Lithium battery residual life analysis method, apparatus and device, and storage medium

Abstract

The invention discloses a lithium battery residual life analysis method, device and equipment and a storage medium, and relates to the technical field of lithium battery management and control, and the method comprises the steps: obtaining a first battery capacity sequence corresponding to a source domain battery and a second battery capacity sequence corresponding to a target domain battery; carrying out capacity regeneration detection on the first battery capacity sequence and the second battery capacity sequence, and determining a current transfer learning scene according to a detection result; based on the current migration learning scene, performing adaptive migration prediction on the first battery capacity sequence and the second battery capacity sequence to obtain final capacity prediction data corresponding to the target domain battery; and determining a residual service life prediction result corresponding to the target domain battery according to the capacity failure threshold value and the capacity prediction data corresponding to the target domain battery. Through capacity regeneration detection and scene type division, adaptive transfer learning in different capacity regeneration scenes is realized, capacity regeneration interference is effectively inhibited, and the prediction precision of the battery transfer learning residual life is improved.

Description

Technical Field

[0001] This application relates to the field of lithium battery management technology, and in particular to a method, apparatus, device and storage medium for analyzing the remaining life of a lithium battery. Background Technology

[0002] Lithium-ion batteries have been widely used in new energy vehicles and energy storage systems. Transfer learning, which can effectively utilize small sample data to improve prediction performance, has been widely used in the field of battery remaining life prediction.

[0003] However, in actual use, lithium-ion batteries inevitably experience capacity regeneration, which leads to a decrease in the capacity knowledge transfer effect of the transfer learning model, ultimately affecting the accuracy of remaining life prediction.

[0004] Therefore, how to suppress interference from capacity regeneration and improve the accuracy of lithium battery transfer learning remaining lifetime prediction has become an urgent problem to be solved. Summary of the Invention

[0005] The main objective of this application is to provide a method, apparatus, device, and storage medium for analyzing the remaining life of lithium batteries, aiming to solve the technical problem of how to suppress interference from capacity regeneration and improve the accuracy of lithium battery transfer learning remaining life prediction.

[0006] To achieve the above objectives, this application proposes a method for analyzing the remaining life of lithium batteries, the method comprising: Obtain the first battery capacity sequence corresponding to the source domain battery and the second battery capacity sequence corresponding to the target domain battery; Capacity regeneration detection is performed on the first battery capacity sequence and the second battery capacity sequence, and the current transfer learning scenario is determined based on the detection results; Based on the current transfer learning scenario, adaptive transfer prediction is performed on the first battery capacity sequence and the second battery capacity sequence to obtain the final capacity prediction data corresponding to the target domain battery. The remaining service life prediction result of the target domain battery is determined based on the capacity failure threshold corresponding to the target domain battery and the capacity prediction data.

[0007] In one embodiment, the step of performing capacity regeneration detection on the first battery capacity sequence and the second battery capacity sequence, and determining the current transfer learning scenario based on the detection result, includes: The first capacity regeneration feature identification result is determined based on the capacity decay characteristics of the first battery capacity sequence. Based on the capacity decay characteristics of the second battery capacity sequence, the identification result of the second capacity regeneration feature is determined; Data matching is performed on the preset transfer learning scenario type, the first capacity regeneration feature recognition result, and the second capacity regeneration feature recognition result to obtain the current transfer learning scenario; the preset transfer learning scenario type includes: both the source domain and the target domain contain capacity regeneration phenomena, the source domain contains capacity regeneration phenomena and the target domain does not contain capacity regeneration phenomena, and the source domain does not contain capacity regeneration phenomena and the target domain contains capacity regeneration phenomena.

[0008] In one embodiment, the step of performing adaptive transfer prediction on the first battery capacity sequence and the second battery capacity sequence based on the current transfer learning scenario to obtain the final capacity prediction data corresponding to the target domain battery includes: Based on the current transfer learning scenario, variational mode decomposition is performed on the first battery capacity sequence and / or the second battery capacity sequence to obtain a set of mode components; Adaptive transfer learning training is performed based on the modal component set to obtain the final capacity prediction data corresponding to the target domain battery.

[0009] In one embodiment, the modal component set includes: a source domain modal component set and a target domain modal component set; The step of performing variational mode decomposition on the first battery capacity sequence and / or the second battery capacity sequence based on the current transfer learning scenario to obtain a set of mode components includes: In the current transfer learning scenario where both the source domain and the target domain contain capacity regeneration phenomena, variational mode decomposition is performed on the first battery capacity sequence to obtain the source domain mode component set, which contains multiple intrinsic mode function components. Variational mode decomposition is performed on the second battery capacity sequence to obtain the target domain mode component set, which contains the same number of intrinsic mode function components as the source domain mode component set; Accordingly, the step of performing adaptive transfer learning training based on the modal component set to obtain the final capacity prediction data corresponding to the target domain battery includes: Based on each intrinsic mode function component in the source domain modal component set, a corresponding number of bidirectional long short-term memory neural network sub-models are trained, and the capacity prediction of each intrinsic mode function component in the target domain modal component set is performed by the bidirectional long short-term memory neural network sub-models to obtain the model prediction sub-results corresponding to the target domain battery. The model prediction results are fused to generate the final capacity prediction data corresponding to the target domain battery.

[0010] In one embodiment, the modal component set includes: a source domain modal component set; The step of performing variational mode decomposition on the first battery capacity sequence and / or the second battery capacity sequence based on the current transfer learning scenario to obtain a set of mode components includes: In the case where the current transfer learning scenario is that the source domain contains capacity regeneration and the target domain does not contain capacity regeneration, variational mode decomposition is performed on the first battery capacity sequence to obtain the source domain mode component set, which contains multiple intrinsic mode function components. Accordingly, the step of performing adaptive transfer learning training based on the modal component set to obtain the final capacity prediction data corresponding to the target domain battery includes: Calculate the correlation coefficient between each intrinsic mode function component in the source domain modal component set and the second battery capacity sequence; Based on the correlation coefficient, the intrinsic mode function component with the highest correlation is selected from the source domain modal component set as the transfer feature component; The bidirectional long short-term memory neural network model is trained based on the aforementioned transfer feature components to obtain a target capacity prediction model; The target capacity prediction model is used to predict the capacity of the second battery capacity sequence to obtain the final capacity prediction data corresponding to the target domain battery.

[0011] In one embodiment, the modal component set includes: a target domain modal component set; The step of performing variational mode decomposition on the first battery capacity sequence and / or the second battery capacity sequence based on the current transfer learning scenario to obtain a set of mode components includes: In the current transfer learning scenario where the source domain does not contain capacity regeneration phenomena and the target domain does contain capacity regeneration phenomena, variational mode decomposition is performed on the second battery capacity sequence to obtain the target domain mode component set, which contains multiple intrinsic mode function components. Accordingly, the step of performing adaptive transfer learning training based on the modal component set to obtain the final capacity prediction data corresponding to the target domain battery includes: The bidirectional long short-term memory neural network model is trained based on the first battery capacity sequence to obtain the target capacity prediction model. The target capacity prediction model is then used to predict the capacity of each intrinsic mode function component in the target domain modal component set to obtain the model prediction sub-result corresponding to the target domain battery. The model prediction results are fused to generate the final capacity prediction data corresponding to the target domain battery.

[0012] In one embodiment, before obtaining the first battery capacity sequence corresponding to the source domain battery and the second battery capacity sequence corresponding to the target domain battery, the method further includes: Obtain the initial battery capacity sequence corresponding to the source domain battery; Perform a translation transformation on the initial battery capacity sequence to obtain enhanced samples; The initial battery capacity sequence and the enhanced sample are combined into a first battery capacity sequence.

[0013] Furthermore, to achieve the above objectives, this application also proposes a lithium battery remaining life analysis device, which includes: The data acquisition module is used to acquire the first battery capacity sequence corresponding to the source domain battery and the second battery capacity sequence corresponding to the target domain battery. The scenario analysis module is used to perform capacity regeneration detection on the first battery capacity sequence and the second battery capacity sequence, and determine the current transfer learning scenario based on the detection results; The capacity prediction module is used to perform adaptive transfer prediction on the first battery capacity sequence and the second battery capacity sequence based on the current transfer learning scenario, so as to obtain the final capacity prediction data corresponding to the target domain battery. The remaining life assessment module is used to determine the remaining life prediction result of the target domain battery based on the capacity failure threshold corresponding to the target domain battery and the capacity prediction data.

