Lithium battery health state estimation method and system based on disturbance recovery enhancement and cross-domain adaptation mechanism

Abstract

The invention discloses a lithium battery health state estimation method and system based on disturbance recovery enhancement and a cross-domain adaptation mechanism, relates to the technical field of lithium batteries, and aims to simulate degradation regeneration of original data with SOH labels, generate high-fidelity extended samples and construct an enhanced data set to solve the problem of scarcity of original labeled data; performing hierarchical analysis on the enhanced data, mining long and short term dynamic laws of the lithium battery operation data, and outputting an initial SOH prediction result; by means of a cross-domain knowledge adaptation optimization strategy, the feature distribution difference between a source domain and a target domain is quantified through a distribution similarity measurement mechanism, new environment data distribution self-adaptation alignment is achieved through dynamic parameter adjustment, a prediction result is corrected, and a high-precision SOH value is output. According to the method, high-precision estimation of the SOH of the distributed energy storage lithium battery is realized, the method is suitable for an original scene with SOH label data volume scarcity, and the applicability and the accuracy of the prediction model are improved.

Description

Technical Field

[0001] The present invention relates to the field of lithium battery technology, and more particularly to a method and system for estimating the health status of a lithium battery based on disturbance recovery enhancement and cross-domain adaptation mechanisms. Background Art

[0002] With the rapid development of new energy technologies, lithium batteries are increasingly being used in distributed energy storage systems. However, due to complex operating environments and incomplete data collection, existing SOH estimation models often face challenges such as scarce raw data, incomplete labels, and distribution shifts. This results in poor model generalization and large prediction errors in real-world scenarios.

[0003] Traditional lithium-ion battery state-of-health (SOH) estimation methods often rely on large amounts of labeled historical operating data for training. In practical applications, particularly at new deployment sites or in scenarios with changing operating conditions, sufficient labeled data is often unavailable. Existing data augmentation techniques fail to fully account for degradation mechanisms during data evolution, and traditional transfer learning methods struggle to effectively align significant distribution differences between the source and target domains.

[0004] Therefore, how to provide a lithium battery health status estimation method and system based on disturbance recovery enhancement and cross-domain adaptation mechanism, combined with generative enhancement and efficient migration technology, to perform high-precision estimation of the SOH of distributed energy storage lithium batteries is a problem that technical personnel in this field urgently need to solve. Summary of the Invention

[0005] In view of this, the present invention provides a lithium battery state of health estimation method and system based on disturbance recovery enhancement and cross-domain adaptation mechanism, which combines data disturbance enhancement strategy, deep sequence feature extraction technology and feature distribution adaptive alignment method to overcome the problems of existing SOH estimation methods in data scarcity, insufficient model generalization ability and poor cross-scenario adaptability. Through the data disturbance enhancement model, the present invention can generate diversified training samples on a small amount of lithium battery raw data with SOH labels; based on the sequence feature extraction module, the system can effectively capture the dynamic features in the lithium battery operation data; combined with the cross-domain knowledge adaptation optimization strategy, the present invention achieves adaptive processing of cross-domain data by measuring and minimizing the feature differences between different data distributions. Overall, the present invention provides an accurate and highly generalized SOH estimation method and system, which can obtain stable health status prediction results when the amount of raw data is scarce.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for estimating the health status of a lithium battery based on disturbance recovery enhancement and cross-domain adaptation mechanism, comprising: Collect raw data of lithium batteries with health status labels and generate redistributed data through data perturbation enhancement model; Perform sequence feature extraction and health status estimation on the redistributed data, output the preliminary prediction results of the lithium battery health status, and calculate the health status estimation deviation; Feature distribution alignment is performed based on the distribution mean difference measurement method to obtain feature distribution differences; A joint loss function is constructed based on the health state estimation deviation and the feature distribution difference. By iteratively updating parameters, the cross-domain generalization ability of the model is improved to obtain the final lithium battery health state prediction result.

[0007] Preferably, the data disturbance enhancement model includes a disturbance generation unit and a dynamic feature identification unit, which are used to introduce disturbance into the original data of the lithium battery to generate disturbance data samples, and then restore the disturbance data samples to obtain restored redistributed data.

[0008] Preferably, the disturbance generation unit is used to gradually add Gaussian noise to the lithium battery raw data based on a preset disturbance scheduling strategy until a preset disturbance level is reached, thereby generating multi-stage degradation samples; The dynamic feature identification unit is used to fuse time step information, distinguish the differences between samples generated at different stages and real data during the training process, and improve the reverse recovery capability of the disturbance generation unit through an adversarial optimization mechanism.

[0009] The disturbance generation unit simulates data degradation by gradually introducing Gaussian noise, and then the inverse recovery module iteratively restores the disturbed samples to restore the data representation close to the original distribution; the dynamic feature identification unit is used to identify the difference features between the restored samples and the real data, and optimize the accuracy and stability of the restoration process through an adversarial training mechanism.

[0010] Preferably, after the perturbation generation unit is trained, it can generate initial pseudo samples from random Gaussian noise and reconstruct its characteristic structure through an inverse recovery mechanism.

