Battery health status assessment method and device considering enhanced causal relationship

Abstract

The present invention provides a battery health status assessment method and device that considers causal enhancement. The method includes the following steps: obtaining battery measurement data, obtaining multi-time series nonlinear causal relationships between battery health status and health factors based on a causal inference method, and constructing a common battery multi-time series nonlinear causal relationship directed graph; establishing a causal relationship enhancement model based on causal-observation fusion deduction based on the battery multi-time series nonlinear causal relationship directed graph and a battery health factor and health status deduction method based on causal cumulative response, and obtaining battery multi-time series nonlinear causal enhancement features; using the causal enhancement features as input, and outputting the battery health status using a causal enhancement-bidirectional gated recurrent unit model, thereby improving the interpretability, accuracy, and generalization of health status estimation.

Description

Technical Field

[0001] The present invention belongs to the technical field of battery health status assessment, and relates to a battery health status assessment method and device, and in particular to a battery health status assessment method and device that considers enhanced causal relationships. Background Art

[0002] With the rapid development of new power systems, various new energy storage technologies, represented by batteries, are becoming increasingly widespread to support high penetration of renewable energy into the power system. During cycling, the loss of ions and active materials within batteries leads to continuous aging and performance degradation, potentially causing safety incidents such as battery leakage, insulation damage, and local short circuits. Accurately estimating battery health status and real-time monitoring of battery capacity degradation provide parameter support for battery safety warnings, fault diagnosis, and balancing, ensuring safe and reliable battery operation.

[0003] Existing battery health status estimation methods mainly include model-based estimation methods and data-driven estimation methods. Model-based estimation methods describe battery aging behavior by building a spatial state model, but in reality, electrochemical mechanisms are complex and changeable, making practical application difficult. To avoid complex electrochemical mechanism analysis and solutions, most current research uses data-driven methods to estimate battery health status. The data-driven method measures battery parameters such as current, voltage, and temperature, extracts health factors that are highly correlated with changes in health status, and uses them as training data to build an estimation model, thereby achieving health status estimation.

[0004] Existing research shows that there are significant differences in the capacity degradation curves of batteries under different operating conditions, which places higher demands on the generalization ability of data-driven models. Under the condition of sufficient data samples, data-driven models based on correlation analysis built for specific operating conditions can, to a certain extent, guarantee the generalization ability and accuracy within the distribution. However, for scenarios with unknown operating conditions such as newly built energy storage power stations, traditional data-driven models based on correlation analysis are prone to pseudo-correlation problems, resulting in insufficient interpretability of the model and affecting its generalization ability outside the distribution. To address the above problems, some studies use attention mechanisms to calculate the contribution of each health factor in health state estimation under different operating conditions of the battery, in order to improve the interpretability and generalization ability of health state estimation. However, health state estimation methods based on attention mechanisms still rely on correlation analysis to weigh the contribution of health factors, fail to fundamentally solve the pseudo-correlation problem, and limit the cross-condition generalization ability of battery health state estimation.

[0005] In recent years, causal inference theory has provided new ideas for solving the above problems. Compared with traditional data-driven methods that rely solely on correlation analysis, causal inference can reveal the intrinsic mechanism between variables and effectively avoid the problem of spurious correlation, thereby improving the interpretability and cross-operating generalization ability of battery health state estimation. Causal inference has been applied to fields such as power load and new energy forecasting, and power system state estimation. However, in the field of battery health state estimation, existing research mainly applies causal inference to health factor screening, and fails to fully incorporate causal relationships containing deep causal information such as causal graphs and causal effects into the health state estimation model. As a result, the accuracy and generalization ability of health state estimation methods that take causal relationships into account still have room for improvement. Therefore, it is necessary to develop battery health state estimation methods that consider causal relationship enhancement. Summary of the Invention

[0006] To solve the above problems, the present invention proposes a battery health status assessment method and device that considers causal relationship enhancement. The causal inference method is used to obtain the multi-time series nonlinear causal relationship between the battery health status and health factors, and the battery multi-time series nonlinear causal enhancement features are generated through causal-observation fusion deduction. On the basis of the battery multi-time series nonlinear causal enhancement features, the battery health status is estimated based on a bidirectional gated recurrent unit network to improve the interpretability, accuracy and generalization of the health status estimation.

