Lithium ion battery health state prediction method, apparatus and device, and storage medium

Abstract

The invention provides a lithium ion battery health state prediction method, apparatus and device, and a storage medium, and relates to the technical field of lithium ion batteries, and the method comprises the steps of collecting real-time voltage, current and temperature data of a target battery system; inputting the real-time voltage, current and temperature data of the target battery system into a battery health state prediction model, and dynamically predicting the health state of the target battery system; wherein the battery health state prediction model is obtained by performing transfer learning fine tuning on a global pre-training model based on a private data set of the target battery system; the global pre-training model is obtained by performing distributed training through a federated learning framework based on operation data of a plurality of client lithium batteries under different charging strategies, federated learning and transfer learning technologies are combined, battery health prediction in a scene with privacy protection requirements can be processed, and the method and the system have high practicability. Moreover, the problem of data scarcity can be solved, and the real-time evaluation of the health state of the battery in a charging scene is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium-ion batteries, and in particular, to a method, device, equipment and storage medium for predicting the state of health of a lithium-ion battery. Background Art

[0002] Lithium-ion batteries are widely used in fields such as electric vehicles and energy storage systems, and their state of health is crucial for battery life management. In practical applications, due to the difficulty of battery data collection, the prediction of the state of health of batteries usually faces the problem of insufficient data. Traditional model-based methods usually require accurate battery models and a large amount of experimental data. In related technologies, transfer learning is used to optimize model calculation. Although transfer learning shows good application prospects in the case of insufficient data, existing transfer learning methods still have problems such as data privacy restrictions and poor cross-scenario adaptability. Summary of the Invention

[0003] The present invention provides a method, device, equipment and storage medium for predicting the state of health of a lithium-ion battery, so as to solve the defects of the traditional method for predicting the state of health of a lithium-ion battery, which has data privacy restrictions and poor cross-scenario adaptability.

[0004] The present invention provides a method for predicting the state of health of a lithium-ion battery, including: Collecting real-time voltage, current and temperature data of a target battery system; Inputting the real-time voltage, current and temperature data of the target battery system into a battery state of health prediction model to dynamically predict the state of health of the target battery system; Wherein, the battery state of health prediction model is obtained by fine-tuning a global pre-trained model through transfer learning based on a private data set of the target battery system; the global pre-trained model is obtained by distributed training through a federated learning framework based on the operation data of multiple client lithium batteries under different charging strategies.

[0005] According to the method for predicting the state of health of a lithium-ion battery provided by the present invention, the global pre-trained model includes: a feature extraction module and a state of health estimation module; The feature extraction module is used to extract associated features from the operation data of the multiple client lithium batteries under different charging strategies; The state of health estimation module is used to predict the state of health of the battery based on the associated features; When fine-tuning the global pre-trained model through transfer learning, only the parameters of the state of health estimation module are adjusted, and the pre-trained weights of the feature extraction module are retained.

[0006] According to the method for predicting the state of health of a lithium-ion battery provided by the present invention, the feature extraction module includes: a plurality of convolutional layers, a pooling layer, and a self-attention mechanism; The plurality of convolutional layers are used to extract local temporal features of voltage and current signals; The pooling layer is used to compress the local temporal features by max pooling or average pooling; The self-attention mechanism dynamically evaluates the importance weights of each channel where the local temporal features are located, enhances the contribution of key signals sensitive to the prediction of the battery state of health, so as to extract the correlation features between the battery state of health and the charging data.

[0007] According to the method for predicting the state of health of a lithium-ion battery provided by the present invention, the global pre-trained model is obtained by distributed training through a federated learning framework based on the operation data of a plurality of client lithium-ion batteries under different charging strategies, and includes: Constructing a local dataset based on the operation data of its own lithium-ion battery under different charging strategies in each client; Each client trains its own model on the local dataset and uploads it to the central server, so that the central server performs federated learning to obtain a global pre-trained model; the federated learning includes weighted averaging of the model parameters from each client according to a weighted average aggregation strategy to learn common features from multiple clients.

