Lithium battery health state prediction method and system fusing user behaviors and physical information

Abstract

The invention belongs to but not limited to the technical field of battery health, and particularly relates to a lithium battery health state prediction method and system fusing user behaviors and physical information, and the method comprises the steps: S1, feature extraction and data processing; s2, a Kolmogorov-Arnold network optimized by an Attention mechanism is constructed, and S2, the Kolmogorov-Arnold network is constructed; s3, designing a physical information embedded neural network; and S4, performing multi-task joint optimization and prediction. The Attention-KAN-PINN model used in the method is obviously superior to other comparison models in the aspect of mean absolute error (MAE) index, the error distribution is more concentrated, and the stability is higher. In terms of average absolute percentage error (MAPE) indexes, the Attention-KAN-PINN model is also excellent in performance and minimum in distribution range, and the average value and the standard deviation are obviously lower than those of other models. In addition, the model shows remarkable advantages in the aspect of a root mean square error (RMSE) index, and the robustness of the model to abnormal values is further proved.

Description

Technical Field

[0001] The present invention belongs to but is not limited to the field of battery health technology, and in particular relates to a lithium battery health status prediction method and system that integrates user behavior and physical information. Background Art

[0002] As the core technology of modern energy storage and conversion, lithium-ion batteries have been widely used in consumer electronics, automobiles, energy storage systems and other fields. However, the degradation of their health status, including capacity decay and shortened life, has brought severe challenges to practical applications. Therefore, how to accurately predict the health status of lithium batteries has become a hot topic of research at home and abroad. At present, the research on lithium battery SOH prediction mainly focuses on two directions:

[0003] The first is a data-driven machine learning method. With the rapid development of artificial intelligence technology, researchers have tried to use machine learning models such as support vector machines, random forests, and deep neural networks to build a nonlinear mapping relationship between battery health status and usage conditions by learning from historical cycle data.

[0004] The second is the simulation method based on physical models. The physical model describes the behavior of lithium batteries through electrochemical principles and predicts the SOH of lithium batteries, such as the Doyle-Fuller-Newman (DFN) model and the Pseudo-two-Dimensional Model (P2D) model based on electrochemical thermodynamics, which provide theoretical support for the prediction of battery health status.

[0005] Although data-driven methods and physical model methods have their own advantages, it is difficult for a single technical solution to take into account both high efficiency and physical consistency at the same time. The data-driven method mainly relies on a large amount of historical data. Although it performs well in prediction accuracy and model flexibility, it is difficult to effectively integrate the physical laws inside the battery, resulting in insufficient robustness and reliability of the model under extreme conditions. At the same time, due to the neglect of modeling the battery degradation mechanism, the prediction results lack physical meaning, which limits its promotion in practical applications.

[0006] On the other hand, the physical model method can provide prediction results with high physical consistency by combining the theoretical electrochemical behavior of lithium batteries. However, such methods usually involve solving complex differential equations with high computational complexity. Not only is it difficult to meet real-time requirements, but it also places high demands on hardware computing power. In addition, the physical model has limited modeling capabilities for user behavior data and cannot flexibly reflect the dynamic impact of user charging patterns, usage frequency, etc. on battery health status, thus limiting its applicability in complex application scenarios.

[0007] In existing research, there are few attempts to deeply integrate user behavior data with battery physical information, and the model therefore lacks the ability to optimize the entire link from user to battery. This fragmented approach means that the potential value of user behavior features has not been fully explored, and accurate prediction of battery health status cannot be achieved. Therefore, there is an urgent need for a new method that can effectively combine user behavior data with battery physical models to improve prediction accuracy while enhancing the reliability and computational efficiency of the model, providing a full-link solution for battery health management. This has laid a solid research background and technical foundation for the proposal of this invention.