[0014] In addition, to achieve the above objectives, this application also proposes a lithium battery remaining life analysis device, the device including: a memory, a processor, and a lithium battery remaining life analysis program stored in the memory and executable on the processor, the lithium battery remaining life analysis program being configured to implement the steps of the lithium battery remaining life analysis method as described above.

[0015] In addition, to achieve the above objectives, this application also provides a storage medium, which is a computer-readable storage medium, on which a program for implementing a lithium battery remaining life analysis method is stored. The program for implementing the lithium battery remaining life analysis method is executed by a processor to implement the steps of the lithium battery remaining life analysis method as described above.

[0016] This application provides a method, apparatus, device, and storage medium for analyzing the remaining life of lithium batteries. The method includes: acquiring a first battery capacity sequence corresponding to a source domain battery and a second battery capacity sequence corresponding to a target domain battery; performing capacity regeneration detection on the first and second battery capacity sequences, and determining the current transfer learning scenario based on the detection results; performing adaptive transfer prediction on the first and second battery capacity sequences based on the current transfer learning scenario to obtain the final capacity prediction data corresponding to the target domain battery; and determining the remaining life prediction result corresponding to the target domain battery based on the capacity failure threshold and the capacity prediction data. This application constructs an adaptive transfer learning framework through capacity regeneration phenomenon detection and scenario type classification, realizing adaptive transfer learning under different capacity regeneration scenarios, thereby effectively suppressing capacity regeneration interference, improving cross-domain prediction accuracy and robustness, and further improving the accuracy of battery transfer learning remaining life prediction based on the optimized capacity prediction data. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the first embodiment of the lithium battery remaining life analysis method of this application; Figure 2 This is a schematic diagram of the sample generation process in the first embodiment of the lithium battery remaining life analysis method of this application; Figure 3 This is a flowchart illustrating the second embodiment of the lithium battery remaining life analysis method of this application; Figure 4 This is a schematic diagram of the energy difference curve of the second embodiment of the lithium battery remaining life analysis method of this application; Figure 5 This is a schematic diagram of the first transfer learning process in the second embodiment of the lithium battery remaining life analysis method of this application; Figure 6 This is a schematic diagram of the second transfer learning process in the second embodiment of the lithium battery remaining life analysis method of this application; Figure 7 This is a schematic diagram of the third transfer learning process in the second embodiment of the lithium battery remaining life analysis method of this application; Figure 8This is a schematic diagram of the module structure of the lithium battery remaining life analysis device according to an embodiment of this application; Figure 9 This is a schematic diagram of the equipment structure of the hardware operating environment involved in the lithium battery remaining life analysis method in the embodiments of this application.

[0020] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0021] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0022] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0023] The main solution of this application is as follows: obtaining the first battery capacity sequence corresponding to the source domain battery and the second battery capacity sequence corresponding to the target domain battery; performing capacity regeneration detection on the first battery capacity sequence and the second battery capacity sequence, and determining the current transfer learning scenario based on the detection results; performing adaptive transfer prediction on the first battery capacity sequence and the second battery capacity sequence based on the current transfer learning scenario to obtain the final capacity prediction data corresponding to the target domain battery; and determining the remaining service life prediction result corresponding to the target domain battery based on the capacity failure threshold and the capacity prediction data.

[0024] Currently, during actual charge-discharge cycles, lithium-ion batteries often experience a brief capacity recovery phenomenon, known as capacity regeneration, after a certain period of rest. This phenomenon manifests as a local peak within the overall capacity decay trend. Since the core of transfer learning is to achieve knowledge transfer through the distributional correlation between the source and target domains, capacity regeneration can disrupt this correlation, leading to a significant increase in prediction error. Therefore, within the transfer learning framework, capacity regeneration is not only a disruptive factor in battery life prediction but may also trigger a "domain shift" problem—a mismatch in feature distribution between the source and target domains—directly affecting the model's generalization ability and prediction reliability. To ensure the stability and accuracy of the transfer learning model under different capacity regeneration scenarios, effective handling of the capacity regeneration phenomenon is essential.

[0025] To address this issue, this application constructs an adaptive transfer learning framework by detecting capacity regeneration phenomena and classifying scenario types. This framework enables adaptive transfer learning under different capacity regeneration scenarios, thereby effectively suppressing capacity regeneration interference, improving cross-domain prediction accuracy and robustness, and further enhancing battery transfer learning lifetime prediction accuracy based on the optimized capacity prediction results.

[0026] It should be noted that the executing entity in this embodiment can be a lithium battery remaining life analysis system, or a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or a lithium battery remaining life analysis device capable of performing the above functions. This embodiment does not specifically limit it in this way. The following uses a lithium battery remaining life analysis device (hereinafter referred to as the analysis device) as the executing entity to describe this embodiment and the following embodiments.

[0027] Based on this, embodiments of this application provide a method for analyzing the remaining life of a lithium battery, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the lithium battery remaining life analysis method of this application.

[0028] In this embodiment, the lithium battery remaining life analysis method includes steps S10~S40: Step S10: Obtain the first battery capacity sequence corresponding to the source domain battery and the second battery capacity sequence corresponding to the target domain battery; It is important to understand that the aforementioned first battery capacity sequence can be a set of capacity data recorded by the source domain battery during charge-discharge cycles, containing capacity values ​​at different cycle numbers. This first battery capacity sequence can come from publicly available datasets or measured data, such as the NASA Battery Dataset (NASA Prognostics Center of Excellence Data Repository-Battery DataSet) or the CALCE Battery Dataset (CALCE Battery Research Group Battery Data), for example, the capacity sequence of the B0006 battery [2.0, 1.98, 1.95, ..., 1.4] (unit: Ah). Correspondingly, the aforementioned second battery capacity sequence can be a set of capacity data from the target domain battery, with the same format as the first battery capacity sequence, and is the core object for remaining lifetime prediction.

[0029] Step S20: Perform capacity regeneration detection on the first battery capacity sequence and the second battery capacity sequence, and determine the current transfer learning scenario based on the detection results; It is easy to understand that during the data preprocessing stage, the analysis device can use the built-in capacity detection algorithm to identify whether there is a "brief capacity recovery" phenomenon in the first and second battery capacity sequences, that is, to perform the above-mentioned capacity regeneration detection. The identification basis is to judge whether there is a local peak in the capacity decay trend.

[0030] At this point, the current transfer learning scenario can be the type of transfer learning scenario determined based on the capacity regeneration detection results of the current source domain and target domain. The analysis device can dynamically adapt the scenario features of different transfer learning scenario types through the real-time capacity regeneration detection results, and then adopt different transfer learning strategies for different transfer learning scenarios to predict the battery capacity of the target domain in the subsequent process.

[0031] In one feasible implementation, step S20 may include steps S21 to S23: Step S21: Determine the first capacity regeneration feature identification result based on the capacity decay characteristics of the first battery capacity sequence; Step S22: Based on the capacity decay characteristics of the second battery capacity sequence, determine the second capacity regeneration feature identification result; It should be noted that the aforementioned capacity decay characteristics can be the law governing the change of battery capacity with the number of cycles. When the battery capacity sequence contains capacity regeneration phenomena, it can be characterized by "overall decay + local peaks," while without them, it can be characterized by stable and monotonic decay. Therefore, the aforementioned first capacity regeneration feature identification result can be a detection conclusion of the first battery capacity sequence, which can include two categories: "the source domain contains capacity regeneration phenomena" or "the source domain does not contain capacity regeneration phenomena." Similarly, the aforementioned second capacity regeneration feature identification result can be a detection conclusion of the second battery capacity sequence, and its format is consistent with the first identification result, which can include two results: "the target domain contains capacity regeneration phenomena" or "the target domain does not contain capacity regeneration phenomena."

[0032] For example, this embodiment can use a capacity regeneration detection algorithm of "first-order difference + capacity health index (CHI)" to analyze the capacity decay characteristics of the first battery capacity sequence and the second battery capacity sequence. The capacity regeneration detection algorithm first determines the first-order difference Δx_i (Δx_i=x_i-x_i-1, i=2, 3, ..., n) corresponding to each capacity data in the first and second battery capacity sequences, and determines the capacity threshold by the formula CHI_n=min(0, μ_n+3σ); when the difference change Δx_n+1 of the (n+1)th capacity data at any time is detected to be greater than the capacity threshold CHI_n, it can be determined that there is a capacity regeneration phenomenon.

[0033] Step S23: Perform data matching on the preset transfer learning scenario type, the first capacity regeneration feature recognition result, and the second capacity regeneration feature recognition result to obtain the current transfer learning scenario; the preset transfer learning scenario type includes: both the source domain and the target domain contain capacity regeneration phenomena, the source domain contains capacity regeneration phenomena and the target domain does not contain capacity regeneration phenomena, and the source domain does not contain capacity regeneration phenomena and the target domain contains capacity regeneration phenomena.