[0011] Preferably, the reverse recovery mechanism is used to gradually remove noise and reconstruct the original data features until the noise-free state is restored to obtain a pseudo sample; A neighborhood feature matching strategy is adopted to calculate the feature similarity between the pseudo samples and the original samples, determine the corresponding health status labels, and merge the pseudo samples with the original dataset to construct an extended dataset for subsequent training.

[0012] The initial pseudo sample is restored to the characteristic scale of the original data after denormalization, and the health status label is back-mapped using the nearest neighbor distribution strategy based on the neighborhood similarity with the original sample in the feature space; The pseudo samples with health status labels are merged with the original samples to construct a sample dataset for subsequent modeling training.

[0013] Based on the time step encoding strategy, temporal information is embedded into data features and a probabilistic evaluation of the data authenticity is output.

[0014] Preferably, performing sequence feature extraction and health status estimation on the redistributed data includes: A selective state space sequence model including multi-layer nonlinear transformation and state control mechanism is used to mine the time series features of redistributed data.

[0015] Preferably, the selective state space sequence model includes a deep dynamic filtering unit, an information screening unit, a multi-scale dependency capturing unit and a health state estimation unit which are connected in sequence.

[0016] Preferably, the deep dynamic filtering unit processes the input sequence data based on an adaptive filtering strategy; The information screening unit screens the processed data through a gated linear structure; The multi-scale dependency capture unit extracts local and global temporal dependency features of the time series in parallel in multiple feature spaces; The health state estimation unit performs aggregate analysis on the extracted time series dependency features and outputs a preliminary prediction result of the health state of the lithium battery.

[0017] Preferably, feature distribution alignment is performed based on a distribution mean difference measurement method to reduce the impact of different data distribution differences on model performance; obtaining feature distribution differences includes: calculating the feature distribution differences between source domain and target domain samples in a high-dimensional feature space, and dynamically adjusting parameters in the feature extraction process using an adaptive matching strategy based on the measurement results to minimize the feature distribution differences; Among them, during the feature distribution alignment process, the health state estimation deviation is dynamically monitored; When the estimation error exceeds the preset threshold, the weight of the distribution mean difference metric is automatically adjusted, the feature distribution difference is recalculated and the parameters are updated until the prediction error converges to the preset range.

[0018] Preferably, a lithium battery health status estimation system based on disturbance recovery enhancement and cross-domain adaptation mechanism includes: The data redistribution module is used to collect raw data of lithium batteries with health status labels and generate redistributed data through the data perturbation enhancement model; The feature extraction and estimation module is used to extract sequence features and estimate the health status of the redistributed data, output the preliminary prediction results of the health status of the lithium battery, and calculate the health status estimation deviation; The cross-domain knowledge adaptation optimization module is used to align feature distributions based on the distribution mean difference measurement method, obtain feature distribution differences, and reduce the impact of cross-domain deviations on prediction accuracy; The result output module is used to construct a joint loss function based on the health state estimation deviation and the feature distribution difference, and to improve the cross-domain generalization ability of the model by iteratively updating parameters to obtain the final lithium battery health state prediction result.

[0019] Through the above technical solutions, it can be seen that compared with the existing technology, the present invention discloses a method and system for estimating the health status of lithium batteries based on disturbance recovery enhancement and cross-domain adaptation mechanism, including: collecting original lithium battery operating data with SOH labels, using a data disturbance enhancement model to perform expansion processing on the original lithium battery operating data to obtain an expanded data set; using a selective state space sequence model to extract features from the expanded data set, learning the potential dynamic laws in the lithium battery time series, and obtaining a preliminary prediction result of the lithium battery SOH; using a strategy of minimizing feature distribution differences to perform cross-domain knowledge adaptation optimization, correcting the preliminary prediction result, and outputting the final SOH prediction value. The present invention achieves high-precision estimation of the SOH of distributed energy storage lithium batteries, is suitable for scenarios where the amount of original SOH-labeled data is scarce, and can improve the applicability and accuracy of the prediction model by generating effective expanded training samples and optimizing the data distribution of the new environment through cross-domain adaptation. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0021] Figure 1 A schematic flow chart of a lithium battery health status estimation method based on disturbance recovery enhancement and cross-domain adaptation mechanism provided in an embodiment of the present invention.

[0022] Figure 2 This is a diagram of the SOH prediction results provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0023] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0024] The embodiment of the present invention discloses a method for estimating the health status of a lithium battery based on disturbance recovery enhancement and cross-domain adaptation mechanism, such as Figure 1 As shown in the figure, the entire process from data collection, data generation, feature learning and prediction to cross-domain knowledge adaptation and optimization and result output is clearly presented. The specific steps are as follows: S100, collecting raw data of lithium batteries with health status labels, and generating redistributed data through a data perturbation enhancement model to obtain an expanded data set; S200, performing sequence feature extraction and health state estimation on the redistributed data, outputting a preliminary prediction result of the lithium battery health state, and calculating the health state estimation deviation; wherein, a selective state space sequence model is used to model and extract features of the expanded data set, learn the potential dynamic laws in the lithium battery time series, and obtain a preliminary prediction result of the lithium battery health state; S300: Align feature distributions based on a distribution mean difference metric to obtain feature distribution differences; construct a joint loss function based on the health state estimation deviation and the feature distribution differences, iteratively update parameters to improve the model's cross-domain generalization capability, and obtain a final lithium battery health state prediction result. This embodiment of the present invention performs cross-domain feature adaptation optimization, modifies the preliminary lithium battery health state prediction result, and outputs a final SOH prediction value.