[0008] In view of the above analysis, the technical problems that urgently need to be solved in the existing technology are: there is currently a lack of a new method that can effectively combine user behavior data with battery physical models to improve prediction accuracy while enhancing the reliability and computing efficiency of the model, providing a full-link solution for battery health management. Summary of the invention

[0009] In view of the problems existing in the prior art, the present invention provides a lithium battery health status prediction method and system that integrates user behavior and physical information, innovatively combining user behavior data, physical modeling and deep learning to significantly improve the accuracy and reliability of SOH prediction. The user data set used in the experiment contains data from 124 lithium battery charge and discharge processes, including voltage, current, temperature and other characteristics.

[0010] The present invention is implemented as follows: a lithium battery health status prediction method integrating user behavior and physical information, comprising:

[0011] S1. Feature extraction and data processing;

[0012] S2. Construction of Kolmogorov-Arnold network (KAN) with optimized Attention mechanism;

[0013] S3. Design of Physical Information Neural Network (PINN);

[0014] S4. Multi-task joint optimization and prediction.

[0015] Furthermore, S1 specifically includes: The discharge process of the battery varies depending on the user's behavior. In contrast, the charging process is more fixed and regular. Users rarely fully discharge the battery when using it, and it is more common to charge it to full power. Therefore, 16 health indicators related to the time, voltage, current, temperature, internal resistance and charging power during the charging process are selected as the input features of the model. These feature data are processed by the 3σ principle, normalized in the [-1,1] interval, and finally input into the model.

[0016] Furthermore, S2 specifically includes: introducing the Attention mechanism, giving the KAN model the ability to dynamically assign weights to key user behavior features, so that the model can capture the deep correlation between user usage habits and changes in battery health status. Through the KAN model, user behavior features are effectively combined with physical information to achieve multi-dimensional mapping and prediction of battery health status.

[0017] Furthermore, S3 specifically includes: in model training, the physical laws of battery degradation are explicitly embedded into the loss function as physical constraints to ensure the consistency of the prediction results with the battery operation mechanism. The introduction of PINN overcomes the lack of physical consistency of traditional data-driven methods and improves the accuracy and reliability of the model.

[0018] Furthermore, S4 specifically includes: optimizing prediction capabilities through a multi-task learning framework, and designing joint loss functions, such as partial differential equation loss, data term loss, physical loss, and L2 regularization loss, to balance the relationship between user behavior, physical information, and model SOH prediction accuracy.

[0019] Another object of the present invention is to provide a lithium battery health status prediction system integrating user behavior and physical information to implement the lithium battery health status prediction method integrating user behavior and physical information, comprising:

[0020] Data processing module, used for feature extraction and data processing;

[0021] Network building module, used to build Kolmogorov-Arnold network (KAN) for optimizing the Attention mechanism;

[0022] Neural network design module, used to design physical information embedded neural network (PINN);

[0023] The joint optimization and prediction module is used to perform multi-task joint optimization and prediction.

[0024] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows:

[0025] First, the Attention-KAN-PINN model used in the present invention performs significantly better than other comparative models in terms of mean absolute error (MAE) indicators, with more concentrated error distribution and higher stability. The Attention-KAN-PINN model introduces an attention mechanism that can identify and increase the weights of key features, thereby more accurately capturing important information in the battery degradation process. At the same time, the Attention-KAN-PINN model combines the powerful high-dimensional complex function approximation capability of Kolmogorov-Arnold Networks (KAN) with the physical constraints of the physical information neural network (PINN). This combination not only improves the data fitting capability, but also enhances the physical consistency of the model, which better compensates for the problem that traditional deep learning models are not good at simulating battery degradation trends. Experimental results show that the KAN-PINN model based on the attention mechanism performs well in tracking battery degradation trajectories and is superior to other common models in prediction accuracy. Specifically, the mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE) of KAN-PINN reached 0.0056, 0.0069, and 0.5949%, respectively. In addition, the coefficient of determination (R 2 ) exceeds 0.94, which further verifies the efficiency and reliability of the model in processing complex battery degradation data.

