joint discriminator through a formula for calculating the true health state, ensuring that the shared features capture only common, task-independent information.

[0121] In addition, the joint predictor can also be used to predict general labels. :

[0122] ;

[0123] in It is a supervised loss function, such as mean squared error (MSE) in regression tasks or cross-entropy loss in classification tasks. This supervision ensures that the extracted shared features retain task-related knowledge, rather than collapsing into a noise-dominated representation.

[0124] In order to improve the robustness and generalization ability of the model, this application proposes two knowledge enhancement strategies: feature-level fusion and logit-level enhancement. These two strategies can make full use of the advantages of shared features while preserving the inherent information unique to the battery task.

[0125] The shared feature extraction model also includes a first fusion module, used to fuse shared features and task-specific features to generate more informative representations for each battery task. Element-wise addition can be used as the fusion method. For task-specific features, fuse feature vectors The calculation formula is:

[0126] ;

[0127] This enables the two information sources to be directly and efficiently integrated. Through feature-level fusion, the shared feature extraction model can not only generalize better to different tasks, but also capture the unique features of each learning task, thereby improving overall performance.

[0128] This application utilizes multi-battery task learning to mine and leverage shared features and dependencies among battery tasks, thereby reducing the risk of overfitting in a single task, improving data utilization, and enhancing the model's adaptability to complex scenarios. It can be applied in multiple fields such as computer vision, natural language processing, and battery health monitoring.

[0129] Existing physics-based methods rely on battery electrochemical principles and physical degradation mechanisms to construct degradation models, requiring complex calculations to simulate microscopic degradation processes, resulting in high computational complexity. Furthermore, these physical models struggle to efficiently adapt to the nonlinear and time-varying degradation characteristics of batteries in multi-battery configurations, directly impacting the real-time performance and adaptability of battery health state estimation across multiple configuration scenarios. The proposed solution, however, eliminates the need for complex physical modeling to capture degradation processes. Instead, it integrates common degradation patterns across multiple configurations through data-driven multi-battery task joint learning, fundamentally reducing complexity while ensuring the adaptability of battery health state estimation across multi-battery configurations.

[0130] The structure diagram of the battery health status analysis model is as follows: Figure 3 As shown, it includes a weighting module, a dedicated feature extraction module, a second fusion module, and a task prediction module;

[0131] The input terminal of the weighting module and the first input terminal C1 of the dedicated feature extraction module are both used as input terminals of the battery health status analysis model;

[0132] The output of the weighting module is connected to the second input C2 of the dedicated feature extraction module and the first input D1 of the task prediction module, respectively. The output of the dedicated feature extraction module is connected to the input of the second fusion module, and the output of the second fusion module is connected to the second input D2 of the task prediction module.

[0133] The output of the task prediction module serves as the output of the battery health status analysis model.

[0134] Specifically, feature-level fusion integrates knowledge during the representation stage, while logit-level enhancement refines task-specific predictions by introducing logit-level guidance. This method utilizes the task-specific prediction contribution of shared features to guide task-specific learning. During supervised training, the contribution of shared features to the prediction of each sample can be estimated. This contribution originates from the joint prediction unit, and in classification tasks, this probability distribution corresponds to the probability distribution of the true labels. :

[0135] ;

[0136] In regression tasks, this metric represents the error relative to the true value:

[0137] ;

[0138] in, It is a scaling factor that controls the sensitivity of the weights. Indicates the first One sample.

[0139] To integrate this knowledge into the learning of specific features, a weighted module is used for each sample. distribute This is used to control its impact on the training of the pool health status analysis model.

[0140] ;

[0141] in, This represents the training loss function of the battery health status analysis model (task-specific layer). Indicates the total number of samples. This indicates the task prediction module. This dynamic weighting mechanism ensures that the dedicated feature extraction module pays more attention to samples that provide reliable task-irrelevant information, thereby avoiding the battery health status analysis model from overfitting the battery task-specific features.