[0008] According to the method for predicting the state of health of a lithium-ion battery provided by the present invention, the constructing a local dataset based on the operation data of its own lithium-ion battery under different charging strategies in each client includes: Collecting the operation data of its own lithium-ion battery under different charging strategies in each client; Performing resampling processing on the operation data of its own lithium-ion battery under different charging strategies in each client, adjusting the number of sample points to ensure that each data point is evenly distributed in the time dimension; Performing maximum-minimum normalization processing on the resampled data, compressing the data values into a unified range to obtain a local dataset.

[0009] According to the method for predicting the state of health of a lithium-ion battery provided by the present invention, the battery state of health prediction model is obtained by fine-tuning the global pre-trained model through transfer learning based on the private dataset of the target battery system, and includes: Fine-tuning the global pre-trained model through transfer learning and performing personalized training on the private dataset of the target battery system; Randomly divide the training dataset of the global pre-trained model into a training set, a validation set, and a test set; when the fine-tuned model achieves the best prediction accuracy on the private dataset, during the training process of the battery health state prediction model, use the validation set to monitor the performance of the model in real time to prevent overfitting, and perform early stopping adjustment according to the loss function of the validation set during the model training process; evaluate the prediction ability of the battery health state prediction model in the real world through the test set, and output the final battery health state prediction model when the prediction ability meets the preset requirements.

[0010] According to the lithium-ion battery health state prediction method provided by the present invention, the private dataset of the target battery system includes: voltage, current, temperature, and capacity data of the target battery system.

[0011] The present invention also provides a lithium-ion battery health state prediction device, including: An acquisition module, configured to acquire real-time voltage, current, and temperature data of the target battery system; A prediction module, configured to input the real-time voltage, current, and temperature data of the target battery system into the battery health state prediction model, and dynamically predict the health state of the target battery system; Wherein, the battery health state prediction model is obtained by performing transfer learning fine-tuning on the global pre-trained model based on the private dataset of the target battery system; the global pre-trained model is obtained by performing distributed training through a federated learning framework based on the operation data of multiple client lithium batteries under different charging strategies.

[0012] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein when the processor executes the program, it implements the lithium-ion battery health state prediction method described in any one of the above.

[0013] The present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the lithium-ion battery health state prediction method described in any one of the above.

[0014] The lithium-ion battery state of health prediction method, device, equipment and storage medium provided by the present invention collect the real-time voltage, current and temperature data of the target battery system; input the real-time voltage, current and temperature data of the target battery system into the battery state of health prediction model to dynamically predict the state of health of the target battery system; wherein, the battery state of health prediction model is obtained by fine-tuning the global pre-trained model through transfer learning based on the private data set of the target battery system; the global pre-trained model is obtained by distributed training through the federated learning framework based on the operation data of multiple client lithium batteries under different charging strategies. The present invention combines federated learning and transfer learning technologies, can handle battery health prediction in scenarios with privacy protection requirements, and can solve the problem of scarce data, realizing real-time evaluation of the state of health of the battery in the charging scenario. Description of the Drawings

[0015] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0016] Figure 1 It is a schematic flowchart of the lithium-ion battery state of health prediction method provided by an embodiment of the present invention; Figure 2 It is a schematic framework diagram of the lithium-ion battery state of health prediction method provided by an embodiment of the present invention; Figure 3 It is a schematic model structure diagram of the battery state of health prediction model provided by an embodiment of the present invention; Figure 4 It is a schematic structure diagram of the feature extraction module provided by an embodiment of the present invention; Figure 5 It is a schematic diagram of the lithium battery state of health prediction result provided by an embodiment of the present invention; Figure 6 It is a schematic structure diagram of the lithium-ion battery state of health prediction device provided by an embodiment of the present invention; Figure 7 It is a schematic functional structure diagram of the electronic device provided by an embodiment of the present invention.

[0017] In the figure, 601, acquisition module; 602, prediction module; 710, processor; 720, communication interface; 730, memory; 740, communication bus. Detailed Embodiments

[0018] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without creative efforts shall fall within the protection scope of the present invention.