[0026] Second, as auxiliary evidence of the inventiveness of the claims of the present invention, it is also reflected in the following important aspects:

[0027] First of all, the technical solution of the present invention can improve the accuracy and robustness of lithium battery SOH prediction after conversion. By combining user behavior data characteristics, KAN based on attention mechanism and physical information neural network, this method can comprehensively and accurately capture the key features that affect the battery health state and fit the dynamic attenuation of battery SOH. Compared with the traditional single data-driven or physical modeling method, the present invention significantly improves the prediction accuracy of the model and its adaptability to complex usage scenarios. In addition, high-precision SOH prediction can extend the battery life and reduce the situation where the battery is replaced prematurely due to misjudgment of health status, thereby reducing user costs and improving the market competitiveness of battery manufacturers.

[0028] Secondly, it can reduce the risk of battery degradation and improve user experience. By accurately predicting the battery health status, the present invention can help users optimize charging and discharging behaviors, avoid accelerated battery degradation due to improper operation, and thus maximize the actual service life of the battery. For electric vehicles, energy storage equipment and other fields, good user behavior optimization suggestions will significantly improve product reliability and user satisfaction, thereby enhancing brand loyalty and market reputation.

[0029] Finally, it can support the optimization and upgrading of the intelligent battery management system (BMS). By integrating the prediction method of the present invention into the battery management system (BMS), the battery health status can be monitored and predicted in real time, providing intelligent support for battery management. Based on the accurately predicted health status information, the BMS can optimize the charging strategy, thermal management, and usage recommendations. This function can help companies develop smarter and more reliable battery management systems, and further promote technological innovation and industrial upgrading in the fields of new energy vehicles and energy storage systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 is a flow chart of a lithium battery health status prediction method integrating user behavior and physical information provided by an embodiment of the present invention;

[0031] Figure 2 This is a structural diagram of a lithium battery health status prediction system that integrates user behavior and physical information provided by an embodiment of the present invention;

[0032] Figure 3 It is a violin plot of the experimental results of the multi-layer perceptron (MLP), convolutional neural network (CNN), long short-term memory network (LSTM) and the proposed Attention-KAN-PINN on the user data set provided by the embodiment of the present invention;

[0033] Figure 4 It is a graph of SOH estimation results of a multi-layer perceptron (MLP), a convolutional neural network (CNN), a long short-term memory network (LSTM) and the proposed Attention-KAN-PINN on a user data set provided by an embodiment of the present invention;

[0034] Figure 5 The figure shows the fit between the prediction results and the true values ​​of the three models KAN, Attention-KAN and Attention-KAN-PINN on the user dataset;

[0035] Figure 6 The error distribution graphs of the three models KAN, Attention-KAN, and Attention-KAN-PINN are shown;

[0036] Figure 7 Distribution plots of mean absolute error (MAE), mean absolute percentage error (RMSE), and root mean square error (MAPE) of the proposed Attention-KAN-PINN model, ordinary Kolmogorov-Arnold Networks (KAN), and Attention-KAN models;

[0037] Figure 8It is a detailed flow chart of a lithium battery health status prediction method integrating user behavior and physical information provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0039] like Figure 1 As shown, the lithium battery health status prediction method based on user behavior and physical information provided by the present invention includes:

[0040] S1. Feature extraction and data processing;

[0041] S2. Construction of Kolmogorov-Arnold network (KAN) with optimized Attention mechanism;

[0042] S3. Design of Physical Information Neural Network (PINN);

[0043] S4. Multi-task joint optimization and prediction.

[0044] In step S1, features are extracted from user behavior data and battery physical sensor data. User behavior data includes information such as charging mode, discharging mode, and usage frequency, while physical sensor data includes key parameters such as voltage, current, temperature, and internal resistance. Data consistency and validity are ensured through data processing techniques such as normalization, denoising, and dimensionality reduction. At the same time, time series data is segmented for subsequent neural network training.

[0045] In step S2, the Kolmogorov-Arnold network (KAN) is used to model the high-dimensional nonlinear relationship of the battery state. In order to improve the prediction ability of the network, the Attention mechanism is introduced to highlight the most important features for predicting the battery health status by dynamically assigning weights. For example, weights will be assigned to core features such as the battery's internal resistance change and temperature anomalies, thereby enhancing the network's ability to perceive key states.