[0007] The technical solution adopted in the present invention is as follows:

[0008] A battery health status assessment method considering causal relationship enhancement includes the following steps: obtaining battery measurement data, obtaining multi-time series nonlinear causal relationships between battery health status and health factors based on a causal inference method, and constructing a common battery multi-time series nonlinear causal relationship directed graph; establishing a causal relationship enhancement model based on causal-observation fusion deduction according to the battery multi-time series nonlinear causal relationship directed graph and a battery health factor and health status deduction method based on causal cumulative response, and obtaining battery multi-time series nonlinear causal enhancement features; taking the causal enhancement features as input, and outputting the battery health status using a causal enhancement-bidirectional gated recurrent unit model.

[0009] Furthermore, the battery measurement data includes the mean, minimum, maximum, standard deviation and integral of the voltage, current and temperature measurement data in each charge and discharge cycle, as well as the time taken for each charge and discharge cycle.

[0010] Furthermore, the causal inference method is used to obtain the multi-time series nonlinear causal relationship between the battery health state and the health factor, specifically including: using the potential Peter-Clark instantaneous conditional independence algorithm to discover the causal graph between the battery health state and the health factor; using the optimal adjustment set theory based on the causal graph to estimate the causal effect of the nonlinearity between the battery health state and the health factor, and combining the causal graph discovery results and the causal effect estimation results to obtain the multi-time series nonlinear causal relationship between the battery health state and the health factor.

[0011] Furthermore, the common multi-time-series nonlinear causal relationship directed graph of batteries includes a node set and an edge set, wherein the node set is composed of a health factor set and a battery health status indicator with common causality, and the edge set describes the multi-time-series nonlinear common causal relationship between the battery health status and the health factors.

[0012] Furthermore, according to the directed graph of multi-time-series nonlinear causality of batteries and the method for deducing battery health factors and health states based on causal cumulative responses, a causal enhancement model based on causal-observation fusion deduction is established to obtain multi-time-series nonlinear causal enhancement features of batteries. The specific steps include: establishing a battery health state and health factor deduction model based on causal cumulative responses according to the directed graph of multi-time-series nonlinear causality of batteries, and using the deduction model to obtain the health state and health factor deduction values ​​within each charge and discharge cycle according to the historical health state and health factor observation values; with the goal of minimizing the difference between the health factor deduced value and the health factor observation value, introducing the optimal health state proxy value to replace the unknown health state observation value, and constructing a causal enhancement model based on causal-observation fusion deduction; solving the causal enhancement model to obtain the deduced values ​​of the battery health state and health factor, which together constitute the multi-time-series nonlinear causal enhancement features of batteries.

[0013] Furthermore, the inference model starts from the health state and health factor observations of the first charge and discharge cycle, and calculates the average causal response of each health state and health factor in each charge and discharge cycle in chronological order as affected by all other health states or health factors at historical moments.

[0014] Furthermore, a loss function based on the invariant risk minimization theory is adopted in the training process of the causal enhancement-bidirectional gated recurrent unit model.

[0015] Furthermore, before the causal enhancement feature is input, it needs to undergo domain calibration batch normalization processing, and each different operating condition has independent batch normalization parameters.

[0016] Furthermore, for battery samples with known operating conditions, the batch normalization parameters corresponding to the operating conditions are directly applied for processing; for battery samples with unknown operating conditions, their similarity distance with all known operating conditions is calculated, and the batch normalization parameters corresponding to the operating condition with the smallest distance are selected for processing.

[0017] A computer device, comprising:

[0018] one or more processors;

[0019] a memory for storing one or more programs;

[0020] When the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned battery health status assessment method considering enhanced causal relationships.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] 1. Traditional data-driven methods based on correlation analysis suffer from insufficient interpretability and struggle to ensure the accuracy and generalizability of battery health state estimation beyond training samples. This paper establishes a battery causal relationship enhancement model based on causal-observation fusion deduction, constructing multi-time series nonlinear causal enhancement features from battery measurement data, improving the interpretability and in-distribution generalization of health state estimation.

[0023] 2. Based on the multi-time series nonlinear causal enhancement characteristics of batteries, this paper proposes a lithium-ion battery health state estimation method based on a bidirectional gated recurrent unit network and invariant risk minimization theory, which further improves the accuracy and distribution generalization of health state estimation. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figures 1 to 4 This is a multi-time series nonlinear causal relationship diagram between the health status inferred by different batteries and the battery health factor in an embodiment of the present invention.