[0019] Figure 1 The flowchart of the method for predicting the state of health of a lithium-ion battery provided by an embodiment of the present invention is as Figure 1 shown. The method for predicting the state of health of a lithium-ion battery provided by an embodiment of the present invention includes: Step 101: Collect the real-time voltage, current, and temperature data of the target battery system; Step 102: Input the real-time voltage, current, and temperature data of the target battery system into the battery state of health prediction model to dynamically predict the state of health of the target battery system; Among them, the battery state of health prediction model is obtained by fine-tuning the global pre-trained model through transfer learning based on the private data set of the target battery system; the global pre-trained model is obtained by distributed training through a federated learning framework based on the operation data of multiple client lithium batteries under different charging strategies.

[0020] The method for predicting the state of health of a lithium-ion battery provided by an embodiment of the present invention includes a federated learning stage and a transfer learning stage. In the transfer learning stage, transfer learning is used to fine-tune the source model. As Figure 2 shown, through transfer learning, the pre-trained model obtained by federated learning can be used as the source model and applied to the target battery data set. Since the data sets of target batteries may differ in charging strategies and battery types, transfer learning adjusts the weights of the source model to enable it to better adapt to the specific working conditions of the target battery. This process enables the model to utilize the knowledge learned by the source model from a large amount of data and at the same time make personalized adjustments according to the specific data of the target battery to ensure the prediction accuracy of the model.

[0021] In traditional model calculation optimized by transfer learning, although transfer learning shows good application prospects in the case of insufficient data, existing transfer learning methods still have problems such as data privacy restrictions and poor cross-scenario adaptability.

[0022] The method for predicting the state of health of a lithium-ion battery provided by the embodiments of the present invention collects real-time voltage, current, and temperature data of a target battery system; inputs the real-time voltage, current, and temperature data of the target battery system into a battery state-of-health prediction model to dynamically predict the state of health of the target battery system; wherein, the battery state-of-health prediction model is obtained by fine-tuning a globally pre-trained model through transfer learning based on a private data set of the target battery system; the globally pre-trained model is obtained by distributed training through a federated learning framework based on the operation data of multiple client lithium batteries under different charging strategies. The present invention combines federated learning and transfer learning technologies, can handle battery health prediction in scenarios with privacy protection requirements, and can solve the problem of scarce data, realizing real-time evaluation of the state of health of the battery under charging scenarios.

[0023] Based on any of the above embodiments, as Figure 3 shown, the globally pre-trained model includes: a feature extraction module and a state-of-health estimation module; The feature extraction module is used to extract associated features from the operation data of the multiple client lithium batteries under different charging strategies; The state-of-health estimation module is used to predict the state of health of the battery based on the associated features; When fine-tuning the globally pre-trained model through transfer learning, only the parameters of the state-of-health estimation module are adjusted, and the pre-trained weights of the feature extraction module are retained.

[0024] In the embodiments of the present invention, the specific configuration parameters of the battery state-of-health prediction model are shown in Table 1.

[0025] Table 1 Configuration parameters of the battery state-of-health prediction model

[0026] In the embodiments of the present invention, the feature extraction module includes: a plurality of convolutional layers, pooling layers, and a self-attention mechanism; The plurality of convolutional layers are used to extract local temporal features of voltage and current signals; In the embodiments of the present invention, the multi-convolutional layers include shallow convolutional layers and deep convolutional layers. The shallow convolutional layers identify transient features in the original signal (such as current spikes at the beginning of charging); the deep convolutional layers combine low-level features to detect high-order patterns (such as the capacity decay trend in multiple consecutive cycles). Extract local temporal dependencies related to the state of health from the battery operation data (such as the association between the morphological changes of a certain charging curve and capacity loss).

[0027] The pooling layers are used to compress the local temporal features through max pooling or average pooling; In the embodiment of the present invention, one-dimensional max pooling halves the sequence length and retains the maximum value within each window, such as highlighting abnormal current fluctuations. It gradually compresses non-critical details and focuses on the macroscopic trends that have a significant impact on battery health (such as the relationship between the duration of the temperature rise segment and battery aging).

[0028] The self-attention mechanism dynamically evaluates the importance weights of each channel where local temporal features are located, enhances the contribution of key signals sensitive to the prediction of battery health status, and extracts the correlation features between battery health status and charging data.