[0034] It is easy to understand that the aforementioned preset transfer learning scenario types can be three predefined types of transfer learning scenarios, covering all combinations of capacity regeneration features in the source and target domains. Specifically, these include scenarios where both the source and target domains contain capacity regeneration, the source domain contains capacity regeneration but the target domain does not, and the source domain does not contain capacity regeneration but the target domain does. In this case, the analysis device can compare the first and second capacity regeneration feature identification results with the three preset scenarios to determine the uniquely matching transfer learning scenario, thereby identifying the current transfer learning scenario.

[0035] Step S30: Based on the current transfer learning scenario, perform adaptive transfer prediction on the first battery capacity sequence and the second battery capacity sequence to obtain the final capacity prediction data corresponding to the target domain battery; It is easy to understand that the aforementioned final capacity prediction data can be the set of capacity prediction values ​​for the target domain battery corresponding to the future cycle number output by the transfer learning model. In this embodiment, to address the data distribution differences between the source and target domains caused by capacity regeneration, a transfer learning framework based on the Bidirectional Long Short-Term Memory (BiLSTM) model can be constructed as the aforementioned transfer learning model. The advantage of the BiLSTM model is that it can simultaneously capture the forward and backward dependencies of time-series data, which better reflects the dynamic decay characteristics of battery capacity.

[0036] It should be understood that, in order to convert battery capacity data into samples suitable for supervised learning, the first battery capacity sequence and the second battery capacity sequence mentioned above in this embodiment can be generated by segmenting the original capacity data using a sliding window method to maintain the correlation between adjacent capacities.

[0037] Furthermore, to meet the needs of engineering applications and considering both the long-term trend and short-term fluctuations of battery capacity degradation, this embodiment proposes a 7-step prediction strategy based on a sliding time window. This strategy predicts the capacity value for the next 7 cycles based on the current window data, capturing both medium- and long-term degradation patterns while maintaining prediction accuracy. The sample generation process can be found in [reference needed]. Figure 2 , Figure 2 This is a schematic diagram illustrating the sample generation process of the first embodiment of the lithium battery remaining life analysis method of this application. Figure 2 As shown, each sample can consist of a fixed-length input window (i.e., X_t-n to X_t-1) and capacity values ​​for the next 7 time steps (i.e., X_t-n to X_t+6).

[0038] It is important to understand that, compared to single-step prediction methods, the seven-step prediction method proposed in this embodiment may result in a reduction in the amount of source domain data used during the prediction process, especially when the dataset is small. Insufficient data may affect prediction accuracy. Specifically, the seven-step prediction method involves more time steps in each prediction, and the use of source domain data decreases as the number of prediction steps increases. This reduction in data may lead to a decrease in the prediction accuracy of the transfer learning model.

[0039] Therefore, in one feasible implementation, in this embodiment, steps A1 to A3 may be included before step S10: Step A1: Obtain the initial battery capacity sequence corresponding to the source domain battery; Step A2: Perform a translation transformation on the initial battery capacity sequence to obtain enhanced samples; Step A3: Combine the initial battery capacity sequence and the enhanced sample into a first battery capacity sequence.

[0040] It is understandable that the initial battery capacity sequence mentioned above can be the original capacity sequence of the source domain batteries without data augmentation, i.e., the original data source of the first battery capacity sequence. In this embodiment, data augmentation is required on the initial battery capacity sequence to ensure model accuracy. Available data augmentation methods include traditional transformation methods, i.e., generating new samples through linear transformations (such as translation, scaling, rotation, etc.); and data generation methods based on generative models, typically represented by variational autoencoders (VAEs) and generative adversarial networks (GANs).

[0041] Considering the complexity and stability requirements of the experiment, this embodiment chooses to use translation transformation, a traditional transformation method, for data augmentation. This method not only preserves the statistical properties of the original data but also effectively expands the sample size with relatively low computational cost.

[0042] For example, the analysis device can generate a new capacity sequence, namely the aforementioned enhanced sample, by adding a fixed offset δ (e.g., δ = 0.05 Ah, ensuring that the capacity value of the enhanced sample is within a reasonable range and does not exceed the battery's rated capacity and failure threshold) to each data point of the initial battery capacity sequence. The enhanced sample generated after this translational transformation process follows the same decay trend as the initial battery capacity sequence, with only an overall offset of δ. This effectively expands the sample size of the source domain without affecting the model's prediction trend.

[0043] Finally, the analysis device can merge the initial battery capacity sequence with all augmented samples to form an expanded first battery capacity sequence, which can be used for subsequent transfer learning training. Meanwhile, this embodiment can be experimentally verified to show that the RMSE (Root Mean Square Error) of the model after data augmentation can be reduced by 15%-30%, especially in small sample scenarios, making the seven-step prediction accuracy proposed in this embodiment reach the level of single-step prediction.

[0044] In this embodiment, to enhance the feature sharing between the source and target domains, a virtual capacity sequence consistent with the distribution of the source domain is generated by combining data augmentation methods, thereby strengthening the source domain sample foundation for transfer learning and further improving the prediction accuracy and stability of transfer learning.

[0045] Step S40: Determine the remaining service life prediction result of the target domain battery based on the capacity failure threshold corresponding to the target domain battery and the capacity prediction data.

[0046] It is easy to understand that the above-mentioned capacity failure threshold can be the critical value at which the target domain battery is judged to be in failure, usually 80% of the rated capacity (e.g., for a battery with a rated capacity of 2.0Ah, the failure threshold can be 1.4Ah). Its specific value can be adjusted according to the actual application scenario.

[0047] At this point, the predicted Remaining Useful Life (RUL) can be determined by analyzing the number of cycles required for the target domain battery to reach the capacity failure threshold from its current cycle count. For example, the analysis device can iterate through the capacity prediction data, find the cycle count N_pre corresponding to the first drop to the capacity failure threshold, and record the current cycle count N_now of the target domain battery. Accordingly, the predicted Remaining Useful Life can be: RUL = N_pre - N_now.

[0048] In this embodiment, existing transfer learning methods for predicting the remaining life of lithium batteries do not consider the mismatch between the source and target domains (domain shift) caused by capacity regeneration, resulting in decreased knowledge transfer effectiveness and low prediction accuracy. This embodiment addresses this issue by using capacity regeneration detection and scenario classification to accurately match transfer strategies, thereby suppressing the interference of capacity regeneration on transfer learning at its source. Furthermore, by fully utilizing the general features of the source domain data and the personalized features of the target domain data through adaptive transfer prediction, the accuracy of capacity prediction is significantly improved. Simultaneously, this embodiment ensures the objectivity and reliability of the remaining life prediction results based on the quantitative calculation of capacity prediction data and failure thresholds, providing an accurate reference for the safe use of batteries.

[0049] Furthermore, this embodiment can also rapidly expand the source domain sample size by using translation transformation while preserving the original sequence decay trend, so that the enhanced first battery capacity sequence contains more variant samples, effectively solving the data scarcity problem of seven-step prediction and improving the stability and accuracy of migration prediction.

[0050] This embodiment provides a method for analyzing the remaining life of lithium batteries. The method includes: acquiring a first battery capacity sequence corresponding to a source domain battery and a second battery capacity sequence corresponding to a target domain battery; performing capacity regeneration detection on the first and second battery capacity sequences, and determining the current transfer learning scenario based on the detection results; performing adaptive transfer prediction on the first and second battery capacity sequences based on the current transfer learning scenario to obtain the final capacity prediction data corresponding to the target domain battery; and determining the remaining life prediction result of the target domain battery based on the capacity failure threshold and the capacity prediction data. This application constructs an adaptive transfer learning framework through capacity regeneration phenomenon detection and scenario type classification, realizing adaptive transfer learning under different capacity regeneration scenarios. This effectively suppresses capacity regeneration interference, improves cross-domain prediction accuracy and robustness, and further improves the accuracy of battery transfer learning remaining life prediction based on the optimized capacity prediction data results.

[0051] Based on the first embodiment of this application, in the second embodiment of this application, the same or similar content as the first embodiment described above can be referred to the above description, and will not be repeated hereafter.

[0052] As the above analysis shows, to address the feature differences in different scenarios, this paper proposes an adaptive transfer learning framework based on the capacity regeneration phenomenon. Furthermore, this embodiment can design differentiated variational mode decomposition (VMD) and transfer strategies for different scenarios based on this framework, realizing a gradual optimization process from component-specific transfer to feature selection transfer, and then to target domain-specific transfer.