[0025] Figure 5 This is a comparison chart of the actual observed values ​​of the B0005 battery health status and discharge current integral and the causal-observation fusion deduction results in an embodiment of the present invention.

[0026] Figure 6-Figure 7 This is a comparison chart of the t-distributed stochastic nearest neighbor embedding (t-SNE) dimensionality reduction visualization results of the original health factor feature distribution and causal enhancement feature distribution of different batteries in an embodiment of the present invention.

[0027] Figures 8 to 15This is a comparison chart of the health status estimation curves of the causal enhancement-bidirectional gated recurrent unit in an embodiment of the present invention and the comparison method in 29 groups of test batteries (in the figure, SVR, LSTM, BRU, BiLSTM, BiGRU, GC-LSTM, and CE-BiGRU are support vector regression, long short-term memory network, gated recurrent unit, bidirectional long short-term memory network, bidirectional gated recurrent unit, Granger causal-long short-term memory network, and causal enhancement-bidirectional gated recurrent unit, respectively).

[0028] Figure 16 This is a comparison chart of battery health status estimation results of different methods B0042 in an embodiment of the present invention.

[0029] Figures 17 to 19 This is a comparison chart of battery health status estimation results under three typical working conditions using different methods in an embodiment of the present invention. DETAILED DESCRIPTION

[0030] The following uses lithium-ion batteries as an example to further clearly and in detail describe the technical solution of the present invention in conjunction with the accompanying drawings.

[0031] A battery health status assessment method considering causal relationship enhancement includes the following steps: obtaining battery measurement data, obtaining multi-time series nonlinear causal relationships between battery health status and health factors based on a causal inference method, and constructing a common battery multi-time series nonlinear causal relationship directed graph; establishing a causal relationship enhancement model based on causal-observation fusion deduction according to the battery multi-time series nonlinear causal relationship directed graph and a battery health factor and health status deduction method based on causal cumulative response, and obtaining battery multi-time series nonlinear causal enhancement features; taking the causal enhancement features as input, and outputting the battery health status using a causal enhancement-bidirectional gated recurrent unit model.

[0032] First, to improve the interpretability and in-distribution generalization of battery health state estimation, this paper studies the multi-time series nonlinear causal relationship between battery measurement data and battery health state, establishes a battery causal relationship enhancement model based on causal-observation fusion deduction, and thus constructs battery multi-time series nonlinear causal enhancement features. The specific method is as follows:

[0033] Relevant research on the lithium-ion capacity attenuation process shows that the operating conditions such as voltage, current, temperature and time in the lithium-ion charge and discharge cycle have an important impact on the health status of the battery throughout the entire life cycle. Specifically, if the battery operating temperature is too high, it will accelerate the internal side reactions of the battery, while if it is too low, it will lead to lithium deposition and loss of active materials; if the battery charging voltage is too high, it will cause overcharging and trigger problems such as electrolyte decomposition and lithium deposition. If the discharge cut-off voltage is too low, it will cause over-discharge and loss of active materials; if the battery charge and discharge rate is too high, it will also cause active material loss and temperature increase, thereby accelerating internal side reactions. Based on the above analysis, in order to avoid the lack of explainability of causal relationships caused by complex processing of battery measurement data, starting from the basic statistical characteristics of voltage, current, temperature and time measurement data, the mean, minimum, maximum, standard deviation and integral of the voltage, current and temperature measurement data in each charge and discharge cycle, as well as the time of each charge and discharge cycle, are extracted to form the battery health factor time series data to be screened, which is used for subsequent causal inference and battery health status estimation. The time series data of battery sample i to be screened can be obtained as shown below :

[0034]

[0035] Where, The nth health factor data of battery sample i in the tth charge and discharge cycle; is the set of charge and discharge cycles of battery sample i; A set of battery health factors to be screened; A collection of battery samples.