[0029] In the embodiment of the present invention, through the lightweight attention mechanism, the importance weights of each channel (i.e., different sensor signals such as voltage, current, and temperature) are dynamically evaluated. It enhances the contribution of key signals such as the voltage inflection point at the end of charging and suppresses irrelevant noise. It captures the synergistic effect of multimodal data (voltage, current, temperature) and automatically identifies the combination of input signals that is most sensitive to the prediction of health status (for example: the change in current at high temperature can reflect the degree of aging better than voltage alone).

[0030] The collaborative work of the multiple convolutional layers, pooling layers, and self-attention mechanism is as follows. The convolutional layer initially extracts local temporal features (such as "the voltage drop rate during a certain charge"); the pooling layer compresses redundant information and retains key patterns (such as "the peak points of capacity decline in 10 consecutive cycles"); the attention mechanism recalibrates the channel weights (such as "the prediction weight of the temperature signal increases under high-temperature working conditions"). After stacking multiple layers, the model finally outputs high-order abstract features (such as "fast charging strategy + temperature fluctuation → non-linear acceleration of capacity decay"). Through this division of labor, the model can accurately locate multi-scale features related to battery health status from complex and high-noise battery data.

[0031] Based on any of the above embodiments, the global pre-trained model is obtained through distributed training in a federated learning framework based on the operation data of multiple client lithium batteries under different charging strategies, including: Construct a local dataset in each client based on the operation data of its own lithium battery under different charging strategies; Each client trains its own model on the local dataset and uploads it to the central server, so that the central server performs federated learning to obtain the global pre-trained model; the federated learning includes weighted averaging of the model parameters from each client according to the weighted average aggregation strategy to learn common features from multiple clients.

[0032] The embodiments of the present invention adopt federated learning technology and perform distributed training based on multiple clients (i.e., datasets under different battery operating conditions). Each client trains its own model on its local data and uploads it to the central server. The central server uses a weighted average aggregation strategy to perform weighted averaging on the model parameters from each client, thereby obtaining a general global model. During this process, the local models of each client do not need to share their sensitive training data, which can effectively protect data privacy and enable the model to learn common features from multiple source data through shared knowledge. After this process, the obtained global model can adapt to different charging strategies and battery types.

[0033] In the embodiments of the present invention, constructing the local dataset based on the operation data of its own lithium battery under different charging strategies in each client includes: Collect the operation data of its own lithium battery under different charging strategies in each client; Perform resampling processing on the operation data of its own lithium battery under different charging strategies in each client, adjust the number of sample points, and ensure that each data point is evenly distributed in the time dimension; Perform maximum-minimum normalization processing on the resampled data, compress the data values to a unified range, and obtain the local dataset.

[0034] In the embodiments of the present invention, a series of preprocessing is performed on the collected raw data to meet the input requirements of the deep learning model. First, by performing resampling processing on the collected data, the number of sample points is adjusted to ensure that each data point is evenly distributed in the time dimension, which can be achieved by methods such as linear interpolation. Second, in order to accelerate the training process and reduce the differences between different data dimensions, all the collected data will be subjected to maximum-minimum normalization processing to compress the data values to a unified range (such as [0, 1]). This enables different battery operation data to better adapt to the deep learning model, especially for the processing of high-dimensional data. In addition, abnormal data also needs to be cleaned to ensure the quality of the input data.

[0035] Based on any of the above embodiments, the battery health state prediction model is obtained by performing transfer learning fine-tuning on the global pre-trained model based on the private dataset of the target battery system, and includes: Perform fine-tuning on the global pre-trained model through transfer learning and perform personalized training on the private dataset of the target battery system; Randomly divide the training dataset of the global pre-trained model into a training set, a validation set, and a test set; when the fine-tuned model achieves the best prediction accuracy on the private dataset, monitor the performance of the model in real time during the training process of the battery health state prediction model through the validation set to prevent overfitting, and perform early stopping adjustment according to the loss function of the validation set during the model training process; evaluate the prediction ability of the battery health state prediction model in the real world through the test set, and output the final battery health state prediction model when the prediction ability meets the preset requirements.