[0053] Therefore, based on the first embodiment, please refer to Figure 3 , Figure 3 This is a flowchart illustrating the second embodiment of the lithium battery remaining life analysis method of this application. In this embodiment, step S30 includes steps S31 to S32: Step S31: Based on the current transfer learning scenario, perform variational mode decomposition on the first battery capacity sequence and / or the second battery capacity sequence to obtain a set of mode components; Step S32: Perform adaptive transfer learning training based on the modal component set to obtain the final capacity prediction data corresponding to the target domain battery.

[0054] It should be understood that the aforementioned modal component set can be a set of multiple intrinsic mode function (IMF) components obtained by variational mode decomposition of source domain data and / or target domain data. Each component corresponds to a different frequency component of the original capacity sequence. For example, variational mode decomposition of the first battery capacity sequence corresponding to the source domain can yield the source domain modal component set, and variational mode decomposition of the second battery capacity sequence corresponding to the target domain can yield the target domain modal component set.

[0055] Among them, the variational mode decomposition mentioned above is an adaptive signal decomposition method. By iteratively optimizing and solving the variational model, it decomposes the non-stationary signal into multiple stationary IMF components. The core of the variational mode decomposition process is to determine the optimal number of mode components K.

[0056] Taking the B0006 battery, which exhibits significant capacity regeneration, as an example, the key to VMD decomposition lies in determining the optimal number of mode components K. If the value of K is too small, important information in the signal may be lost; if the value of K is too large, frequency aliasing may occur. In this case, this embodiment can select the optimal number of mode components K (i.e., the integer k in formula (1)) through the energy difference index. The specific formula for the energy difference is as follows: (1) (2) (3) In the formula, This is the sum of the energies of all modal components when the number of modal components is k; This is the sum of the energies of all modal components when the number of modal components is k-1; This represents the energy difference. To the first under mode decomposition The i-th sampling point of a component sequence; n is the number of sampling points.

[0057] At this time, refer to Figure 4 It can be seen that, Figure 4 This is a schematic diagram of the energy difference curve of the second embodiment of the lithium battery remaining life analysis method of this application. Figure 4 The horizontal axis represents the number of modal components K, and the vertical axis represents the energy difference. .like Figure 4 As shown, the energy difference curve reaches an inflection point when K is 4; however, the energy difference increases further when K is increased to 5. This indicates that when K=5, spurious modal components may be generated during the decomposition process, affecting the decomposition effect. Therefore, K=4 can be selected as the optimal number of modal components for the B0006 battery.

[0058] Therefore, in this embodiment, the analysis device can use the energy difference method to calculate the total energy of the modal components under different K values, find the K corresponding to the inflection point of the energy difference as the optimal value, and then perform VMD decomposition on the capacity sequence based on the optimal K value to obtain the modal component set (the optimal modal components corresponding to different battery types may be different, and the appropriate optimal modal components can be selected for modal decomposition according to the selected energy difference index or actual situation).

[0059] It is easy to understand that during the adaptive transfer learning training process described above, the analysis device can dynamically adjust the model training strategy (such as training data and network parameter update rules) based on the current scene characteristics and BiLSTM as the base model, and transfer the knowledge learned by the source domain model to the target domain for capacity prediction analysis of the target domain.

[0060] For example, in this embodiment, it can be pre-set that each BiLSTM (sub) model is a three-layer network structure when training the BiLSTM (sub) model, and a "parameter fine-tuning" strategy is adopted during transfer: that is, the first two layers of network parameters of each BiLSTM (sub) model trained according to the first battery capacity sequence corresponding to the source domain are locked (in order to retain the core knowledge of the source domain), and only a small amount of data (such as 20%) in the second battery capacity sequence corresponding to the target domain is used to fine-tune the output layer of the pre-trained model (in order to adapt to the distribution of the target domain), and then the capacity prediction of the remaining data in the target domain is performed based on the adapted BiLSTM (sub) model.

[0061] It should be noted that this embodiment can adopt different variational mode decomposition strategies according to different transfer learning scenarios, and adjust the training data of the model according to the scenario differences (training of all components of source domain data, training of source domain data with selected features, and training of source domain data components separately).

[0062] In the first feasible implementation, the modal component set includes a source domain modal component set and a target domain modal component set. In this embodiment, step S31 may include steps B1~B2: Step B1: In the case that the current transfer learning scenario contains capacity regeneration phenomena in both the source domain and the target domain, variational mode decomposition is performed on the first battery capacity sequence to obtain the source domain mode component set, which contains multiple intrinsic mode function components. Step B2: Perform variational mode decomposition on the second battery capacity sequence to obtain the target domain mode component set, wherein the target domain mode component set contains the same number of intrinsic mode function components as the source domain mode component set; Accordingly, in this embodiment, step S32 may include steps C1~C2: Step C1: Train a corresponding number of bidirectional long short-term memory neural network sub-models for each intrinsic mode function component in the source domain modal component set, and use the bidirectional long short-term memory neural network sub-models to predict the capacity of each intrinsic mode function component in the target domain modal component set, thereby obtaining the model prediction sub-results corresponding to the target domain battery. Step C2: The model prediction sub-results are fused to generate the final capacity prediction data corresponding to the target domain battery.

[0063] It should be noted that when both the source and target domain battery data exhibit capacity regeneration, the capacity sequence data displays nonlinear and non-stationary characteristics, making transfer learning methods susceptible to regeneration fluctuations. Therefore, this embodiment introduces VMD decomposition into both the source and target domains to preprocess the data. By decomposing the complex capacity sequence data into several stationary mode functions (IMFs), the regeneration characteristics and the primary degradation trend characteristics are separated. In this case, the aforementioned intrinsic mode function components can be stationary signal components obtained through VMD decomposition, each with independent frequency characteristics, which can separate the primary degradation trend and capacity regeneration fluctuations in the original capacity sequence.

[0064] Understandably, the aforementioned bidirectional long short-term memory neural network sub-model can be a BiLSTM sub-model trained separately for each IMF component corresponding to the source domain data. The structure is consistent with the basic BiLSTM model mentioned above, and the only difference between the bidirectional long short-term memory neural network sub-models is the training data and the target of adaptation.

[0065] Accordingly, the aforementioned model prediction sub-results can be the capacity prediction values ​​of the target domain battery corresponding to the IMF component output by a single BiLSTM sub-model, and are components of the final capacity prediction data. At this point, the analysis device can add the multiple model prediction sub-results element-wise, i.e., perform the aforementioned fusion process, to obtain capacity prediction data that integrates the features of all IMF components.

[0066] It should be noted that, typically, the optimal K value for adaptation can be determined using the energy difference method for the capacity sequences of the source and target domains respectively. However, in this embodiment, it is preferable to use the same number of modal components in both the source and target domains to ensure better matching and correspondence between the data in the source and target domains. This avoids data inconsistency caused by different numbers of modal components, ensuring smoother data transformation and alignment during the migration process, thereby improving the model's migration performance. Therefore, in this embodiment, the modal component set of the target domain contains the same number of intrinsic mode function components as the modal component set of the source domain.

[0067] Furthermore, in the transfer learning stage of the scenario where "both the source and target domains contain capacity regeneration phenomena," this implementation proposes a "component-specific transfer" strategy. Specifically, this implementation trains a corresponding BiLSTM sub-model for each IMF component in the source domain modal component set and performs transfer learning. Then, it locks the first two layers of the trained base BiLSTM sub-model and fine-tunes the output layer of the base BiLSTM sub-model using a small amount of data from the target domain modal component set corresponding to the IMF component. The remaining target domain data is then transferred based on the fine-tuned sub-models. Finally, the transfer prediction results of each IMF component, i.e., the prediction sub-results of the above models, are summed to obtain the final capacity prediction data. Each BiLSTM sub-model is trained independently, using the same network structure and training parameters (e.g., epochs=100, batch_size=32) to ensure consistency in the training process. Furthermore, each BiLSTM sub-model uses the same sliding window parameters when predicting corresponding components to ensure time alignment of the prediction results.

[0068] Furthermore, this embodiment can employ a parameter fine-tuning strategy during the aforementioned transfer learning process, referring to... Figure 5 It can be seen that, Figure 5 This is a schematic diagram of the first transfer learning process in the second embodiment of the lithium battery remaining life analysis method of this application. In this embodiment, after constructing a BiLSTM basic model based on different modal components in the source domain data, the parameters of the first two layers of the BiLSTM basic model (i.e., ...) are locked. Figure 5 The first two BiLSTM layers shown are fine-tuned using only a small amount of data from the target domain in the output layer to adapt to the data distribution of the target domain, and finally output the migration prediction results.