[0036] To investigate the multi-time series nonlinear causal relationship between the battery state of health (SOH) and battery health factors (BHFs), appropriate causal inference methods are needed to analyze this causal relationship. The capacity degradation process of lithium-ion batteries is often accompanied by the continuous shedding and embedding of solid-state electrodes, as well as the simultaneous oxidation and reduction reactions of the positive and negative electrodes. This leads to the possibility of a multi-time series nonlinear causal relationship between the BOH and BHFs, including unobserved health factors. Causal inference typically involves two steps: causal graph discovery and causal effect estimation. Time series variable causal discovery methods, such as Granger causality-based methods and constraint-based methods, can be used to discover the causal graph between the BOH and BHFs. Considering that the simplified causal graphs generated by Granger causality-based methods generally focus on the causal relationship between the BOH and BHFs from a global perspective, while the window causal graphs generated by constraint-based methods can clearly demonstrate the dynamic causal relationship between the BOH and BHFs across multiple time series, constraint-based methods are more suitable for discovering multi-time series causal graphs between the BOH and BHFs during battery degradation. The constrained latent Peter-Clark instantaneous conditional independence algorithm is an extension of the Peter-Clark algorithm and the instantaneous conditional independence test in the presence of potential unobserved variables. It can effectively discover multi-time series causal graphs for batteries under the influence of potential unobserved health factors. For estimating the causal effect between the battery state of health and battery health factors, methods such as propensity score-based methods and regression-based methods can be used. Considering that the estimation results of propensity score-based methods are generally expressed as the expected probability of the battery state of health or battery health factor values, while regression-based methods can explicitly estimate the nonlinear influence of battery health factors on battery state of health values, regression-based methods are more interpretable in estimating the nonlinear causal effect between the battery state of health and battery health factors. The optimal adjustment set theory based on regression can accurately estimate the nonlinear causal effect between the battery state of health and battery health factors by minimizing the asymptotic variance of the causal effect estimate based on a nonlinear nonparametric estimator.

[0037] In summary, the present invention adopts a battery causal inference method based on the latent Peter-Clark instantaneous conditional independence algorithm and the optimal adjustment set theory to obtain the multi-time series nonlinear causal relationship between the battery health state and the battery health factors under the influence of unobserved health factors, providing a basis for the subsequent construction of battery multi-time series nonlinear causal enhancement features. Among them, the battery causal graph discovery based on the latent Peter-Clark instantaneous conditional independence algorithm is mainly achieved through three main steps: first, the possible causal connection between the battery health state and the battery health factors to be screened is preliminarily identified through a conditional independence test based on nonlinear Gaussian process distance correlation; second, the causal direction is determined based on the instantaneous conditional independence test and the time series relationship; finally, the results of the first two steps are combined to construct a complete battery multi-time series causal graph taking into account the influence of unobserved health factors. The battery causal effect estimation based on the optimal adjustment set theory is mainly achieved through three steps: first, according to the battery multi-time series causal graph obtained by the potential Peter-Clark instantaneous conditional independence algorithm, all battery health factors that have deterministic causal paths with the battery health status are screened; secondly, the optimal adjustment set theory is applied to identify the optimal adjustment set that can simultaneously block the battery non-causal path and minimize the estimation variance; the random forest nonlinear nonparametric regression estimator is used to estimate the nonlinear causal effect between the screened battery health factors and the battery health status; finally, the battery multi-time series causal graph discovery results and the nonlinear causal effect estimation results are combined to form a multi-time series nonlinear causal relationship between the battery health status and the battery health factors.

[0038] For the i-th battery sample, according to the battery causal inference method based on the potential Peter-Clark instantaneous conditional independence algorithm and the optimal adjustment set theory, the battery health factors that have a causal path with the battery health status can be screened, and the multi-time series nonlinear causal relationship between the battery health status and the screened battery health factors can be obtained. The above process expression is as follows:

[0039]

[0040] Where, for representing a battery of causal inference methods based on the latent Peter-Clark instantaneous conditional independence algorithm and optimal tuning set theory; is the time series data of the health factor of battery sample i that has a deterministic causal path with the battery health status after screening by the causal inference method, is the set of battery health factors that have a deterministic causal path between battery sample i and battery health status; is the actual health status time series data of battery sample i; It is a directed graph of multi-time series nonlinear causal relationships between the battery health status and health factors obtained after battery sample i is identified by the causal inference method.