[0036] To ensure the training effect and generalization ability of the model, the preprocessed dataset is divided according to a certain ratio. Usually, the dataset is randomly divided into a training set, a validation set, and a test set in a ratio of 7:1:2. The training set is used to train the deep learning model, and the validation set is used to monitor the performance of the model in real time during the training process to prevent overfitting, and perform early stopping adjustment according to the loss function (such as mean square error) of the validation set during the model training process; the test set is used to finally evaluate the prediction ability of the model in the real world to ensure the reliability and accuracy of the model.

[0037] The fine-tuned model needs to be further verified and evaluated on the target dataset. In this stage, the validation set is used to monitor the performance of the model and evaluate its prediction accuracy. By calculating evaluation metrics such as mean square error and mean absolute error of the model on the validation set, the model is further optimized to ensure its best performance on the target battery dataset. At the same time, the test set is used to comprehensively evaluate the finally trained model to determine its prediction ability and stability in practical applications.

[0038] In the embodiment of the present invention, the private dataset of the target battery system includes: voltage, current, temperature, and capacity data of the target battery system.

[0039] Based on any of the above embodiments, the lithium-ion battery health state prediction method provided by the embodiment of the present invention specifically includes: Step 1: Monitor and collect the operation data of the lithium battery under different fast charging strategies, including voltage, current, temperature, and charging capacity, and establish a training dataset of the battery; Step 2: Preprocess the collected original data, including data resampling and maximum-minimum normalization processing; Step 3: Randomly divide the preprocessed dataset into 7:1:2 according to the ratio of the training set, the validation set, and the test set; Step 4: Construct a lightweight convolutional neural network model (battery health state prediction model), including a feature extraction module and a health state estimation module, for predicting the health state of the lithium battery; Step 5: Conduct federated learning training on the source dataset and obtain the global model using the weighted average aggregation strategy; Step 6: Fine-tune the pre-trained model through transfer learning and conduct personalized training for the target battery dataset; Step 7: Evaluate the performance of the fine-tuned model through the validation set and the test set; Step 8: Based on the trained model, make real-time predictions on the health status of the target battery.

[0040] In the embodiment of the present invention, each client, such as different battery manufacturers / devices, independently trains a lightweight neural network model using local data to learn the correlation features between battery voltage, current, etc. and the battery health status, such as the capacity attenuation pattern and the internal resistance change trend. The central server collects the model parameters uploaded by each client, assigns weights according to the data volume or quality of the client, and performs weighted averaging. For example, the parameter weights of clients with a large data volume are higher. The aggregated global model parameters fuse the common features of multi-source data, such as the capacity attenuation law commonly existing under different charging strategies, and at the same time filter out client-specific noises. Since each client only uploads the model parameters and the original data always remains local, privacy protection is achieved.

[0041] For example, if the data of multiple clients all show that high-temperature fast charging leads to accelerated capacity attenuation, the global model will strengthen this correlation feature.

[0042] In the embodiment of the present invention, during fine-tuning, the feature extraction module is frozen, and the general feature representation learned by federated learning is directly reused without training from scratch, solving the problem of scarce target data. Only the last fully connected layer is fine-tuned to adapt to the specificity of the target battery.

[0043] After the model training is completed and evaluated, real-time predictions are made on the battery health status. By loading the trained deep learning model, parameters such as the voltage, current, and temperature of the target battery can be monitored in real time, and its health status can be predicted based on the real-time operation data of the battery. For example, for a certain lithium iron phosphate material battery, the test conditions are: multi-stage fast charging and constant current discharging at 25°C. Use this federated transfer learning network to predict its health status. The prediction results are as Figure 5 shown, and it can be seen from Figure 5 that the predicted value of this method is highly consistent with the true value.