[0025] This embodiment of the present invention generates diverse data samples by gradually introducing perturbations to a small amount of raw data from lithium batteries with health status labels. This process, based on a pre-set perturbation scheduling strategy, gradually applies Gaussian noise to the raw data, generating multiple noisy data versions until a preset perturbation level is reached. Based on the disturbed data samples, the noise interference is gradually removed through the reverse recovery process to obtain the restored redistributed data. This effectively restores the original data structure and maintains the correspondence between the health status label and the restored sample. Based on the restored data, a structure including multi-layer nonlinear transformation and state control mechanism is used to explore the potential temporal evolution laws in the lithium battery operation data and output preliminary health status prediction results; For lithium battery samples from different data distributions, the distribution difference between their features is calculated and minimized through a cross-domain knowledge adaptation optimization strategy, so that the distribution of source and target domain samples in the high-dimensional feature space tends to be consistent; A joint loss function is constructed to incorporate the deviation of health state estimation and the difference in feature distribution into the optimization objective. The parameters are iteratively updated to minimize the joint loss, thus achieving stable and accurate health state estimation.

[0026] Specifically, the data disturbance enhancement model includes a disturbance generation unit and a dynamic feature identification unit, which are used to introduce disturbance into the original data of the lithium battery to generate disturbance data samples, and then restore the disturbance data samples to obtain restored redistributed data.

[0027] Specifically, the disturbance generation unit is used to gradually add Gaussian noise to the lithium battery raw data based on a preset disturbance scheduling strategy until a preset disturbance level is reached, thereby generating multi-stage degradation samples; The dynamic feature identification unit is used to fuse time step information, distinguish the differences between samples generated at different stages and real data during the training process, and improve the reverse recovery capability of the disturbance generation unit through an adversarial optimization mechanism.

[0028] The data perturbation enhancement model combines a noise perturbation mechanism with dynamic feature identification. The noise perturbation mechanism gradually introduces noise into the raw data of lithium batteries through a forward process to simulate the degradation process of operating data. The perturbation strategy is then combined with a reverse recovery strategy to iteratively restore the perturbed samples to an effective representation close to the original distribution. The dynamic feature identification is used to identify the distribution difference between the restored sample and the real data, and improve the discrimination accuracy by introducing time step embedding information.

[0029] Specifically, after the data perturbation enhancement model is trained, pseudo samples are generated from Gaussian noise, and the SOH labels are determined by denormalization and spatial proximity analysis methods, and then merged with the original samples to form an expanded dataset.

[0030] Specifically, the data perturbation enhancement model integrates the noise perturbation mechanism and the dynamic feature identification unit, including: In the forward perturbation, the pre-set variance is used for scheduling , gradually moving towards the original lithium battery running data Add Gaussian noise to get noise-contaminated data ; During the reverse recovery process, the Gradually restore the original data ; The input of the dynamic feature identification unit includes the data in the forward perturbation process and time step t, the time step t is encoded through a multi-layer perceptron and converted into a time feature vector; the output of the dynamic feature identification unit is the probability of data authenticity.

[0031] Specifically, pseudo samples are generated from Gaussian noise, and the SOH labels are determined by denormalization and spatial proximity analysis methods. These pseudo samples are then merged with the original samples to form an expanded dataset, including: Randomly sample the initial noise sample from the standard normal distribution; Then, using the trained diffusion reverse process model, reverse diffusion processing is iterated from t=T to t=1 to obtain pseudo samples; T is the total time step of the forward perturbation process; According to the position of the sample in the original feature space, the spatial proximity analysis method is used to find the K neighbors closest to the generated sample features in the original sample set, and the weighted average method is used to generate the SOH label to form a pseudo sample pair; The generated pseudo sample set is merged with the original real sample set to form the expanded dataset.

[0032] In one specific embodiment of the present invention, to overcome the limited amount of raw SOH data, the originally collected lithium battery operating data with SOH labels is first enhanced. This embodiment of the present invention utilizes a data perturbation enhancement model for data expansion. The raw data includes input feature data and corresponding SOH labels. Due to the limited number of samples, directly training the model is prone to overfitting and is difficult to generalize to new scenarios.

[0033] The data perturbation enhancement model innovatively integrates the noise perturbation mechanism and the dynamic feature identification unit. Its theoretical basis is based on a reversible mapping strategy that gradually perturbs and restores the distribution of time series data. Its core includes two stages: the forward perturbation process and the reverse recovery process.