[0069] In this process, locking the first two layers has two main functions: First, locking the first two layers can avoid repeatedly training these layers in the target domain, thereby saving computing resources and time and improving transfer efficiency; Second, locking the first two layers helps to retain the core knowledge learned from the source domain, preventing the model from forgetting this knowledge during fine-tuning with a small amount of target domain data and avoiding catastrophic forgetting.

[0070] For example, refer to Figure 5 To illustrate, in a current transfer learning scenario where the source domain exhibits capacity regeneration but the target domain does not, the capacity prediction process based on transfer learning in this embodiment may include the following steps: 1) Source and target domain VMD decomposition: Input source domain data, i.e., the capacity sequence of the first battery mentioned above (e.g., the capacity data of B0006), and set candidate values ​​for K from 1 to 10; then calculate the total energy of the modal components corresponding to each K value. and energy difference The optimal K=4 is determined based on the inflection point of the energy difference, and VMD decomposition is performed based on the optimal K to obtain the source domain modal component set {IMF_S1, IMF_S2, IMF_S3, IMF_S4} (which can correspond to respectively). Figure 5 IMF1~IMF4 were generated after the source domain data was decomposed.

[0071] Then, for the target domain data, i.e., the second battery capacity sequence mentioned above (e.g., the capacity data of B0005, the battery type corresponding to the target domain can be the same as or different from the source domain), VMD decomposition is performed based on the determined optimal K to obtain the target domain modal component set {IMF_T1, IMF_T2, IMF_T3, IMF_T4} (which can correspond to respectively) Figure 5 The target domain data is decomposed into IMF1~IMF4 to ensure that the number of intrinsic mode function components contained in the target domain modal component set is consistent with that in the source domain modal component set.

[0072] 2) Training of the bidirectional long short-term memory neural network sub-model: For each IMF component (e.g., IMF_S1~IMF_S4) in the source domain modal component set, the samples are divided and expanded using the sliding window method (window length 8, prediction steps 7) to generate enhanced samples, and the training set and validation set are divided according to a certain ratio (e.g., 8:2).

[0073] Then, four identical BiLSTM sub-models are constructed (each sub-model may include a three-layer network: two BiLSTM layers and one output layer). Sub-model 1 is trained using IMF_S1, sub-model 2 is trained using IMF_S2, and so on. During model training, the RMSE of the validation set is monitored simultaneously. When the RMSE does not decrease for 10 consecutive training epochs, model training is stopped, and the parameters of each sub-model are saved.

[0074] 3) Model transfer and capacity prediction: Load four pre-trained base BiLSTM sub-models and lock the parameters of the first two BiLSTM layers (to avoid forgetting source domain knowledge). Then, fine-tune the output layer of the corresponding base BiLSTM sub-model using partial data from the intrinsic mode function components in the target domain modal component set to adapt to the target domain data. For example, fine-tune the output layer of sub-model 1 using 20% ​​of the data from IMF_T1, fine-tune sub-model 2 using IMF_T2, and so on. Then, using the finely tuned four BiLSTM sub-models, the future capacity values ​​of IMF_T1-IMF_T4 are predicted respectively, resulting in four sets of model prediction sub-results P1, P2, P3 and P4.

[0075] Finally, the four sets of model prediction sub-results are summed element by element to obtain the target domain battery capacity prediction data P_final[i], that is, P_final[i]=P1[i]+P2[i]+P3[i]+P4[i] (i is the prediction step).

[0076] It is easy to understand that in subsequent transfer learning scenarios where the source domain contains capacity regeneration phenomena but the target domain does not, or in transfer learning scenarios where the source domain does not contain capacity regeneration phenomena but the target domain does, the model structure and training process of the basic bidirectional long short-term memory neural network model are similar to the above process. The only difference between the different scenarios is the number of bidirectional long short-term memory neural network models to be trained and the source of model samples. This will not be elaborated further.

[0077] Therefore, in this implementation, the BiLSTM model is first pre-trained using all source domain data to fully learn general temporal features. When transferring to the target domain task, the parameters of the first two layers of the pre-trained model are locked (not updated), and the fully connected layer (output layer) of the model is fine-tuned only using a small amount of target domain data to adapt to the target domain data distribution. Finally, the fine-tuned model is tested using the remaining target domain data to verify its prediction performance. This preserves the main trend features while effectively weakening the interference of regeneration fluctuations, thereby improving the prediction accuracy of the model in scenarios where both the source and target domains contain capacity regeneration.

[0078] In a second feasible implementation, the modal component set includes: a source domain modal component set. In this embodiment, step S31 may include step B3: Step B3: In the case where the current transfer learning scenario is that the source domain contains capacity regeneration and the target domain does not contain capacity regeneration, variational mode decomposition is performed on the first battery capacity sequence to obtain the source domain mode component set, which contains multiple intrinsic mode function components. Accordingly, in this embodiment, step S32 may include steps C3 to C6: Step C3: Calculate the correlation coefficient between each intrinsic mode function component in the source domain modal component set and the second battery capacity sequence; Step C4: Based on the correlation coefficient, select the intrinsic mode function component with the highest correlation from the source domain modal component set as the transfer feature component; Step C5: Train the bidirectional long short-term memory neural network model based on the transfer feature components to obtain the target capacity prediction model; Step C6: Perform capacity prediction on the second battery capacity sequence using the target capacity prediction model to obtain the final capacity prediction data corresponding to the target domain battery.

[0079] It should be noted that when the first battery capacity sequence in the source domain contains battery data with capacity regeneration, and the second battery capacity sequence in the target domain contains battery data without capacity regeneration, the components of the target domain after VMD decomposition do not contain capacity regeneration fluctuation characteristics, while the components obtained from the source domain decomposition do contain regeneration fluctuations. If all modal components of the source domain are directly transferred for capacity prediction, the modal components of the target domain, lacking fluctuation characteristics, will cause the model to incorrectly learn irrelevant fluctuations, misleading the transfer process and thus affecting the final prediction results.

[0080] Therefore, in the current transfer learning scenario where the source domain contains capacity regeneration phenomena but the target domain does not, if the above-mentioned "component-by-component transfer" method is directly used to transfer all components of the source domain during the transfer learning process, the components of the target domain may be affected by irrelevant fluctuations, thereby reducing the prediction accuracy.

[0081] Therefore, for transfer learning in this scenario, this implementation proposes a feature component selection method based on correlation, namely the "feature selection transfer" strategy. Here, the aforementioned correlation coefficient can be the Pearson correlation coefficient. The analysis device can determine this correlation coefficient by quantifying the linear correlation between the source domain IMF component and the target domain capacity sequence. Its value range can be [-1, 1], with the absolute value closer to 1 indicating a stronger correlation.

[0082] It is important to understand that the aforementioned migration feature components represent the IMF components in the source domain modal component set that have the highest correlation with the target domain capacity sequence. These components contain the main degradation trend shared by both the source and target domains and do not contain interfering regenerative fluctuations. Therefore, the analysis device can sort the correlation coefficients of all source domain IMF components and select the component with the largest absolute value as the migration feature component (usually r>0.9 indicates high correlation).

[0083] Understandably, referring to Figure 6 It can be seen that, Figure 6 This is a schematic diagram of the second transfer learning process in the second embodiment of the lithium battery remaining life analysis method of this application, as shown below. Figure 6 As shown, the target capacity prediction model described above can be based solely on the IMF component in the source domain modal component set that is most relevant to the target domain data (i.e., Figure 6 The IMF1 shown is used as the transfer feature component. The BiLSTM model is trained and fine-tuned with the partial data of the target domain without modal decomposition to ensure that the target capacity prediction model trained based on the source domain data is adapted to the smooth decay characteristics of the target domain.

[0084] For example, refer to Figure 6To illustrate, in a current transfer learning scenario where the source domain exhibits capacity regeneration but the target domain does not, the capacity prediction process based on transfer learning in this implementation may include the following steps: 1) Source Domain VMD Decomposition: The first battery capacity sequence is input and VMD decomposition is performed based on the optimal K to obtain the source domain mode component set {IMF_S1, IMF_S2, IMF_S3, IMF_S4} (which can correspond to respectively). Figure 6 IMF1~IMF4 are generated after variational mode decomposition of the source domain data.

[0085] Then, the Pearson correlation coefficient r between each IMF component and the target domain data, i.e., the second battery capacity sequence X_T, is calculated. The correlation coefficients are determined as follows: r1 = corr(IMF_S1, X_T) = 0.9946 (highly correlated); r2 = corr(IMF_S2, X_T) = 0.0547 (weakly correlated); r3 = corr(IMF_S3, X_T) = 0.0089 (no correlation); r4 = corr(IMF_S4, X_T) = 0.0045 (no correlation). At this point, IMF_S1, which has the largest r, can be selected as the migration feature component, and other weakly correlated / no correlated components can be discarded (to avoid interference from regeneration fluctuations).