[0044] The method for predicting the state of health of a lithium-ion battery provided by the embodiments of the present invention collects various monitoring data generated during the rapid charging process of a lithium battery, such as parameters like voltage, current, temperature, and charging capacity, through various charging strategies and battery operating conditions. The collection of these data is to establish a dataset for predicting the state of health of the battery with wide applicability. Specifically, during implementation, charge-discharge experiments on the battery at various fast charging rates are carried out to obtain the operating data of the battery at different fast charging rates. The ultimate goal of this process is to construct a high-quality and diverse training dataset covering the operating conditions of batteries of different clients for predicting the state of health in subsequent steps. The embodiments of the present invention can improve the accuracy of predicting the state of health of the battery while ensuring data privacy and demonstrate excellent adaptability in the rapid charging scenario.

[0045] The lithium-ion battery state-of-health prediction device provided by the present invention will be described below. The lithium-ion battery state-of-health prediction device described below can be mutually referred to in correspondence with the lithium-ion battery state-of-health prediction method described above.

[0046] Figure 6 It is a schematic structural diagram of the lithium-ion battery state-of-health prediction device provided by the embodiments of the present invention, as Figure 6 shown, the lithium-ion battery state-of-health prediction device provided by the embodiments of the present invention includes: An acquisition module 601, configured to acquire real-time voltage, current, and temperature data of a target battery system; A prediction module 602, configured to input the real-time voltage, current, and temperature data of the target battery system into a battery state-of-health prediction model to dynamically predict the state of health of the target battery system; Among them, the battery state-of-health prediction model is obtained by performing transfer learning fine-tuning on a global pre-trained model based on a private dataset of the target battery system; the global pre-trained model is obtained by performing distributed training through a federated learning framework based on the operating data of multiple client lithium batteries under different charging strategies.

[0047] The lithium-ion battery state of health prediction device provided by the embodiment of the present invention collects the real-time voltage, current and temperature data of the target battery system; inputs the real-time voltage, current and temperature data of the target battery system into the battery state of health prediction model to dynamically predict the state of health of the target battery system; wherein, the battery state of health prediction model is obtained by performing transfer learning fine-tuning on the global pre-trained model based on the private data set of the target battery system; the global pre-trained model is obtained by performing distributed training through a federated learning framework based on the operation data of multiple client lithium batteries under different charging strategies. The present invention combines federated learning and transfer learning technologies, can handle battery health prediction in scenarios with privacy protection requirements, and can solve the problem of data scarcity to realize real-time evaluation of the state of health of the battery under the charging scenario.

[0048] Figure 7 An entity structure diagram of an electronic device is exemplified, as Figure 7 shown, the electronic device may include: a processor 710, a communication interface 720, a memory 730 and a communication bus 740. Among them, the processor 710, the communication interface 720, and the memory 730 complete mutual communication through the communication bus 740. The memory 730 includes a computer program, an operating system, and the acquired data. The processor 710 can call the logical instructions in the memory 730 to execute the lithium-ion battery state of health prediction method, and the method includes: collecting the real-time voltage, current and temperature data of the target battery system; inputting the real-time voltage, current and temperature data of the target battery system into the battery state of health prediction model to dynamically predict the state of health of the target battery system; wherein, the battery state of health prediction model is obtained by performing transfer learning fine-tuning on the global pre-trained model based on the private data set of the target battery system; the global pre-trained model is obtained by performing distributed training through a federated learning framework based on the operation data of multiple client lithium batteries under different charging strategies.

[0049] In addition, when the logical instructions in the above-mentioned memory 730 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the related technology, or a part of this 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 for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0050] On the other hand, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is used to implement the lithium-ion battery state of health prediction method provided by the above-mentioned various methods. The method includes: collecting real-time voltage, current, and temperature data of a target battery system; inputting the real-time voltage, current, and temperature data of the target battery system into a battery state of health prediction model to dynamically predict the state of health of the target battery system; wherein, the battery state of health prediction model is obtained by performing transfer learning fine-tuning on a global pre-trained model based on a private data set of the target battery system; the global pre-trained model is obtained by performing distributed training through a federated learning framework based on the operation data of multiple client lithium batteries under different charging strategies.

[0051] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without creative labor.