[0046] In step S3, a physical information neural network (PINN) is constructed by combining physical constraint information. The physical behavior of the battery is embedded into the neural network through the battery equivalent circuit model and thermodynamic model. The physical model provides theoretical constraints on parameters such as voltage, current and temperature, so that the network is not only data-driven, but also follows physical laws, thereby improving the reliability and generalization ability of the prediction results.

[0047] In step S4, a multi-task learning framework is used to jointly optimize the prediction tasks of battery health status (such as remaining capacity and internal resistance) and battery life. By setting different loss function weights, the network can simultaneously learn short-term status (such as current capacity) and long-term trends (such as life prediction). This joint optimization strategy ensures that the prediction model performs optimally in multiple dimensions.

[0048] In the model training phase, user behavior and physical information data are used to train KAN and PINN respectively, and the outputs of the two networks are fused. The model performance is evaluated by cross-validation technology, and indicators such as mean square error (MSE) and mean absolute error (MAE) are used to evaluate the prediction accuracy. Experimental results show that the introduction of the Attention mechanism and physical constraints significantly improves the accuracy and stability of the model.

[0049] After the model is deployed, it is combined with an embedded system to achieve real-time prediction of the battery status. When new data enters the system, the model automatically updates parameters through the online learning module to adapt to the dynamic changes of the battery. The prediction results can be provided to users for battery maintenance and uploaded to the cloud through the Internet of Things technology to provide support for large-scale battery status monitoring.

[0050] like Figure 8 As shown in the figure, in the initial stage of the method, the user behavior data (such as charging and discharging behavior, temperature fluctuation, etc.) and physical information (such as battery capacity, internal resistance, number of cycles, etc.) are feature extracted and preprocessed. For user behavior data, the time series analysis method is used to extract the characteristics of the charging and discharging mode; for physical information, the nonlinear characteristics of battery performance changes are captured through methods such as wavelet transform. After normalization, the feature data is uniformly mapped to the same feature space to eliminate the scale differences between different data sources and provide high-quality input for subsequent model training.

[0051] During the model building phase, the Attention mechanism is introduced to optimize the Kolmogorov-Arnold network. The Kolmogorov-Arnold network has good nonlinear mapping capabilities and can fit the relationship between complex user behavior and battery performance. The Attention mechanism is used to weight input features and automatically focus on key features that have an important impact on the prediction of battery health status. By dynamically adjusting feature weights, the model can more efficiently capture the nonlinear interactive relationship between user behavior and physical information, improving the accuracy and robustness of predictions.

[0052] In order to combine the physical mechanism information of the battery, a physical information embedding neural network module is designed to embed the physical parameters of the battery (such as internal resistance and voltage change rate) into the hidden nodes of the network. This embedding method avoids the overfitting problem caused by the model's complete reliance on data-driven learning by adding physical constraints. The physical embedding module also enhances the network's ability to characterize the battery's health status by simulating a mathematical model of the battery's aging and loss process, making the prediction results closer to the actual working mechanism of the battery.

[0053] In the prediction stage, a multi-task learning framework is used to jointly optimize battery capacity decay, health status assessment, and remaining life prediction as multiple subtasks. Through shared feature representation, the model shares information between different tasks, improving the generalization performance of the overall prediction. During the joint optimization process, the training balance of each subtask is coordinated through a weighted loss function to ensure the accuracy of health status prediction, while providing accurate remaining life and capacity degradation trend prediction results. Ultimately, this method achieves efficient prediction of lithium battery health status by integrating user behavior and physical information.

[0054] S1 specifically includes: The discharge process of the battery varies depending on the user's behavior. In contrast, the charging process is more fixed and regular. Users rarely fully discharge the battery when using it, but it is more common to charge it to full power. Therefore, the time, voltage, current, temperature, internal resistance and charging capacity during the charging process are selected as the input features of the model. After the feature data is processed by the 3σ principle, it is normalized in the [-1,1] interval.