[0034] The model's input consists of raw lithium battery operating data samples with a time series structure. Each sample consists of input features and a corresponding SOH label. All feature data is normalized and time-aligned before input to ensure the effectiveness and consistency of subsequent noise perturbation and sequence modeling.

[0035] S101, forward perturbation process In this process, the pre-set variance is used for scheduling. , gradually towards the initial data Adding Gaussian noise, in the embodiment of the present invention, Represents the original lithium battery characteristic data, that is, the vector composed of the voltage, current, temperature and other characteristic values ​​of the lithium battery collected. As the time step t advances from 1 to T, the data The noise gradually dominates the data, and the characteristics of the original data are gradually obscured, eventually approaching a pure noise distribution. T is the total number of time steps in the forward perturbation process, which determines the number of steps required to go from the original state to the fully noised state.

[0036] To ensure the stability and effectiveness of the noise addition process, variance scheduling Using fixed linear scheduling, the expression is as follows: (1) Where T represents the total number of diffusion steps, and Represent the minimum and maximum intensity of the noise respectively. , , ensuring that the noise increases gradually but not excessively.

[0037] The entire forward perturbation process can be accurately described by the following formula: (2) (3) in, The initial data is known In the case of all data samples from time step 1 to T The joint probability distribution of Indicates that the previous time step is known Data sample In the case of Data sample The conditional probability distribution of Represents Gaussian distribution, and I represents the identity matrix. Choosing the identity matrix I as the covariance matrix of Gaussian noise means that each feature is independent of each other when adding noise, and the noise of each feature has the same variance This method of gradually adding noise simulates the degradation process of data in a complex environment and provides a basis for the subsequent reverse recovery of the original data characteristics.

[0038] S102, reverse recovery process In contrast to the forward perturbation process, the reverse recovery process is dedicated to recovering the Gradually restore the original data x 0. The process can be expressed as: (4) (5) in, Indicates the model parameters Under the action of The joint probability distribution of . Represents the data sample at the known current time step t In the case of the data sample at the previous time step t-1 The conditional probability distribution of In this process, the model uses the learned parameters , continuously estimating the mean of each time step and covariance , gradually removing the noise and restoring the original characteristics of the data. The process of training perturbation recovery is essentially to optimize the variational lower bound so that the model can accurately learn the distribution characteristics of the data, thereby achieving effective recovery from noise to original data.

[0039] S103, introducing dynamic feature identification unit The data perturbation enhancement model utilizes step-by-step noise perturbation and a reverse recovery mechanism to synthesize diverse data samples to enhance the training dataset. However, traditional high-order step-by-step recovery processes often rely on long noise restoration paths, which not only increases computational overhead but can also affect the quality of the synthesized samples due to error accumulation. Therefore, the present invention introduces a dynamic feature identification unit that improves the efficiency of the noise recovery phase through an adversarial training mechanism and reduces the structural degradation of the generated samples, making the reconstructed lithium battery operating data more consistent with the actual distribution.

[0040] The main function of the dynamic feature identification unit is to distinguish the data generated at different time steps t from the real battery data, and to guide the optimization of the reverse recovery strategy through adversarial training so that the generated data is more consistent with the real SOH data distribution. Its application process mainly includes the following steps: Dynamic feature identification unit The input includes the data of the diffusion process (Contains real data sampled from the battery dataset and data generated by the data augmentation module) and time step t (representing the current stage of the diffusion process, taking discrete integer values ​​t = 1, 2, ..., T). In order to better utilize the time step information, the time step t is encoded through a multi-layer perceptron (MLP), that is, , first normalize t and then convert it into a time feature vector through MLP.

[0041] Discriminator Taking data and the encoding of time step t as input, output the probability that its source is real data or generated data: (6) in, is the weight matrix of data features, is the weight matrix of the time step feature, is the bias term, It is a Sigmoid function that is used to output the probability of data authenticity (ranging from 0 to 1).

[0042] The core goal of the enhancement model is to learn a reverse recovery process from random noise Restore real data The training process is mainly based on mean square error (MSE) loss, and the optimization goal is as follows: (7) in, is the data in the forward perturbation process; is Gaussian noise, is the noise predicted by the model, and the goal is to make the noise predicted by the model Close to real noise , in order to ensure the stability of the reverse recovery process, is a cumulative scaling factor used to gradually adjust the mixing ratio of data and noise during the forward perturbation process, defined as: (8) Where t is the time step, Variance scheduling parameter at the sth time step.

[0043] In the data perturbation enhancement model, a dynamic feature identification unit is added to further improve the authenticity and quality of the generated data. , therefore, adversarial loss It consists of two parts: (9) Among them, the first item indicates that the discriminator hopes to Give a higher probability (i.e., predict as true); the second term indicates that the discriminator hopes to Gives a lower probability (i.e. the prediction is false).

[0044] Finally, the data perturbation enhancement model is trained by combining the adversarial training loss of the dynamic feature identification unit with the inverse recovery loss, and the optimization goal is: (10) Through this adversarial training, the discriminator continuously optimizes its discrimination ability, and the data enhancement module generates data that is closer to the real SOH data.