[0086] 2) Training the target capacity prediction model: A three-layer BiLSTM model is constructed for the transfer feature component IMF_S1 to obtain a pre-trained model; then, based on partial data of the second battery capacity sequence X_T, the pre-trained model with the parameters of the first two layers is fine-tuned to generate the target capacity prediction model.

[0087] 3) Capacity forecasting: The remaining data of the second battery capacity sequence X_T are input into the target capacity prediction model, and the capacity prediction data is output.

[0088] In this embodiment, the negative impact of the source domain capacity regeneration component on the target domain can be avoided by using the "feature selection transfer" strategy, thereby improving the prediction accuracy of transfer learning.

[0089] In a third feasible implementation, the modal component set includes: a target domain modal component set. In this embodiment, step S31 may include step B4: Step B4: In the case where the current transfer learning scenario is that the source domain does not contain capacity regeneration and the target domain does contain capacity regeneration, variational mode decomposition is performed on the second battery capacity sequence to obtain the target domain mode component set, which contains multiple intrinsic mode function components. Accordingly, in this embodiment, step S32 may include steps C7~C8: Step C7: Train the bidirectional long short-term memory neural network model based on the first battery capacity sequence to obtain the target capacity prediction model, and use the target capacity prediction model to predict the capacity of each intrinsic mode function component in the target domain modal component set to obtain the model prediction sub-result corresponding to the target domain battery. Step C8: The model prediction sub-results are fused to generate the final capacity prediction data corresponding to the target domain battery.

[0090] It should be noted that when the source domain does not exhibit capacity regeneration but the target domain does, both the aforementioned "component-specific migration" and "feature selection migration" methods may not be effective in addressing the capacity regeneration fluctuation problem. In this scenario (i.e., the source domain does not exhibit capacity regeneration, but the target domain does), mode decomposition of the source domain data is unnecessary—because the source domain does not exhibit capacity regeneration, its capacity changes primarily exhibit a stable degradation trend. Decomposition would instead introduce redundant components, leading to increased fluctuation interference during the migration process and ultimately increasing the prediction error.

[0091] Specifically, the component-based transfer method requires VMD decomposition of both the source and target domains. However, since the source domain data does not contain capacity regeneration fluctuations, directly transferring the IMF components from the source domain decomposition will interfere with the regeneration fluctuation characteristics of the target domain, resulting in an inability to accurately model the volatility of the target domain. The feature selection transfer method, on the other hand, filters out the components most relevant to the target domain trend through correlation, but since the source domain data does not contain capacity regeneration fluctuations, this filtering is not possible.

[0092] Therefore, to address this issue, this implementation proposes a "target domain separate migration" strategy. First, VMD decomposition is performed only on the target domain data, decomposing it into a primary degradation trend and a fluctuation component (based on Pearson correlation analysis, the component with the highest correlation is taken as the primary degradation trend component, and the other components with low correlation are taken as fluctuation components), while the source domain data is not subjected to VMD decomposition. Then, the stationary features of the source domain are migrated to the components decomposed in the target domain.

[0093] At this time, refer to Figure 7 It can be seen that, Figure 7 This is a schematic diagram of the third transfer learning process in the second embodiment of the lithium battery remaining life analysis method of this application, as shown below. Figure 7 As shown, the analysis device can pre-train a target capacity prediction model based on the source domain data without mode decomposition, i.e., the stationary capacity sequence corresponding to the first battery capacity sequence, to learn the general battery degradation law. Then, the stationary features of the pre-trained and transfer-fine-tuned target capacity prediction model are transferred to each IMF component in the target domain, i.e., the multiple intrinsic mode function components (which can respectively correspond to...) contained in the target domain modal component set. Figure 7The IMF1~IMF4 generated from the variational mode decomposition of the target domain data are input into the target capacity prediction model to perform separate capacity prediction for each component, ensuring that regeneration fluctuations and the main degradation trend are accurately modeled. Finally, the capacity prediction sub-results corresponding to each component are superimposed element by element to reconstruct the complete capacity decay trend (including regeneration phenomena) of the target domain battery, thus obtaining the capacity prediction data.

[0094] In this embodiment, the stable knowledge in the source domain can be fully utilized in the target domain through the "target domain separate migration" strategy, and the capacity regeneration fluctuation can be explicitly modeled through the "component superposition" method, thereby ensuring the effectiveness and robustness of the model in the complex environment of the target domain.

[0095] The above are only three feasible implementation methods of steps S31 and S32 provided in this embodiment. This embodiment does not specifically limit the specific implementation methods of steps S31 and S32.

[0096] In summary, this embodiment addresses the impact of capacity regeneration on remaining lifetime prediction accuracy in transfer learning scenarios by proposing a multi-step remaining lifetime prediction method based on a combination of variational mode decomposition and data augmentation. This method effectively suppresses the interference of capacity regeneration during transfer learning on the prediction model and improves the accuracy of remaining lifetime prediction.

[0097] In the current transfer learning scenario where both the source and target domains contain capacity regeneration phenomena, this embodiment separates the main degradation trend (low-frequency IMF component) and capacity regeneration fluctuations (high-frequency IMF component) of the source and target domains through VMD decomposition, allowing each sub-model to focus on learning a single frequency feature and improving the model's fitting ability. Furthermore, by combining component-by-component transfer learning with parameter fine-tuning, the general variation patterns of each component in the source domain are preserved while adapting to the personalized characteristics of each component in the target domain. This ensures that the capacity prediction data after fusing the prediction results of each model accurately reflects the capacity decay trend (including regeneration phenomena) of the battery in the target domain. Experimental verification shows that the RMSE and MAE of this method are lower than those of the traditional MMD-BiLSTM method, which can significantly improve the prediction accuracy in scenarios where both the source and target domains contain capacity regeneration phenomena.

[0098] In the current transfer learning scenario where the source domain contains capacity regeneration but the target domain does not, this embodiment can accurately identify the common features (main degradation trend) of the source and target domains through correlation coefficient screening, eliminate irrelevant components containing regeneration fluctuations, and avoid interference from the source. At the same time, the target capacity prediction model after data augmentation is trained only on effective transfer features, adapts to the stable decay characteristics of the target domain, and significantly improves prediction accuracy (experimental verification shows that RMSE can be reduced to below 1.38%).

[0099] In the current transfer learning scenario where the source domain does not contain capacity regeneration phenomena but the target domain does, this embodiment only performs VMD decomposition on the target domain, explicitly separating the main degradation trend and regeneration fluctuations, facilitating targeted modeling. Furthermore, this embodiment transfers stationary features from the source domain to each component of the target domain, utilizing both the general degradation knowledge of the source domain and predicting the fluctuation characteristics of the target domain individually for each component. At this point, the fused capacity prediction data can accurately reconstruct the capacity changes (including regeneration phenomena) of the battery in the target domain. Experiments show that the AE error can be reduced to within 17 cycles, significantly outperforming traditional methods.

[0100] This embodiment discloses a method for handling a transfer learning scenario where both the source and target domains exhibit capacity regeneration. The method involves performing variational mode decomposition on the first battery capacity sequence to obtain a source domain modal component set, which contains multiple intrinsic mode function (EMF) components. Variational mode decomposition is then performed on the second battery capacity sequence to obtain a target domain modal component set, which contains the same number of EMF components as the source domain modal component set. For each EMF component in the source domain modal component set, a corresponding number of bidirectional long short-term memory (LSTM) neural network sub-models are trained. These sub-models then predict the capacity of each EMF component in the target domain modal component set, yielding model prediction sub-results for the target domain battery. Finally, the model prediction sub-results are fused to generate the final capacity prediction data for the target domain battery.

[0101] In the current transfer learning scenario where the source domain contains capacity regeneration phenomena and the target domain does not, variational mode decomposition is performed on the first battery capacity sequence to obtain a set of source domain mode components, which contains multiple intrinsic mode function components. The correlation coefficient between each intrinsic mode function component in the source domain mode component set and the second battery capacity sequence is calculated. Based on the correlation coefficient, the intrinsic mode function component with the highest correlation is selected from the source domain mode component set as the transfer feature component. The bidirectional long short-term memory neural network model is trained according to the transfer feature component to obtain the target capacity prediction model. The target capacity prediction model is used to predict the capacity of the second battery capacity sequence to obtain the final capacity prediction data corresponding to the target domain battery.