[0055] S2 specifically includes: introducing the Attention mechanism, giving the KAN model the ability to dynamically assign weights to key user behavior features, so that the model can capture the deep correlation between usage habits and changes in battery health status. Through the KAN model, user behavior features are effectively combined with physical information to achieve multi-dimensional mapping and prediction of battery health status.

[0056] S3 specifically includes: In model training, the physical laws of the battery are explicitly embedded into the loss function as physical constraints to ensure the consistency of the prediction results with the battery operation mechanism. The introduction of PINN overcomes the black box nature of traditional data-driven methods and improves the robustness and reliability of the model.

[0057] S4 specifically includes: optimizing prediction capabilities through a multi-task learning framework, and designing joint loss functions, such as partial differential equation loss, data term loss, physical loss, and L2 regularization loss, to balance the relationship between user behavior, physical information, and model SOH prediction accuracy.

[0058] like Figure 2As shown, another object of the present invention is to provide a lithium battery health status prediction system integrating user behavior and physical information to implement the lithium battery health status prediction method integrating user behavior and physical information, comprising:

[0059] Data processing module, used for feature extraction and data processing;

[0060] Network building module, used to build Kolmogorov-Arnold network (KAN) for optimizing the Attention mechanism;

[0061] Neural network design module, used to design physical information embedded neural network (PINN);

[0062] The joint optimization and prediction module is used to perform multi-task joint optimization and prediction.

[0063] The data processing module is responsible for collecting multi-dimensional data of lithium batteries, including user behavior data (such as charging and discharging modes, temperature changes) and physical information (such as internal resistance, voltage, current, etc.). Through the feature extraction algorithm, the original data is reduced in dimension, and noise and outliers are removed to ensure high quality and consistency of the data. After feature extraction, feature vectors are generated for subsequent model training so that the system can accurately capture the health status characteristics of lithium batteries.

[0064] The network building module uses the Kolmogorov-Arnold network (KAN) optimized by the Attention mechanism. KAN dynamically assigns weights to different features through the Attention mechanism, allowing the model to better focus on key features. Specifically, the Attention mechanism calculates the feature importance score based on the weight distribution of user behavior and physical information, thereby improving the sensitivity to key indicators of lithium battery health status and optimizing prediction accuracy.

[0065] The neural network design module achieves deep learning and modeling of physical information through physical information embedding neural network (PINN). PINN embeds electrochemical knowledge (such as lithium ion diffusion equation and battery aging model) into the neural network structure as prior information to enhance the model's ability to understand the internal physical mechanism of lithium batteries. The embedding process effectively combines theoretical models and data-driven models by adding a physical constraint layer to the neural network.

[0066] The joint optimization and prediction module incorporates lithium battery life prediction and health status assessment into the optimization objectives through multi-task learning methods. The multi-task loss function is used to coordinate the weight balance of user behavior data and physical information, so that the model can simultaneously predict the remaining useful life (RUL) and health status (SOH). During the optimization process, the system dynamically adjusts the loss weight to ensure the balance between tasks, thereby improving the overall prediction effect.

[0067] The system adopts a phased training method, first training the KAN and PINN models separately, and then jointly training them. By introducing an online learning mechanism, the system can dynamically correct the model using real-time data. The dynamic correction process fine-tunes the model parameters based on the latest user behavior and physical information data to ensure that the prediction results are always consistent with the actual health status of the lithium battery.

[0068] The prediction results are output through multi-dimensional indicators (such as battery capacity loss rate, remaining life, etc.). The system also supports visual display of the results, providing content including health status trend charts, feature importance analysis, and prediction accuracy evaluation. Users can formulate battery usage optimization strategies or replace batteries in time based on the prediction results, thereby extending the service life of lithium batteries and improving their efficiency.