[0041] Considering that new batteries cannot provide health status time series data in actual situations to infer the causal relationship between health status and health factors, relevant research shows that there are some common causal relationships between battery health status and health factors under different operating conditions. Therefore, a causal relationship directed graph can be obtained from different battery samples with known health status time series data. Extract a common causal directed graph , estimating the battery health status based on the common causal relationship helps to improve the generalization within the distribution. Based on this assumption, a directed graph of multi-time series nonlinear causal relationships of batteries with common characteristics can be obtained as follows:

[0042]

[0043] In the formula, the multi-series nonlinear causal relationship directed graph of batteries with common characteristics is A collection of nodes A set of health factors that have been screened based on common causal factors and battery health indicators Composition, of which The set of health factors after screening derived from each battery sample i ; Edge set Describes the multi-time series nonlinear common causal relationship between battery health status and health factors. Indicates that the mth battery indicator has a nonlinear common causal effect on the nth indicator with a time lag of τ, and the estimator By calculating the nonlinear nonparametric estimator based on random forest in different battery samples The nonlinear common causal effect is characterized by the average value of , which quantifies the expected value of the corresponding n-th indicator of the m-th indicator of the distribution after the time lag τ; is the common causal threshold, which represents a causal relationship in the battery sample The minimum threshold for the number of occurrences in .

[0044] Considering that existing research mainly applies causal inference theory to health factor screening, and fails to fully incorporate causal relationships containing deep causal information such as causal graphs and causal effects into health state estimation models, the present invention adopts a battery health state and health factor deduction method based on causal cumulative response to fully capture the battery multi-time series nonlinear causal relationship directed graph The method uses iterative cumulative causal responses as its core idea. Starting from the health state and health factor observations of the first charge and discharge cycle, the average causal response of each health state and health factor within each charge and discharge cycle is calculated in chronological order, influenced by all other health states or health factors at the same time. For the tth charge and discharge cycle of battery sample i, the battery health state and health factor deduction results based on the causal cumulative response are expressed as follows:

[0045]

[0046] Where, and are the deduction results of the battery health status and the nth health factor in the tth charge and discharge cycle of battery sample i, respectively. Their initial values ​​are and are the initial observation values ​​of battery health status and health factor, respectively and ; and They are used to characterize the battery health state and health factor deduction method based on causal cumulative response, respectively. The input is the deduction results of all battery health states and health factors before the t-1th charge and discharge cycle of battery sample i, and the output is the deduction results of the battery health state and the nth health factor in the tth charge and discharge cycle; and They represent the total number of battery health states or health factors that have a time-lagged causal effect on the battery health state and the nth health factor in the tth charge-discharge cycle of battery sample i; The maximum time delay step considered in the causal inference process.

[0047] Ideally, if the battery's state of health and health factor under different operating conditions strictly adhere to the same and complete battery multi-time series nonlinear causal relationship, the battery health factor and health state deduction results based on the causal cumulative response should be completely consistent with their observed values. However, there are three key limitations in reality: First, there are still some differences in the battery state of health-health factor causal relationship under different operating conditions. This is due to heterogeneity caused by factors such as battery material composition, manufacturing process and environmental conditions; second, some potential health factors cannot be observed, resulting in incomplete inferred causal relationships. These unobserved variables may include factors that are difficult to measure directly, such as changes in the internal microstructure of the battery and decomposition of electrolyte components; third, causal cumulative deduction may have cumulative bias. Early minor deviations will continue to amplify as the deduction process progresses, ultimately causing the later deduction results to deviate significantly from the actual values. These three limitations make the battery health factor and health state deduction method based solely on causal cumulative response insufficiently accurate and difficult to meet the needs of actual applications. At the same time, if the health status is estimated only by relying on the observed value of the battery health factor, it is impossible to fully utilize the intrinsic mechanism information provided by the causal relationship, lack of interpretability and generalization within the distribution, especially when the battery operating conditions vary greatly. Therefore, in order to solve the above problems, the present invention combines causal reasoning with observation data to establish a causal relationship enhancement model based on causal-observation fusion reasoning to form a multi-time series nonlinear causal enhancement feature. The model obtains the multi-time series nonlinear causal relationship between the battery health status and the health factor. Combined with actual observation data, the proposed method introduces adjustable weight parameters to proactively balance the fusion ratio of causal inference results with actual observations based on data quality and causal relationship strength. This approach, to a certain extent, corrects for causal differences between individual batteries, compensates for the impact of potential unidentified causal relationships, and suppresses the spread of cumulative errors. The constructed multi-time series nonlinear causal enhancement feature simultaneously retains the inherent explanatory power of causal relationships and the real-time information of observational data, thereby achieving a balance between accuracy, interpretability, and within-distribution generalization in battery health state estimation. Furthermore, given that battery health state observations are often unknown in real applications, this poses a challenge to causal enhancement models based on causal-observation fusion inference. To address this issue, the present invention minimizes the gap between the health factor inference results based on causal-observation fusion inference and the health factor observations by finding an optimal health state proxy value to replace the unknown health state observations. Because both the health state inference results and the health factor inference results are derived from causal-observation fusion inference, the theoretical gap between these health state inference results and the actual health state observations is small. Therefore, the health state inference results and health factor inference results obtained by causal-observation fusion inference can be used as enhanced features to estimate health state observations, thereby improving the interpretability and in-distribution generalization of health state estimation.