[0102] In the current transfer learning scenario where the source domain does not contain capacity regeneration and the target domain does, variational mode decomposition is performed on the second battery capacity sequence to obtain the target domain mode component set, which contains multiple intrinsic mode function components. A bidirectional long short-term memory neural network model is trained based on the first battery capacity sequence to obtain a target capacity prediction model. The target capacity prediction model is then used to predict the capacity of each intrinsic mode function component in the target domain mode component set, obtaining the model prediction sub-results corresponding to the target domain battery. The model prediction sub-results are then fused to generate the final capacity prediction data corresponding to the target domain battery.

[0103] Therefore, given the different capacity regeneration fluctuations in the source and target domains, directly using a single model for transfer learning can lead to a mixture of regeneration features and the main degradation trend. Furthermore, existing transfer learning methods cannot specifically learn the variation patterns of each component, resulting in low accuracy in remaining lifetime prediction. This embodiment addresses the difference in whether the source and target domains contain capacity regeneration data in transfer learning, dividing the remaining lifetime prediction problem based on transfer learning into three research scenarios. Differentiated variational mode decomposition and transfer strategies are adopted in different scenarios, achieving a layer-by-layer optimization from "overall component individual transfer" to "feature selection transfer" and then to "target domain individual transfer," systematically solving the problem of interference from capacity regeneration to transfer learning.

[0104] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the lithium battery remaining life analysis method of this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0105] This application also provides a lithium battery remaining life analysis device, please refer to... Figure 8 , Figure 8 This is a schematic diagram of the module structure of the lithium battery remaining life analysis device according to an embodiment of this application. In this embodiment, the lithium battery remaining life analysis device includes: Data acquisition module 801 is used to acquire the first battery capacity sequence corresponding to the source domain battery and the second battery capacity sequence corresponding to the target domain battery; The scene analysis module 802 is used to perform capacity regeneration detection on the first battery capacity sequence and the second battery capacity sequence, and determine the current transfer learning scene based on the detection results; The capacity prediction module 803 is used to perform adaptive transfer prediction on the first battery capacity sequence and the second battery capacity sequence based on the current transfer learning scenario, so as to obtain the final capacity prediction data corresponding to the target domain battery. The remaining life assessment module 804 is used to determine the remaining life prediction result of the target domain battery based on the capacity failure threshold corresponding to the target domain battery and the capacity prediction data.

[0106] Optionally, in this embodiment, the scenario analysis module 802 is further configured to determine a first capacity regeneration feature identification result based on the capacity decay characteristics of the first battery capacity sequence; determine a second capacity regeneration feature identification result based on the capacity decay characteristics of the second battery capacity sequence; and perform data matching on a preset transfer learning scenario type, the first capacity regeneration feature identification result, and the second capacity regeneration feature identification result to obtain the current transfer learning scenario; the preset transfer learning scenario type includes: both the source domain and the target domain contain capacity regeneration phenomena, the source domain contains capacity regeneration phenomena and the target domain does not contain capacity regeneration phenomena, and the source domain does not contain capacity regeneration phenomena and the target domain contains capacity regeneration phenomena.

[0107] Optionally, in this embodiment, the capacity prediction module 803 is further configured to perform variational mode decomposition on the first battery capacity sequence and / or the second battery capacity sequence based on the current transfer learning scenario to obtain a set of mode components; and perform adaptive transfer learning training based on the set of mode components to obtain the final capacity prediction data corresponding to the target domain battery.

[0108] Optionally, in this embodiment, the modal component set includes: a source domain modal component set and a target domain modal component set; The capacity prediction module 803 is further configured to, when the current transfer learning scenario involves capacity regeneration in both the source domain and the target domain, perform variational mode decomposition on the first battery capacity sequence to obtain the source domain mode component set, wherein the source domain mode component set contains multiple intrinsic mode function components; and perform variational mode decomposition on the second battery capacity sequence to obtain the target domain mode component set, wherein the target domain mode component set contains the same number of intrinsic mode function components as the source domain mode component set. Accordingly, the capacity prediction module 803 is further configured to train a corresponding number of bidirectional long short-term memory neural network sub-models for each intrinsic mode function component in the source domain modal component set, and to perform capacity prediction on each intrinsic mode function component in the target domain modal component set through the bidirectional long short-term memory neural network sub-models to obtain the model prediction sub-results corresponding to the target domain battery; and to perform fusion processing on the model prediction sub-results to generate the final capacity prediction data corresponding to the target domain battery.

[0109] Optionally, in this embodiment, the modal component set includes: a source domain modal component set; The capacity prediction module 803 is further configured to perform variational mode decomposition on the first battery capacity sequence to obtain the source domain mode component set when the current transfer learning scenario is that the source domain contains capacity regeneration phenomenon and the target domain does not contain capacity regeneration phenomenon, and the source domain mode component set contains multiple intrinsic mode function components. Accordingly, the capacity prediction module 803 is further configured to calculate the correlation coefficient between each intrinsic mode function component in the source domain modal component set and the second battery capacity sequence; based on the correlation coefficient, select the intrinsic mode function component with the highest correlation from the source domain modal component set as the transfer feature component; train the bidirectional long short-term memory neural network model according to the transfer feature component to obtain the target capacity prediction model; and perform capacity prediction on the second battery capacity sequence through the target capacity prediction model to obtain the final capacity prediction data corresponding to the target domain battery.

[0110] Optionally, in this embodiment, the modal component set includes: a target domain modal component set; The capacity prediction module 803 is further configured to perform variational mode decomposition on the second battery capacity sequence when the current transfer learning scenario is that the source domain does not contain capacity regeneration phenomena and the target domain contains capacity regeneration phenomena, to obtain the target domain mode component set, wherein the target domain mode component set contains multiple intrinsic mode function components. Accordingly, the capacity prediction module 803 is further configured to train a bidirectional long short-term memory neural network model based on the first battery capacity sequence to obtain a target capacity prediction model, and to perform capacity prediction on each intrinsic mode function component in the target domain modal component set using the target capacity prediction model to obtain the model prediction sub-result corresponding to the target domain battery; and to perform fusion processing on the model prediction sub-result to generate the final capacity prediction data corresponding to the target domain battery.

[0111] Optionally, in this embodiment, the data acquisition module 801 is further configured to acquire the initial battery capacity sequence corresponding to the source domain battery; perform a translation transformation on the initial battery capacity sequence to obtain an enhanced sample; and summarize the initial battery capacity sequence and the enhanced sample into a first battery capacity sequence.

[0112] The lithium battery remaining life analysis device provided in this application, employing the lithium battery remaining life analysis method in the above embodiments, can solve the technical problem of how to suppress interference from capacity regeneration and improve the accuracy of lithium battery transfer learning remaining life prediction. Compared with the prior art, the beneficial effects of the lithium battery remaining life analysis device provided in this application are the same as those of the lithium battery remaining life analysis method provided in the above embodiments, and other technical features in the lithium battery remaining life analysis device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0113] This application provides a lithium battery remaining life analysis device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the lithium battery remaining life analysis method in the first embodiment described above.

[0114] The following is for reference. Figure 9 , Figure 9 This is a schematic diagram of the hardware operating environment involved in the lithium battery remaining life analysis method in this application embodiment. The lithium battery remaining life analysis device in this application embodiment may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), vehicle terminals (such as vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 9 The lithium battery remaining life analysis device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0115] like Figure 9As shown, the lithium battery remaining life analysis device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the lithium battery remaining life analysis device. The processing unit 1001, the ROM 1002, and the RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the lithium battery remaining life analysis device to communicate wirelessly or wiredly with other devices to exchange data. Although a lithium battery remaining life analysis device with various systems is shown in the figure, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems can be implemented alternatively.

[0116] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, an embodiment disclosed in this application includes a lithium battery remaining life analysis program product, which includes a lithium battery remaining life analysis program carried on a computer-readable medium, the lithium battery remaining life analysis program containing program code for performing the methods shown in the flowcharts. In such an embodiment, the lithium battery remaining life analysis program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the lithium battery remaining life analysis program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0117] The lithium battery remaining life analysis device provided in this application, employing the lithium battery remaining life analysis method described in the above embodiments, can solve the technical problem of how to suppress interference from capacity regeneration and improve the accuracy of lithium battery transfer learning remaining life prediction. Compared with the prior art, the beneficial effects of the lithium battery remaining life analysis device provided in this application are the same as those of the lithium battery remaining life analysis method provided in the above embodiments, and other technical features in this lithium battery remaining life analysis device are the same as those disclosed in the method of the previous embodiment, and will not be repeated here.