[0069] This system achieves high-precision prediction of the health status of lithium batteries by combining user behavior data with physical information. The KAN optimized by the Attention mechanism enhances the feature learning ability, the PINN embedding physical constraints improves the interpretability of the model, and the multi-task joint optimization ensures the comprehensiveness and robustness of the prediction, providing reliable technical support for lithium battery life management.

[0070] In order to verify the technical effect of the Attention-KAN-PINN method based on the fusion of user behavior and physical information proposed in the present invention in the prediction of the health status of lithium batteries, a comparative experiment of multi-layer perceptron (MLP), convolutional neural network (CNN), Kolmogorov-Arnold network (KAN) and Attention-KAN-PINN was designed. Among them, the comparative models (MLP, CNN, KAN) selected in the experiment cover the basic neural network structures and optimization models commonly used in data-driven methods, and can comprehensively evaluate the performance improvement of the Attention-KAN-PINN model. On the user data set, analysis is performed from the aspects of prediction accuracy and error distribution. The experimental results are shown in Table 1. Figure 3The error distribution of each model under the three error indicators is shown through violin plots, and the mean line and standard deviation line are attached for comparative analysis. Among them, MAE is used to measure the mean absolute value of the prediction error, MAPE evaluates the percentage of the prediction error relative to the actual value, and RMSE reflects the sensitivity of the model to large errors. It is an important indicator for measuring the robustness of outliers. The Attention-KAN-PINN model performs significantly better than other comparison models in the mean absolute error (MAE) indicator, with a more concentrated error distribution and higher stability. In terms of the mean absolute percentage error (MAPE) indicator, the Attention-KAN-PINN model also performs well, with the smallest distribution range, and the mean and standard deviation are significantly lower than other models. In addition, the model shows significant advantages in the root mean square error (RMSE) indicator, further proving its robustness to outliers. The SOH estimation results of the four models are shown in the figure. Figure 4 shown.

[0071] Table 1 Experimental results of multi-layer perceptron (MLP), convolutional neural network (CNN), Kolmogorov-Arnold network (KAN) and the proposed Attention-KAN-PINN on user datasets.

[0072]

[0073] In order to reflect the stability and reliability of the model, the training and testing process of each model was repeated 10 times. The above table shows the average of the 10 results.

[0074] Example 1: Electric vehicle lithium battery health status monitoring and prediction system

[0075] In the actual application of electric vehicles, the lithium battery health status prediction method of the present invention is used to achieve accurate monitoring of battery performance and prediction of remaining life. The system collects user behavior data (such as charging frequency, discharge depth, voltage curve) and physical information of the battery (such as internal resistance, capacity, number of cycles) through the battery management system (BMS), and pre-processes the data and inputs it into the Kolmogorov-Arnold network optimized by the Attention mechanism. The network can efficiently integrate user behavior and physical information, and at the same time, by embedding the physical model of battery aging, it further enhances the ability to predict changes in battery health status.

[0076] The system uses a multi-task learning framework to achieve joint prediction of battery capacity decline, state of health (SOH) and remaining life (RUL), providing users with real-time battery health reports and range estimates. At the same time, the prediction results can be used to optimize charging strategies and extend battery life. For example, when the prediction results show that the battery is about to reach the health threshold, the system can remind users to replace the battery module or adjust driving habits in time to avoid the risk of range reduction and failure due to battery aging.

[0077] Example 2: Energy Storage Power Station Lithium Battery Pack Health Status Management System

[0078] In energy storage power station applications, the lithium battery health status prediction method of the present invention is used to monitor battery pack performance in real time to ensure the safety and economy of power station operation. By collecting physical data such as the total current, voltage, temperature fluctuations of the battery pack, as well as user load behavior (such as discharge rate and frequency), combined with the prediction model of the present invention, accurate prediction of the health status and remaining life of the battery pack can be achieved. The Attention mechanism helps the system focus on key influencing factors, such as temperature and number of cycles, and the prediction of the health status assessment (SOH) and capacity degradation trend is more stable.