[0045] S104. Enhanced sample generation and application After the data perturbation enhancement model is trained, it can be used to generate forged samples with realistic feature distribution from Gaussian noise to enhance the data scale and diversity of the SOH prediction task.

[0046] Specifically, Gaussian initialization is first performed, that is, the initial noise sample is randomly sampled from the standard normal distribution ( represents a completely degraded Gaussian noise sample with the same dimension as the original input feature); then, using the trained diffusion reverse process model, iteratively execute the formula from t=T to t=1: (11) in, , To enhance the neural network trained by the model, it is used to predict the current noise. 、 are the preset noise scheduler parameters used in the perturbation process, is the variance adjustment coefficient of the sampling process), after T steps of inverse diffusion processing, the pseudo sample is finally obtained: (12) Since the input features have been normalized to the interval [0, 1] or [-1, 1] before training, it is necessary to perform a denormalization operation on the generated results, that is, Then, according to the position of the sample in the original feature space, the spatial proximity analysis method is used to find the K neighbors closest to the generated sample features in the original sample set, and the SOH label is generated by weighted average. , and finally form a pseudo sample pair Finally, the pseudo sample set generated above is and the original real sample set Merge to form the enhanced complete dataset This enhanced dataset has richer sample distribution and coverage area, which helps to improve the training stability, robustness and generalization ability of subsequent SOH prediction models (such as selective state space sequence models), especially when the scale of original labeled data is limited.

[0047] Specifically, the sequence feature extraction and health status estimation of the redistributed data include: A selective state space sequence model including multi-layer nonlinear transformation and state control mechanism is used to mine the time series features of redistributed data.

[0048] Specifically, the selective state space sequence model includes a deep dynamic filtering unit, an information screening unit, a multi-scale dependency capturing unit and a health state estimation unit connected in sequence; Specifically, the deep dynamic filtering unit processes the input sequence data based on an adaptive filtering strategy; The information screening unit screens the processed data through a gated linear structure; The multi-scale dependency capture unit extracts local and global temporal dependency features of the time series in parallel in multiple feature spaces; The health state estimation unit performs aggregate analysis on the extracted time series dependency features and outputs a preliminary prediction result of the health state of the lithium battery.

[0049] The deep dynamic filtering unit performs a dynamic convolution operation on the input sequence through a learnable state-space filter to obtain filtered feature data; The information screening unit performs information screening on the filtered feature data; The multi-scale dependency capture unit captures the local and global dependencies of lithium batteries at different time scales in parallel in multiple subspaces; The health state estimation unit integrates the processed features and outputs the prediction result of the lithium battery SOH through the fully connected layer.

[0050] In a specific embodiment of the present invention, the data set enhanced by the data perturbation enhancement model is Efficient modeling and feature extraction are performed to learn the underlying dynamic patterns in time series and accurately regress and predict the SOH state corresponding to each input sequence. The enhanced data is processed using a selective state-space sequence model. The selective state-space sequence model is an efficient sequence model architecture that effectively captures the complex dependencies in lithium battery time series data. The entire processing process can be divided into the following stages: S201, Deep Dynamic Filtering Unit: The Selective State Space Sequence Model innovatively introduces a learnable state space filter specifically designed to capture the long-term dependency characteristics of lithium battery time series data. The Selective State Space Sequence Model performs a "dynamic convolution" operation on the input sequence, achieving cross-step information propagation through state transitions at each time step t. Each input sequence is represented by a parameterized state space model: (13) in, is the feature representation of the input sequence at time step t, is the updated hidden state, which integrates the hidden state of the previous moment With the current input This mechanism has lower complexity (approximately linear complexity) than standard convolution / self-attention, and has the ability to remember long-term dependencies, making it suitable for processing long-sequence problems such as battery life.

[0051] S202, information screening unit: After filtering, the features are subjected to information selection control by the information screening unit: (14) Among them, A and B are the linear transformation results extracted from the previous layer (i.e., the deep dynamic filtering unit); It is element-by-element multiplication, which is used to multiply A with the Sigmoid activation function. The processed B is multiplied element by element; Map the value range of B to the interval (0, 1) to achieve information filtering.

[0052] GLU selectively retains important information while shielding against noisy features, making it particularly suitable for processing potentially redundant data. In lithium battery data processing, it can identify key features that truly reflect the changing trends in battery health, eliminate interference from the data generation process, enhance model robustness, and provide higher-quality data features for subsequent processing.

[0053] S203, Multi-scale Dependency Capture Unit: Data enters the multi-scale attention mechanism. The multi-scale attention mechanism is used in the selective state space sequence model to capture the local and global dependencies of the input sequence at different time scales and construct a rich time series representation. First, the time series features generated by the data perturbation enhancement model are Mapped into query, key, and value vectors through linear transformation: (15) in , , is a learnable weight matrix, , is the number of attention sub-channels.