[0142] and Figure 1 Corresponding to the method described above, embodiments of the present invention also provide a battery health status analysis device under a multi-battery configuration, used for analyzing the battery health status of a battery in a multi-battery configuration. Figure 1 The specific implementation of the method, the battery health status analysis device under multi-battery configuration provided in this embodiment of the invention, can be integrated into a computer terminal or various mobile devices. Figure 4 This paper introduces a battery health status analysis device for multi-battery configurations, such as... Figure 4 As shown, the device may include:

[0143] Data acquisition module 10 is used to acquire target charge and discharge timing data of multi-battery task configuration;

[0144] The shared feature extraction module 20 is used to input the target charge-discharge time series data into a pre-trained shared feature extraction model to output shared features and task prediction contribution. The shared feature extraction model is trained using real multi-battery charge-discharge data as training samples and real shared features and task prediction contribution of the multi-battery charge-discharge data as sample labels.

[0145] The battery health status analysis module 30 is used to input the target charge-discharge time series data, shared features, and task prediction contribution into a pre-trained battery health status analysis model to output the battery health status corresponding to each battery task. The battery health status analysis model is trained using real multi-battery charge-discharge data, shared features, and task prediction contribution as training samples, and using the real battery health status of each battery task in the multi-battery charge-discharge data as sample labels.

[0146] As can be seen from the above technical solution, this application first obtains the target charge-discharge time-series data of the multi-battery task configuration; inputs the target charge-discharge time-series data into a pre-trained shared feature extraction model to output shared features and task prediction contributions; the shared feature extraction model is trained using real multi-battery charge-discharge data as training samples and the real shared features and task prediction contributions of the multi-battery charge-discharge data as sample labels; inputs the target charge-discharge time-series data, shared features, and task prediction contributions into a pre-trained battery health status analysis model to output the battery health status corresponding to each battery task; the battery health status analysis model is trained using real multi-battery charge-discharge data, shared features, and task prediction contributions as training samples and the real battery health status of each battery task in the multi-battery charge-discharge data as sample labels. This application obtains target charge-discharge time-series data for multi-battery task configurations, which is a prerequisite for subsequent extraction of shared features. Task configuration means clarifying that the data is collected under different operating conditions or environments. Then, the target charge-discharge time-series data is input into a pre-trained shared feature extraction model to extract shared features. Shared features contain general information that is independent of battery tasks, which can benefit all tasks and minimize task-specific biases. The task prediction contribution can be regarded as a weight value, which can quantify the confidence of the shared feature extraction model in predicting the battery task samples. It can achieve personalized adaptation and enhance interpretability. It can guide the battery health status analysis model to identify which shared features are important and which are secondary. Then, the battery health status analysis model processes the target charge-discharge time-series data, shared features, and task prediction contribution. Combining the information from these three aspects, the final battery health status value is output, which greatly improves accuracy. Therefore, determining a high-precision battery health status will also benefit the safety of equipment operation.

[0147] Furthermore, embodiments of this application provide a battery health status analysis device for multi-battery configurations. Optionally, Figure 5 The hardware structure block diagram of the battery health status analysis device under a multi-battery configuration is shown. (Refer to...) Figure 5 The hardware structure of a battery health status analysis device under a multi-battery configuration may include: at least one processor 01, at least one communication interface 02, at least one memory 03, and at least one communication bus 04.

[0148] In this embodiment, the number of processor 01, communication interface 02, memory 03 and communication bus 04 is at least one, and processor 01, communication interface 02 and memory 03 communicate with each other through communication bus 04.

[0149] Processor 01 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention.

[0150] Memory 03 may include high-speed RAM, and may also include non-volatile memory, such as at least one disk storage device.