[0079] The prediction results are applied to the operation optimization and maintenance decision-making of energy storage power stations. For example, the system can reasonably adjust the load distribution and charge and discharge scheduling plan based on the health status prediction results to avoid further damage to the aging battery packs due to high load operation. At the same time, by predicting the failed battery packs, maintenance work can be arranged in advance to avoid the overall performance degradation of the system or sudden failures. This method reduces the operation risk of energy storage power stations, reduces maintenance costs, and at the same time extends the service life of battery packs, improving the economic benefits and safety of power station operation.

[0080] Relevant evidence of the technical effects achieved by the embodiments of the present invention.

[0081] In order to verify the effectiveness of KAN and its combined attention mechanism, the present invention uses the ordinary Kolmogorov-Arnold network (KAN) model, the Attention-KAN model (Kolmogorov-Arnold network combined with the attention mechanism) and the Attention-KAN-PINN model to predict the battery SOH on the user data set. The experimental results are shown in Table 2.

[0082] Table 2 Experimental results of ordinary Kolmogorov-Arnold Networks (KAN), Attention-KAN and the proposed Attention-KAN-PINN model on the user dataset.

[0083]

[0084] Note: The best results in the table are in bold. In order to reflect the stability and reliability of the model, the training and testing process of each model was repeated 10 times. The above table is the average of the 10 experimental results.

[0085] Figure 5 The fitting between the prediction results and the true values ​​of the three models KAN, Attention-KAN and Attention-KAN-PINN on the user data set is demonstrated. As can be seen from the figure, the prediction curve of the Attention-KAN-PINN model proposed in the present invention has the best fitting effect with the true value, and can accurately capture the degradation trend of the lithium battery state of health (SOH). In contrast, due to the lack of physical constraints, the Attention-KAN model has improved the focus on key features to a certain extent, but there are still errors in the extreme value area of ​​​​SOH. The basic model KAN has a weak learning ability in complex degradation behaviors, resulting in a large deviation between the overall curve and the true value, and the worst fitting effect.

[0086] Figure 5 SOH prediction results. Results of the proposed Attention-KAN-PINN model, ordinary Kolmogorov-ArnoldNetworks (KAN), and Attention-KAN on the user dataset. The predicted SOH and the true SOH are distributed near the diagonal, indicating that the model performs well.

[0087] Figure 6 The error distribution of the three models of KAN, Attention-KAN and Attention-KAN-PINN is shown. As can be seen from the figure, the Attention-KAN-PINN model proposed in the present invention has the smallest error range, and the error distribution is highly concentrated within 2%. In comparison, the error distribution of the Attention-KAN model is more dispersed. Although it shows a certain concentration in the central area, there are still large fluctuations in the extreme error values. The error distribution of the KAN model is the most discrete, not only with a wider error range, but also with more areas deviating from zero. This shows that by introducing the Attention mechanism, the model can better focus on the key features of the degradation process; and after combining with the PINN physical constraints, the model further optimizes the ability to capture global features, thereby significantly reducing the prediction error.

[0088] Figure 6 Error distribution plot. The x-axis represents the error between the predicted value and the true value, and the y-axis represents the number of errors.

[0089] Figure 7The performance of the three models, KAN, Attention-KAN and Attention-KAN-PINN, was quantitatively evaluated, including evaluation indicators such as MAE, MAPE and RMSE. As can be seen from the figure, the Attention-KAN-PINN model proposed in this invention achieves the best performance in all indicators.

[0090] Figure 7 Distribution of mean absolute error (MAE), mean absolute percentage error (RMSE) and root mean square error (MAPE) of the proposed Attention-KAN-PINN model, ordinary Kolmogorov-Arnold Networks (KAN) and Attention-KAN models, with mean and standard deviation lines marked. It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. A person of ordinary skill in the art will understand that the above-mentioned devices and methods can be implemented using computer executable instructions and / or contained in a processor control code, for example, such code is provided on a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The device and its modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., and can also be implemented by software executed by various types of processors, or by a combination of the above hardware circuits and software, such as firmware.

[0091] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principle of the present invention should be covered by the protection scope of the present invention.