[0054] First, each attention sub-channel independently calculates a scaled dot product attention score: (16) Then multiple attention sub-channels execute the above calculation process in parallel to form multiple subspace attention outputs: (17) Finally, the outputs of all attention sub-channels are concatenated and passed through a linear transformation matrix Fusion is performed to obtain the final multi-scale attention output: (18) The multi-scale mechanism models temporal dependencies in parallel across multiple subspaces, accurately capturing the characteristic differences of lithium batteries at different aging stages. Compared to traditional sequence models, it offers advantages such as modeling long-term dependencies, maintaining temporal resolution, and adapting to multiple operating conditions, significantly improving the robustness of SOH prediction.

[0055] S204, health state estimation unit: After processing by the deep dynamic filtering unit, information screening unit, and multi-scale attention mechanism, the selective state space sequence model integrates the learned feature information and outputs the estimated results of the lithium battery SOH through a fully connected layer. During the training phase, the model minimizes the mean square error between the predicted value and the true SOH as the optimization goal. The loss function is expressed as: (19) in, is the number of training samples, For the The true SOH value of the samples, is the corresponding predicted value.

[0056] With the help of the back-propagation algorithm, the model propagates back along the network structure according to the error calculated by the loss function, and automatically adjusts various parameters in the model.

[0057] Specifically, feature distribution alignment is performed based on the distribution mean difference measurement method to reduce the impact of different data distribution differences on model performance. The feature distribution difference is obtained by calculating the feature distribution difference between source domain and target domain samples in the high-dimensional feature space, and using an adaptive matching strategy based on the measurement results to dynamically adjust the parameters in the feature extraction process to minimize the feature distribution difference. Among them, during the feature distribution alignment process, the health state estimation deviation is dynamically monitored; When the estimation error exceeds the preset threshold, the weight of the distribution mean difference metric is automatically adjusted, the feature distribution difference is recalculated and the parameters are updated until the prediction error converges to the preset range.

[0058] Specifically, a distributed mean difference algorithm is used to perform cross-domain knowledge adaptation optimization, and the lithium battery SOH prediction result output by the selective state space sequence model is corrected to output the final SOH prediction value, including: The distributed mean difference is used as the metric, and based on the Gaussian kernel function, the source domain feature vector set and the target domain feature vector set are mapped to the high-dimensional kernel Hilbert space respectively. Then, calculating the feature distribution alignment loss of the source domain feature vector set and the target domain feature vector set in the high-dimensional kernel Hilbert space; The selective state space sequence model is trained based on the feature distribution alignment loss to obtain a new comprehensive loss function; The gradient descent method is used to optimize the weights of the selective state space sequence model, and the weights are updated through continuous iteration to optimize the model's feature extraction capabilities.

[0059] Specifically, during the cross-domain knowledge adaptation optimization process, the prediction error of the selective state space sequence model is continuously monitored; If the prediction error exceeds the set threshold, the weight parameters of the feature distribution alignment loss are dynamically adjusted, the feature distribution alignment loss is recalculated, and the model weights are optimized and trained again. This process is repeated until the prediction error of the selective state space sequence model on the target domain data converges to a certain range, and the final SOH prediction value is output.

[0060] In a specific embodiment of the present invention, to effectively reduce prediction errors caused by different data distributions, enable the model to quickly and accurately adapt to new operating conditions, and enhance the model's generalization and prediction accuracy in different distributed energy storage scenarios, a distributed mean difference algorithm is used to measure and align the feature distributions of source and target domain data, achieving efficient domain adaptation and enabling the model to perform well on new datasets.

[0061] The source domain feature vector set ( )Depend on The intermediate sequence features extracted after the selective state space sequence model is extracted. This type of feature reflects the typical characteristics of battery aging behavior in the training dataset. The target domain feature vector set ( ) The battery operation data from a real deployment environment or target test scenario is processed by the same selective state space sequence model structure to generate a sequence feature representation.

[0062] In order to achieve consistent alignment of the feature distribution of the source domain and the target domain, the distribution mean difference is used as the measurement indicator, and based on the Gaussian kernel function, and They are mapped to a high-dimensional kernel Hilbert space (RKHS) and their mean distance is calculated in this space. The core idea is to minimize the mean difference between the two distributions in RKHS so that the distribution forms of the two feature sets in the latent space are as similar as possible. This mean difference is the feature distribution alignment loss, and its mathematical expression is: (20) in, Represents the result after mapping to high-dimensional feature space through kernel function; , Represent samples in the source domain and target domain respectively; and is the number of samples in the source domain and the target domain.

[0063] Using the Gaussian kernel function, the feature distribution alignment loss can be further expressed as: (twenty one) in, is a kernel function, preferably a Gaussian kernel function, which is defined as follows: (twenty two) Where σ is the kernel bandwidth parameter, which controls the sensitivity to local differences in the mapping space. Its specific value can be determined by cross-validation or automatically optimized as an adjustable hyperparameter.