[0118] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0119] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0120] This application provides a storage medium having computer-readable program instructions (i.e., a lithium battery remaining life analysis program) stored thereon, which are used to execute the lithium battery remaining life analysis method in the above embodiments.

[0121] The storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of the storage medium may include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0122] The aforementioned storage medium may be included in the lithium battery remaining life analysis device; or it may exist independently and not be assembled into the lithium battery remaining life analysis device.

[0123] The aforementioned storage medium carries one or more programs. When the aforementioned one or more programs are executed by the lithium battery remaining life analysis device, the lithium battery remaining life analysis device becomes capable of performing lithium battery remaining life analysis.

[0124] The lithium battery remaining life analysis program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—as well as conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0125] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and lithium battery remaining life analysis program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0126] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0127] The readable storage medium provided in this application is a storage medium storing computer-readable program instructions (i.e., a lithium battery remaining life analysis program) for executing the above-described lithium battery remaining life analysis method. This solves the technical problem of how to suppress interference from capacity regeneration and improve the accuracy of lithium battery transfer learning remaining life prediction. Compared with the prior art, the beneficial effects of the storage medium provided in this application are the same as those of the lithium battery remaining life analysis method provided in the above embodiments, and will not be repeated here.

[0128] The above are only some embodiments of this application and do not limit the scope of the solution of this application. All equivalent structural transformations made under the technical concept of this application and using the content of this application specification and drawings, or direct / indirect applications in other related technical fields, are included within the protection scope of this application.

Claims

1. A method for analyzing the remaining life of a lithium battery, characterized in that, The method includes: Obtain the first battery capacity sequence corresponding to the source domain battery and the second battery capacity sequence corresponding to the target domain battery; Capacity regeneration detection is performed on the first battery capacity sequence and the second battery capacity sequence, and the current transfer learning scenario is determined based on the detection results; Based on the current transfer learning scenario, adaptive transfer prediction is performed on the first battery capacity sequence and the second battery capacity sequence to obtain the final capacity prediction data corresponding to the target domain battery. The remaining service life prediction result of the target domain battery is determined based on the capacity failure threshold corresponding to the target domain battery and the capacity prediction data.

2. The lithium battery remaining life analysis method as described in claim 1, characterized in that, The step of performing capacity regeneration detection on the first battery capacity sequence and the second battery capacity sequence, and determining the current transfer learning scenario based on the detection results, includes: The first capacity regeneration feature identification result is determined based on the capacity decay characteristics of the first battery capacity sequence. Based on the capacity decay characteristics of the second battery capacity sequence, the identification result of the second capacity regeneration feature is determined; Data matching is performed on the preset transfer learning scenario type, the first capacity regeneration feature recognition result, and the second capacity regeneration feature recognition result to obtain the current transfer learning scenario; the preset transfer learning scenario type includes: both the source domain and the target domain contain capacity regeneration phenomena, the source domain contains capacity regeneration phenomena and the target domain does not contain capacity regeneration phenomena, and the source domain does not contain capacity regeneration phenomena and the target domain contains capacity regeneration phenomena.

3. The lithium battery remaining life analysis method as described in claim 2, characterized in that, The step of performing adaptive transfer prediction on the first battery capacity sequence and the second battery capacity sequence based on the current transfer learning scenario to obtain the final capacity prediction data corresponding to the target domain battery includes: Based on the current transfer learning scenario, variational mode decomposition is performed on the first battery capacity sequence and / or the second battery capacity sequence to obtain a set of mode components; Adaptive transfer learning training is performed based on the modal component set to obtain the final capacity prediction data corresponding to the target domain battery.

4. The lithium battery remaining life analysis method as described in claim 3, characterized in that, The modal component set includes: a source domain modal component set and a target domain modal component set; The step of performing variational mode decomposition on the first battery capacity sequence and / or the second battery capacity sequence based on the current transfer learning scenario to obtain a set of mode components includes: In the current transfer learning scenario where both the source domain and the target domain contain capacity regeneration phenomena, variational mode decomposition is performed on the first battery capacity sequence to obtain the source domain mode component set, which contains multiple intrinsic mode function components. Variational mode decomposition is performed on the second battery capacity sequence to obtain the target domain mode component set, which contains the same number of intrinsic mode function components as the source domain mode component set; Accordingly, the step of performing adaptive transfer learning training based on the modal component set to obtain the final capacity prediction data corresponding to the target domain battery includes: Based on each intrinsic mode function component in the source domain modal component set, a corresponding number of bidirectional long short-term memory neural network sub-models are trained, and the capacity prediction of each intrinsic mode function component in the target domain modal component set is performed by the bidirectional long short-term memory neural network sub-models to obtain the model prediction sub-results corresponding to the target domain battery. The model prediction results are fused to generate the final capacity prediction data corresponding to the target domain battery.

5. The lithium battery remaining life analysis method as described in claim 3, characterized in that, The modal component set includes: the source domain modal component set; The step of performing variational mode decomposition on the first battery capacity sequence and / or the second battery capacity sequence based on the current transfer learning scenario to obtain a set of mode components includes: In the case where the current transfer learning scenario is that the source domain contains capacity regeneration and the target domain does not contain capacity regeneration, variational mode decomposition is performed on the first battery capacity sequence to obtain the source domain mode component set, which contains multiple intrinsic mode function components. Accordingly, the step of performing adaptive transfer learning training based on the modal component set to obtain the final capacity prediction data corresponding to the target domain battery includes: Calculate the correlation coefficient between each intrinsic mode function component in the source domain modal component set and the second battery capacity sequence; Based on the correlation coefficient, the intrinsic mode function component with the highest correlation is selected from the source domain modal component set as the transfer feature component; The bidirectional long short-term memory neural network model is trained based on the aforementioned transfer feature components to obtain a target capacity prediction model; The target capacity prediction model is used to predict the capacity of the second battery capacity sequence to obtain the final capacity prediction data corresponding to the target domain battery.

6. The method for analyzing the remaining life of a lithium battery as described in claim 3, characterized in that, The modal component set includes: the target domain modal component set; The step of performing variational mode decomposition on the first battery capacity sequence and / or the second battery capacity sequence based on the current transfer learning scenario to obtain a set of mode components includes: In the current transfer learning scenario where the source domain does not contain capacity regeneration phenomena and the target domain does contain capacity regeneration phenomena, variational mode decomposition is performed on the second battery capacity sequence to obtain the target domain mode component set, which contains multiple intrinsic mode function components. Accordingly, the step of performing adaptive transfer learning training based on the modal component set to obtain the final capacity prediction data corresponding to the target domain battery includes: The bidirectional long short-term memory neural network model is trained based on the first battery capacity sequence to obtain the target capacity prediction model. The target capacity prediction model is then used to predict the capacity of each intrinsic mode function component in the target domain modal component set to obtain the model prediction sub-result corresponding to the target domain battery. The model prediction results are fused to generate the final capacity prediction data corresponding to the target domain battery.

7. The method for analyzing the remaining life of a lithium battery as described in claim 1, characterized in that, Before obtaining the first battery capacity sequence corresponding to the source domain battery and the second battery capacity sequence corresponding to the target domain battery, the method further includes: Obtain the initial battery capacity sequence corresponding to the source domain battery; Perform a translation transformation on the initial battery capacity sequence to obtain enhanced samples; The initial battery capacity sequence and the enhanced sample are combined into a first battery capacity sequence.

8. A lithium battery remaining life analysis device, characterized in that, The lithium battery remaining life analysis device includes: The data acquisition module is used to acquire the first battery capacity sequence corresponding to the source domain battery and the second battery capacity sequence corresponding to the target domain battery. The scenario analysis module is used to perform capacity regeneration detection on the first battery capacity sequence and the second battery capacity sequence, and determine the current transfer learning scenario based on the detection results; The capacity prediction module is used to perform adaptive transfer prediction on the first battery capacity sequence and the second battery capacity sequence based on the current transfer learning scenario, so as to obtain the final capacity prediction data corresponding to the target domain battery. The remaining life assessment module is used to determine the remaining life prediction result of the target domain battery based on the capacity failure threshold corresponding to the target domain battery and the capacity prediction data.

9. A lithium battery remaining life analysis device, characterized in that, The device includes: a memory, a processor, and a lithium battery remaining life analysis program stored in the memory and executable on the processor, the lithium battery remaining life analysis program being configured to implement the steps of the lithium battery remaining life analysis method as claimed in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium stores a lithium battery remaining life analysis program, which, when executed by a processor, implements the steps of the lithium battery remaining life analysis method as described in any one of claims 1 to 7.

Images