Claims

1. A lithium battery health status prediction method integrating user behavior and physical information, characterized in that: include: S1. Feature extraction and data processing: extract features from user behavior data and battery physical sensor data, where user behavior data includes charging mode, discharging mode and usage frequency, and physical sensor data includes voltage, current, temperature and internal resistance; perform normalization, denoising and time series segmentation processing on the data; S2. Construction of Kolmogorov-Arnold network (KAN) optimized by attention mechanism: Based on the high-dimensional nonlinear relationship of battery health status, a Kolmogorov-Arnold network (KAN) is constructed, and weights are dynamically assigned through the attention mechanism to highlight the impact of key features; S3. Design of Physical Information Neural Network (PINN): Combining the battery equivalent circuit model and thermodynamic model, a physical information neural network (PINN) is constructed to improve the reliability and generalization ability of the model by embedding physical behavior constraints; S4. Multi-task joint optimization and prediction: A multi-task learning framework is used to jointly optimize the battery health status prediction task, including remaining capacity, internal resistance, and battery life; different loss function weights are used to balance the learning objectives of short-term status and long-term trends; S5. Model training and validation: The KAN and PINN networks are trained by cross-validation techniques, and the outputs of the two networks are fused; the performance of the prediction model is evaluated using mean square error and mean absolute error; S6. Real-time prediction and dynamic update: Combined with the embedded system, the battery health status is predicted in real time. When new data is input, the model parameters are dynamically updated through the online learning module, and the prediction results are uploaded to the cloud for large-scale monitoring.

2. The lithium battery health status prediction method integrating user behavior and physical information as claimed in claim 1, characterized in that: The step S1 specifically includes: During the charging process, time, voltage, current, temperature, internal resistance and charging capacity are extracted as input features. After the feature data are processed according to the 3σ principle to remove outliers, they are further normalized to the [-1,1] interval to improve the stability and adaptability of the model.

3. The lithium battery health status prediction method integrating user behavior and physical information as claimed in claim 1, characterized in that: The step S2 specifically includes: By introducing the Attention mechanism and dynamically assigning weights to user behavior features, the KAN model can focus on key features and capture the deep correlation between user usage behavior and changes in battery health status, thereby achieving high-precision mapping and prediction of battery health status.

4. The lithium battery health status prediction method integrating user behavior and physical information as claimed in claim 1, characterized in that: The step S3 specifically includes: During the model training process, the physical laws of the battery are embedded in the loss function as constraints to ensure that the model prediction results are consistent with the physical mechanism of battery operation, while improving the physical consistency and generalization ability of the model.

5. The lithium battery health status prediction method integrating user behavior and physical information as claimed in claim 1, characterized in that: The step S4 specifically includes: The prediction ability is optimized through a multi-task learning framework, and a joint loss function is designed. This loss function weighs the relationship between user behavior, physical information and prediction accuracy to simultaneously improve the prediction performance of the battery's remaining capacity, internal resistance and life.

6. The lithium battery health status prediction method integrating user behavior and physical information as claimed in claim 1, characterized in that: The multi-task learning framework is based on a joint optimization strategy to dynamically adjust the weights between short-term health status prediction and long-term trend prediction to ensure the adaptability and accuracy of the prediction model.

7. A lithium battery health status prediction system integrating user behavior and physical information to implement the lithium battery health status prediction method integrating user behavior and physical information as claimed in any one of claims 1 to 6, characterized in that: include: Data processing module, used for feature extraction and data processing; Network construction module, used to construct Kolmogorov-Arnold network for Attention mechanism optimization; Neural network design module, used to design physical information embedded in neural network; The joint optimization and prediction module is used to perform multi-task joint optimization and prediction.

8. A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the lithium battery health status prediction method integrating user behavior and physical information as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to execute the steps of the lithium battery health status prediction method integrating user behavior and physical information as described in any one of claims 1 to 6.

10. An information data processing terminal, comprising the lithium battery health status prediction system integrating user behavior and physical information as claimed in claim 7.

Images