[0064] The calculated feature distribution alignment loss is incorporated into the model’s training objective, and together with the original prediction loss, forms a new comprehensive loss function: (twenty three) Among them, the original prediction loss The health status estimation deviation between the SOH value predicted by the model and the true SOH value is measured by the mean square error (MSE), and the expression is shown in formula (19). λ is an adjustable hyperparameter used to balance the weights of prediction loss and feature distribution alignment loss. Subsequently, gradient descent is used to optimize the weights of the selective state-space sequence model network. By continuously iteratively updating the weights, the overall loss function gradually decreases, allowing the model to better adapt to the distribution characteristics of target domain data while learning the relationship between source domain data features and SOH.

[0065] During the cross-domain knowledge adaptation and optimization process, the model's prediction error is continuously monitored. If the prediction error exceeds a set threshold, indicating that the model is poorly adapted to the target domain, the weight λ is dynamically adjusted, the feature distribution alignment loss is recalculated, and the model weights are optimized and trained again. This process is repeated until it stabilizes and the model's prediction error on the target domain data converges to an acceptable range, ultimately outputting accurate and reliable SOH predictions.

[0066] In a specific embodiment of the present invention, a lithium battery health status estimation system based on disturbance recovery enhancement and cross-domain adaptation mechanism includes: The data redistribution module is used to collect raw data of lithium batteries with health status labels and generate redistributed data through the data perturbation enhancement model; The feature extraction and estimation module is used to extract sequence features and estimate the health status of the redistributed data, output the preliminary prediction results of the health status of the lithium battery, and calculate the health status estimation deviation; The cross-domain knowledge adaptation optimization module is used to align feature distributions based on the distribution mean difference measurement method to obtain feature distribution differences; The result output module is used to construct a joint loss function based on the health state estimation deviation and the feature distribution difference, and obtain the final lithium battery health state prediction result by iteratively updating parameters.

[0067] The embodiments of the present invention propose a lithium battery state of health estimation method and system based on disturbance recovery enhancement and cross-domain adaptation mechanism, which has made a breakthrough in solving the problem of traditional SOH estimation methods' strong dependence on a large amount of labeled data, as well as the predicament of data scarcity and weak generalization ability commonly found in data-driven methods.

[0068] The data perturbation augmentation model efficiently generates highly realistic pseudo samples through a unique forward diffusion and reverse recovery process, combined with the adversarial training mechanism of the dynamic feature identification unit. This module accurately learns the distribution of a small amount of real data and generates samples with realistic feature distributions, greatly expanding the capacity and diversity of the dataset. This provides a rich and high-quality data foundation for subsequent model training, effectively alleviating the problem of insufficient original annotated data.

[0069] The selective state space sequence model uses a deep dynamic filtering unit to capture the long-term dependency characteristics of data, uses an information screening unit to filter key information, and uses a multi-scale attention mechanism to mine dependencies at different time scales, forming a deep understanding and processing capability of the time series of lithium battery operation data, significantly improving the accuracy of SOH prediction.

[0070] A cross-domain knowledge adaptation optimization strategy based on feature distribution alignment accurately measures and aligns the feature distributions of source and target domain data, dynamically adjusting the optimization process to address data distribution shifts. By calculating the feature distribution alignment loss and incorporating it into the training objective, combined with gradient descent to optimize model weights, the model's prediction accuracy is ensured in new scenarios (target domains), effectively addressing the key issue of poor model adaptability.

[0071] Specifically, such as Figure 2 As shown in the figure, in terms of prediction accuracy, most of the data are concentrated in the range of difference close to about 0 to 0.0025, and the frequency in this range is as high as more than 800, indicating that in most cases the predicted value of the model is very close to the true value with very small deviation, which means that the model shows high accuracy in the SOH estimation task and can predict the true health status value more accurately.

[0072] From the perspective of the concentration of data distribution, the difference distribution shows an obvious concentration trend, and a large amount of data is concentrated in a few intervals. This reflects that the prediction results of the model are relatively stable. At different data points, the difference between the predicted value and the true value fluctuates little, and has good stability and consistency.

[0073] The embodiment of the present invention simulates degradation and regeneration of original data with SOH labels through a data perturbation enhancement model, integrates a time series discrimination mechanism to generate high-fidelity extended samples, and constructs an enhanced data set to solve the problem of scarcity of original labeled data; utilizes a selective state space sequence model to implement hierarchical analysis of the enhanced data, and through dynamic filtering, gated screening, and multi-scale dependency capture, mines the long-term and short-term dynamic laws of lithium battery operation data and outputs the initial SOH prediction result; with the help of a cross-domain knowledge adaptation optimization strategy, quantifies the difference in feature distribution between the source domain and the target domain through a distribution similarity measurement mechanism, dynamically adjusts parameters to achieve adaptive alignment of the new environment data distribution, corrects the prediction result, and outputs a high-precision SOH value. The embodiment of the present invention achieves high-precision estimation of the SOH of distributed energy storage lithium batteries, and is suitable for scenarios where the amount of original SOH-labeled data is scarce. It can improve the applicability and accuracy of the prediction model by generating effective extended training samples and adapting to the new environment data distribution across domains.