[0151] The memory stores a program that the processor can call. The program is used to execute the battery health status analysis method under the following multi-battery configuration, including:

[0152] Obtain the target charge / discharge timing data for multi-battery task configuration;

[0153] The target charge-discharge time series data is input into a pre-trained shared feature extraction model to output shared features and task prediction contribution. The shared feature extraction model is trained using real multi-battery charge-discharge data as training samples and real shared features and task prediction contribution of the multi-battery charge-discharge data as sample labels.

[0154] The target charge-discharge time series data, shared features, and task prediction contribution are input into a pre-trained battery health status analysis model to output the battery health status corresponding to each battery task. The battery health status analysis model is trained using real multi-battery charge-discharge data, shared features, and task prediction contribution as training samples, and using the real battery health status of each battery task in the multi-battery charge-discharge data as sample labels.

[0155] Optionally, the refined and extended functions of the program can be found in the description of the battery health status analysis method under multi-battery configuration in the method embodiments.

[0156] This application embodiment also provides a storage medium that can store a program suitable for execution by a processor. When the program runs, it controls the device where the storage medium is located to execute the following battery health status analysis method under a multi-battery configuration, including:

[0157] Obtain the target charge / discharge timing data for multi-battery task configuration;

[0158] The target charge-discharge time series data is input into a pre-trained shared feature extraction model to output shared features and task prediction contribution. The shared feature extraction model is trained using real multi-battery charge-discharge data as training samples and real shared features and task prediction contribution of the multi-battery charge-discharge data as sample labels.

[0159] The target charge-discharge time series data, shared features, and task prediction contribution are input into a pre-trained battery health status analysis model to output the battery health status corresponding to each battery task. The battery health status analysis model is trained using real multi-battery charge-discharge data, shared features, and task prediction contribution as training samples, and using the real battery health status of each battery task in the multi-battery charge-discharge data as sample labels.

[0160] Specifically, the storage medium can be a computer-readable storage medium, which can be an electronic storage device such as flash memory, EEPROM (Electrically Erasable Programmable Read-Only Memory), EPROM, hard disk, or ROM.

[0161] Optionally, the refined and extended functions of the program can be found in the description of the battery health status analysis method under multi-battery configuration in the method embodiments.

[0162] Furthermore, the functional modules in the various embodiments of this disclosure can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part. If the function is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this disclosure, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a live streaming device, or a network device, etc.) to execute all or part of the steps of the methods in the various embodiments of this disclosure.

[0163] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0164] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0165] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for analyzing battery health status in a multi-battery configuration, characterized in that, include: Obtain the target charge / discharge timing data for the multi-battery task configuration; The target charge-discharge time series data is input into a pre-trained shared feature extraction model to output shared features and task prediction contribution. The shared feature extraction model is trained using real multi-battery charge-discharge data as training samples and real shared features and task prediction contribution of the multi-battery charge-discharge data as sample labels. The target charge-discharge time series data, shared features, and task prediction contribution are input into a pre-trained battery health status analysis model to output the battery health status corresponding to each battery task. The battery health status analysis model is trained using real multi-battery charge-discharge data, shared features, and task prediction contribution as training samples, and using the real battery health status of each battery task in the multi-battery charge-discharge data as sample labels.

2. The method according to claim 1, characterized in that, The acquisition of target charge / discharge timing data for multi-battery task configuration includes: Obtain the raw charge / discharge timing data of the multi-battery task configuration; Divide the original charge and discharge timing data of each battery task in the original charge and discharge timing data; For each battery task, the original charge-discharge timing data corresponding to each cycle in the original charge-discharge timing data of the battery task is divided into a separate set of first charge-discharge data. Data filtering is performed on the first charge-discharge data of each group to obtain the target charge-discharge timing data.