[0048] In summary, based on the multi-time series nonlinear causal relationship directed graph between the obtained battery health status and health factors Based on the battery health factor and health state deduction method based on causal cumulative response, a causal relationship enhancement model based on causal-observation fusion deduction is established to obtain the battery multi-time series nonlinear causal enhancement characteristics composed of battery health state deduction results and health factor deduction results. The expression of the causal relationship enhancement model based on causal-observation fusion deduction is as follows:

[0049]

[0050] Wherein, the objective function is to minimize the sum of squares of the difference between the battery health factor deduction result and the observed value. The constraint condition is the causal-observation fusion deduction equation. The heuristic algorithm can be used to solve the nonlinear model to obtain the multi-time series nonlinear causal enhancement characteristics of battery sample i. ; and They are the battery health status deduction result of the tth charge and discharge cycle and the nth health factor deduction result of battery sample i based on causal-observation fusion deduction, which together constitute its multi-time series nonlinear causal enhancement feature ; It is the proxy value of all battery health status before the t-1th charge and discharge cycle of battery sample i. is the weight used to weigh the causal relationship and the proportion of observation data in the causal-observation fusion deduction, and its value range is (0,1). When it is close to 0, the deduction results tend to rely on causal relationships. When it is close to 1, the deduction results tend to adopt the observed value. When it is between 0 and 1, the inference result is a weighted combination of causal inference and observations. The value is suitable for the case where the causal relationship is stable and the observation noise is large. The value is applicable to situations where the causal relationship is incomplete but the observation data is accurate. The value can be optimized for specific batteries and application scenarios through cross-validation.

[0051] Furthermore, considering that the battery health state has obvious time series characteristics and the battery capacity attenuation is closely related to the historical degradation curve, the present invention constructs a multi-time series nonlinear causal enhancement feature of the battery. Based on this, the battery health status is estimated using a causal enhancement-bidirectional gated recurrent unit model to improve the accuracy and distribution generalization of battery health status estimation. The specific method is as follows:

[0052] As a variant of the recurrent neural network, the standard gated recurrent unit network alleviates the gradient vanishing problem by introducing reset gates and update gates, and can effectively extract time series features. The bidirectional gated recurrent unit combines the outputs of the forward and reverse gated recurrent units, can simultaneously obtain historical and future information, and enhance the ability to predict time series data. In the unidirectional gated recurrent unit layer, the tth charge and discharge cycle of the battery sample i is mapped to the multi-time series nonlinear causal enhancement feature. As input, the corresponding candidate hidden state of the gated recurrent unit as follows:

[0053]

[0054] Where tanh is the hyperbolic tangent activation function; 、 and are the input weight matrix, the cyclic weight matrix, and the bias respectively; is the hidden state of the previous gated recurrent unit; Reset the gate for the current Gated Recurrent Unit. Gated Recurrent Unit Hidden State Updates as follows:

[0055]

[0056] Where, Update the gate for the current gated recurrent unit. The bidirectional gated recurrent unit concatenates the outputs of the forward gated recurrent unit and the reverse gated recurrent unit to obtain the final output, as shown in the following expression:

[0057]

[0058] Where, and are the forward and reverse hidden states of the tth charge and discharge cycle, respectively; and They are forward and reverse gated recurrent units respectively. The last hidden state of the bidirectional gated recurrent unit By mapping through a fully connected neural network, the estimated health status of the lithium-ion battery of the tth charge and discharge cycle of battery sample i can be obtained.