[0074] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0075] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A lithium battery health status estimation method based on disturbance recovery enhancement and cross-domain adaptation mechanism, characterized in that: include: Collect raw data of lithium batteries with health status labels and generate redistributed data through data perturbation enhancement model; Perform sequence feature extraction and health status estimation on the redistributed data, output the preliminary prediction results of the lithium battery health status, and calculate the health status estimation deviation; Feature distribution alignment is performed based on the distribution mean difference measurement method to obtain feature distribution differences; A joint loss function is constructed based on the health state estimation deviation and the feature distribution difference, and the final lithium battery health state prediction result is obtained by iteratively updating the parameters.

2. The method for estimating the health status of a lithium battery based on disturbance recovery enhancement and cross-domain adaptation mechanism according to claim 1, characterized in that: The data disturbance enhancement model includes a disturbance generation unit and a dynamic feature identification unit, which are used to introduce disturbance into the original data of the lithium battery to generate disturbance data samples, and then restore the disturbance data samples to obtain restored redistributed data.

3. The method for estimating the health status of a lithium battery based on disturbance recovery enhancement and cross-domain adaptation mechanism according to claim 2, characterized in that: The disturbance generation unit is used to add Gaussian noise to the lithium battery raw data based on a preset disturbance scheduling strategy until a preset disturbance level is reached, thereby generating multi-stage degradation samples; The dynamic feature identification unit is used to fuse time step information, distinguish the differences between samples generated at different stages and real data during the training process, and improve the reverse recovery capability of the disturbance generation unit through an adversarial optimization mechanism.

4. The method for estimating the health status of a lithium battery based on disturbance recovery enhancement and cross-domain adaptation mechanism according to claim 2, characterized in that: After the perturbation generation unit is trained, an initial pseudo sample is generated from random Gaussian noise, and its characteristic structure is reconstructed through an inverse recovery mechanism.

5. The method for estimating the health status of a lithium battery based on disturbance recovery enhancement and cross-domain adaptation mechanism according to claim 4 is characterized in that: The reverse recovery mechanism is used to gradually remove noise and reconstruct the original data features until it is restored to a noise-free state and a pseudo sample is obtained; A neighborhood feature matching strategy is adopted to calculate the feature similarity between the pseudo samples and the original samples, determine the corresponding health status labels, and merge the pseudo samples with the original dataset to construct an extended dataset for subsequent training.

6. The method for estimating the health status of a lithium battery based on disturbance recovery enhancement and cross-domain adaptation mechanism according to claim 1, characterized in that: Extract sequence features and estimate health status of redistributed data, including: A selective state space sequence model including multi-layer nonlinear transformation and state control mechanism is used to mine the time series features of redistributed data.

7. The method for estimating the health status of a lithium battery based on disturbance recovery enhancement and cross-domain adaptation mechanism according to claim 6, characterized in that: The selective state space sequence model includes a deep dynamic filtering unit, an information screening unit, a multi-scale dependency capturing unit and a health state estimation unit which are connected in sequence.

8. The method for estimating the health status of a lithium battery based on disturbance recovery enhancement and cross-domain adaptation mechanism according to claim 7, characterized in that: The deep dynamic filtering unit processes the input sequence data based on an adaptive filtering strategy; The information screening unit screens the processed data through a gated linear structure; The multi-scale dependency capture unit extracts local and global temporal dependency features of the time series in parallel in multiple feature spaces; The health state estimation unit performs aggregate analysis on the extracted time series dependency features and outputs a preliminary prediction result of the health state of the lithium battery.

9. The method for estimating the health status of a lithium battery based on disturbance recovery enhancement and cross-domain adaptation mechanism according to claim 1, characterized in that: Feature distribution alignment is performed based on the distribution mean difference measurement method to obtain the feature distribution difference, including: calculating the feature distribution difference between the source domain and target domain samples in the high-dimensional feature space, and using an adaptive matching strategy based on the measurement results to dynamically adjust the parameters in the feature extraction process to minimize the feature distribution difference; Among them, during the feature distribution alignment process, the health state estimation deviation is dynamically monitored; When the estimation error exceeds the preset threshold, the weight of the distribution mean difference metric is automatically adjusted, the feature distribution difference is recalculated and the parameters are updated until the prediction error converges to the preset range.

10. A lithium battery health status estimation system based on disturbance recovery enhancement and cross-domain adaptation mechanism, applying a lithium battery health status estimation method based on disturbance recovery enhancement and cross-domain adaptation mechanism according to any one of claims 1 to 9, characterized in that: include: The data redistribution module is used to collect raw data of lithium batteries with health status labels and generate redistributed data through the data perturbation enhancement model; The feature extraction and estimation module is used to extract sequence features and estimate the health status of the redistributed data, output the preliminary prediction results of the health status of the lithium battery, and calculate the health status estimation deviation; The cross-domain knowledge adaptation optimization module is used to align feature distributions based on the distribution mean difference measurement method to obtain feature distribution differences; The result output module is used to construct a joint loss function based on the health state estimation deviation and the feature distribution difference, and obtain the final lithium battery health state prediction result by iteratively updating parameters.

Images