3. The method according to claim 2, characterized in that, The first charge-discharge data of each group is filtered to obtain the target charge-discharge time series data, including: For each set of the first charge and discharge data, extract the maximum discharge capacity and the nominal battery capacity of the first set of charge and discharge data; Calculate the actual health status of the battery in this cycle based on the maximum discharge capacity and the battery's nominal capacity. Compare the actual health status of each battery task in each cycle with the preset normal range; The first charge-discharge data corresponding to the actual health state that exceeds the normal range is removed from the original charge-discharge timing data, and the remaining first charge-discharge data are used as the second charge-discharge data. Feature extraction is performed on the second charge-discharge data of each group to obtain the target charge-discharge time series data.

4. The method according to claim 3, characterized in that, The step of extracting features from each group of the second charge-discharge data to obtain target charge-discharge time-series data includes: For each set of the second charge-discharge data, the second charge-discharge data is divided into equal-length intervals according to a preset number to form a multivariate time series; The multivariate time series is interpolated under the conditions of voltage, current, charge capacity, and the differential of charge capacity with respect to voltage, respectively, to obtain the target time series; The target time series corresponding to the second charge and discharge data of each group are fused to obtain the target charge and discharge time series data.

5. The method according to claim 1, characterized in that, The shared feature extraction model includes an initial extraction module, a joint prediction module, a joint identification module, and a first fusion module; The first input terminal of the initial extraction module serves as the input terminal of the shared feature extraction model; The first output of the initial extraction module is connected to the input of the joint prediction module, the second output of the initial extraction module is connected to the input of the joint identification module, and the first output of the joint prediction module and the output of the joint identification module are both connected to the input of the first fusion module. The second output of the joint prediction module and the third output of the initial extraction module both serve as outputs of the shared feature extraction model.

6. The method according to claim 5, characterized in that, The joint authentication module includes a gradient inversion unit and an authentication unit; The input terminal of the gradient inversion unit serves as the input terminal of the joint authentication module, and the output terminal of the gradient inversion unit is connected to the input terminal of the authentication unit; the output terminal of the authentication unit serves as the output terminal of the joint authentication module. The gradient inversion unit inputs the data output by the initial extraction unit to the discrimination unit during the forward propagation process. During the backpropagation process, the gradient inversion unit reverses the data returned by the discrimination unit and outputs it to the initial extraction unit.

7. The method according to any one of claims 1 to 6, characterized in that, The battery health status analysis model includes a weighting module, a dedicated feature extraction module, a second fusion module, and a task prediction module; The input terminal of the weighting module and the first input terminal of the dedicated feature extraction module are both used as input terminals of the battery health status analysis model. The output of the weighting module is connected to the second input of the dedicated feature extraction module and the first input of the task prediction module, respectively. The output of the dedicated feature extraction module is connected to the input of the second fusion module, and the output of the second fusion module is connected to the second input of the task prediction module. The output of the task prediction module serves as the output of the battery health status analysis model.

8. A battery health status analysis device for multi-battery configurations, characterized in that, include: The data acquisition module is used to acquire the target charge and discharge timing data of the multi-battery task configuration; The shared feature extraction module is used to input the target charge-discharge time series data into a pre-trained shared feature extraction model to output shared features and task prediction contribution. The shared feature extraction model is trained using real multi-battery charge-discharge data as training samples and real shared features and task prediction contribution of the multi-battery charge-discharge data as sample labels. The battery health status analysis module is used to input the target charge-discharge time series data, shared features, and task prediction contribution into a pre-trained battery health status analysis model to output the battery health status corresponding to each battery task. The battery health status analysis model is trained using real multi-battery charge-discharge data, shared features, and task prediction contribution as training samples, and using the real battery health status of each battery task in the multi-battery charge-discharge data as sample labels.

9. A battery health status analysis device for multi-battery configurations, characterized in that, Including memory and processor; The memory is used to store programs; The processor is configured to execute the program to implement the various steps of the battery health status analysis method under a multi-battery configuration as described in any one of claims 1-7.

10. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements each step of the battery health status analysis method under a multi-battery configuration as described in any one of claims 1-7.

Images