[0052] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the essence of the above technical solution, or the part that contributes to the related technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0053] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting the health status of a lithium-ion battery, characterized in that: include: Collect real-time voltage, current and temperature data of the target battery system; Inputting the real-time voltage, current and temperature data of the target battery system into a battery health state prediction model to dynamically predict the health state of the target battery system; The battery health status prediction model is obtained by fine-tuning the global pre-trained model through transfer learning based on the private data set of the target battery system; The global pre-training model is obtained by distributed training through a federated learning framework based on the operating data of multiple client lithium batteries under different charging strategies.

2. The method for predicting the health status of a lithium-ion battery according to claim 1, characterized in that: The global pre-training model includes: a feature extraction module and a health status estimation module; The feature extraction module is used to extract relevant features from the operation data of the plurality of client lithium batteries under different charging strategies; The health state estimation module is used to predict the battery health state based on the associated features; When performing transfer learning fine-tuning on the global pre-trained model, only the parameters of the health status estimation module are adjusted, and the pre-trained weights of the feature extraction module are retained.

3. The method for predicting the health status of a lithium-ion battery according to claim 2, characterized in that: The feature extraction module includes: multiple convolutional layers, pooling layers and a self-attention mechanism; The multiple convolutional layers are used to extract local timing features of voltage and current signals; The pooling layer is used to compress the local temporal features through maximum pooling or average pooling; The self-attention mechanism dynamically evaluates the importance weights of each channel where the local time series features are located, enhances the contribution of key signals sensitive to the prediction of battery health status, and extracts the correlation features between the battery health status and the charging data.

4. The method for predicting the health status of a lithium-ion battery according to claim 1, characterized in that: The global pre-training model is obtained through distributed training through a federated learning framework based on the operating data of multiple client lithium batteries under different charging strategies, including: In each client, a local data set is built based on the operating data of its own lithium battery under different charging strategies; Each client trains its own model on the local data set and uploads it to the central server, so that the central server performs federated learning to obtain a global pre-trained model; the federated learning includes weighted averaging the model parameters from each client according to a weighted average aggregation strategy to learn common features from multiple clients.

5. The method for predicting the health status of a lithium-ion battery according to claim 4, characterized in that: The local data set is constructed in each client based on the operation data of its own lithium battery under different charging strategies, including: Collect the operating data of each client's lithium battery under different charging strategies; Resample the operating data of each client based on its own lithium battery under different charging strategies, adjust the number of sample points, and ensure that each data point is evenly distributed in the time dimension; The resampled data is normalized to the maximum and minimum values, and the data values ​​are compressed to a uniform range to obtain a local data set.

6. The method for predicting the health status of a lithium-ion battery according to claim 1, characterized in that: The battery health status prediction model is obtained by fine-tuning the global pre-trained model through transfer learning based on the private data set of the target battery system, and includes: Fine-tune the global pre-trained model through transfer learning, and perform personalized training on a private dataset of the target battery system; The training data set of the global pre-trained model is randomly divided into a training set, a validation set and a test set; when the fine-tuned model achieves the best prediction accuracy on the private data set, the performance of the model is monitored in real time during the training of the battery health state prediction model through the validation set to prevent overfitting, and early stopping adjustment is performed according to the loss function of the validation set during the model training process; the prediction ability of the battery health state prediction model in the real world is evaluated through the test set, and the final battery health state prediction model is output when the prediction ability meets the preset requirements.

7. The method for predicting the health status of a lithium-ion battery according to claim 1 or 6, characterized in that: The private data set of the target battery system includes: voltage, current, temperature and capacity data of the target battery system.

8. A lithium-ion battery health status prediction device, characterized in that: include: Acquisition module, used to collect real-time voltage, current and temperature data of the target battery system; A prediction module, used to input the real-time voltage, current and temperature data of the target battery system into a battery health state prediction model to dynamically predict the health state of the target battery system; The battery health status prediction model is obtained by fine-tuning the global pre-trained model through transfer learning based on the private data set of the target battery system; The global pre-training model is obtained by distributed training through a federated learning framework based on the operating data of multiple client lithium batteries under different charging strategies.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the lithium-ion battery health status prediction method as described in any one of claims 1 to 7 is implemented.

10. A non-transitory readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for predicting the health status of a lithium-ion battery as claimed in any one of claims 1 to 7 is implemented.

Images