[0059] Considering that batteries exhibit different degradation patterns under different operating conditions, this operating condition heterogeneity is one of the main challenges in battery health state estimation. Existing studies have shown that different operating conditions can be regarded as different data distributions, but there is some common causal relationship between the battery health state and health factors under different operating conditions. Therefore, the present invention introduces the invariant risk minimization theory to improve the distribution out-generalization ability of the health state estimation model. The traditional model learning paradigm based on empirical risk minimization theory focuses on minimizing the average loss of training data, and often performs poorly when there are systematic differences between the test distribution and the training distribution. Invariant risk minimization can capture the causal structure of the data rather than the specific correlation of the operating conditions by learning an invariant predictor that performs well in all operating conditions, thereby maintaining good performance in unseen operating conditions. For the battery health state estimation problem, the present invention adopts a causal enhancement-bidirectional gated cyclic unit battery health state estimation model learning method based on the invariant risk minimization theory to learn the health state estimation rules that are valid in different operating conditions. The objective function of the causal enhancement-bidirectional gated cyclic unit model based on the invariant risk minimization theory is as follows:

[0060]

[0061] Wherein, the objective function is to minimize the training loss of the causal reinforcement-bidirectional gated recurrent unit with an invariant risk minimization regularization term. The first term is the standard empirical risk minimization loss, which ensures that the causal reinforcement-bidirectional gated recurrent unit has good average performance under different battery operating conditions. The second term is the invariant risk minimization regularization term. By calculating the sum of the squared norms of the gradients of the loss function with respect to the virtual classifier parameters under each operating condition, the model is encouraged to learn causal feature representations that are invariant across operating conditions. represents the parameters of the battery health status estimation model based on causal enhancement-bidirectional gated recurrent unit; Estimating the training error of the battery health status model, usually using the mean square error function; is the weight parameter that balances the empirical risk minimization objective and the invariant risk minimization regularization term; is a virtual linear classifier parameter used to verify the stability of the model representation under various operating conditions. When the invariant risk minimization regularization term approaches zero, it indicates that the model has learned a near-optimal prediction rule across all operating conditions, which is beneficial for improving the out-of-distribution generalization of battery health state estimation.

[0062] Considering that the degradation characteristics of batteries under different operating conditions are significantly different, the constructed battery multi-time series nonlinear causal enhancement feature There may be distribution shifts between different operating conditions. Traditional batch normalization methods usually assume that all training data come from the same distribution and cannot effectively handle the differences in feature distributions of data from different battery operating conditions. To solve this problem, based on the causal enhancement-bidirectional gated cyclic unit battery health state estimation model learning method based on the invariant risk minimization theory, the present invention introduces a domain calibration batch normalization method to maintain independent normalization parameters for different operating conditions, and handle unknown operating conditions through a feature similarity matching mechanism. For battery sample i with known operating conditions, its multi-time series nonlinear causal enhancement features are calibrated based on the domain calibration batch normalization method. The normalization process is as follows:

[0063]

[0064] Where, Multi-time series nonlinear causal enhancement features for battery sample i The results are normalized by the domain calibration batch normalization method; and is a learnable affine parameter for the specific operating condition of battery sample i, which is used to restore the expressive power of the normalized features; and is the multi-time series nonlinear causal enhancement feature under the working condition of battery sample i The mean and variance of To prevent small constants from dividing by zero. For battery samples under unknown operating conditions outside the distribution, the causal enhancement-bidirectional gated recurrent unit model needs to dynamically select the most appropriate batch normalization parameters. Therefore, the present invention adopts an operating condition matching method based on the similarity of battery causal enhancement feature distribution, calculates the feature distribution distance between unknown operating condition samples and known operating condition samples, and selects the batch normalization parameters of the most similar operating condition for application. The expression of the causal enhancement feature similarity distance between battery samples is as follows:

[0065]

[0066] Where, is the causal enhancement feature similarity distance between battery sample i and battery sample j; is the variance difference weight, which is used to balance the importance of mean difference and variance difference in the calculation of working condition similarity. When faced with a battery sample with unknown working condition, by calculating its similarity distance with all known working conditions, the batch normalization parameters corresponding to the working condition with the smallest distance are selected as follows:

[0067] Where k is the index of the known operating condition battery sample that is most similar to the causal enhancement feature distribution of the unknown operating condition battery sample j. During the training phase, domain calibration batch normalization maintains independent normalization parameters for each known operating condition. During the testing phase, for battery samples with known operating conditions, the domain calibration batch normalization parameters of the corresponding operating condition are directly applied; for battery samples with unknown operating conditions, the similarity of their causal enhancement feature distribution with each known operating condition is first calculated, and then the domain calibration batch normalization parameters of the most similar operating condition are applied. This operating condition matching method based on the similarity of battery causal enhancement feature distribution enables the causal enhancement-bidirectional gated recurrent unit network model to adaptively process the differences in feature distributions inside and outside the distribution, further improving the generalization ability of health state estimation under unknown conditions.

[0068] In summary, the causal enhancement-bidirectional gated recurrent unit model in the present invention can fully utilize the intrinsic mechanism information provided by multi-time series nonlinear causal enhancement features, capture time series features and long-term dependencies through bidirectional gated recurrent units, and improve the generalization ability of the model under unknown working conditions through invariant risk minimization theory and domain calibration batch normalization method, thereby achieving accurate, interpretable and well-generalized health status estimation.

[0069] The technical effects of the present invention are further illustrated below using a NASA lithium-ion battery dataset. This dataset contains cyclic charge-discharge data from multiple sets of lithium-ion batteries under different operating conditions. Each set of batteries undergoes a complete aging process from brand new to capacity decay to 80% or 70% of the rated capacity. In this example, four battery data sets, B0005, B0006, B0007, and B0018, were selected to construct training and validation sets (the training and validation sets were divided in a 4:1 ratio, and both training and validation sets were acquired by sampling data at equal intervals along the complete degradation curve). A total of 29 battery data sets, including B0025 to B0056, were selected as the test set. Each battery data set includes measurement data such as voltage, current, temperature, time, and capacity during the charge and discharge process.

[0070] Table 1 lists the operating parameters for each battery group. Each group used the same charging method: first charging to 4.2V at a constant current of 1.5A, then charging at a constant voltage until the current dropped to 20mA. The differences in operating conditions between the different groups were primarily reflected in ambient temperature, discharge cutoff voltage, and discharge current. Ambient temperatures included room temperature (24°C), low temperature (4°C), and high temperature (43 / 44°C). Discharge currents included 1A, 2A, 4A, and a square wave (4A amplitude, 50% duty cycle, 0.05Hz frequency). Cutoff voltages included 2.0V, 2.2V, 2.5V, and 2.7V.

[0071] Table 2 lists the parameter settings, search space and determination method of the battery health state estimation model constructed based on the causal enhancement-bidirectional gated loop unit. The causal relationship enhancement model based on causal-observation fusion deduction constructed in this embodiment is solved using a double annealing algorithm. The double annealing algorithm is a random global optimization algorithm that combines simulated annealing and local search strategies. It has the advantages of strong global optimization ability and fast convergence speed. The double annealing algorithm sets the initial temperature to 5230.0, the annealing restart temperature ratio to 0.00002, and the convergence criterion is to stop when the maximum number of iterations is reached or the no improvement condition is triggered during the local search process. The causal enhancement-bidirectional gated loop unit model is modeled and tested using the Pytorch architecture of Python 3.8.

[0072] A number of potential health factors were extracted from the battery measurement data, including the mean, minimum, maximum, standard deviation, and integral of the voltage, current, and temperature measurements, as well as the duration of each charge and discharge cycle. Using the latent Peter-Clark transient conditional independence algorithm and optimal adjustment set theory, causal inference was performed on the multi-temporal nonlinear causal relationships between the health states and the battery health factors of the four battery groups B0005, B0006, B0007, and B0018 in the training set, taking into account the influence of unobserved health factors. Figures 1 to 4 The multi-time series nonlinear causal relationship between the inferred health status and battery health factors of four battery groups B0005, B0006, B0007 and B0018 is shown respectively (SOH in the figure represents health status), among which the health factors that do not have a deterministic causal path with the health status have been screened out. It should be pointed out that Figures 1 to 4 The positive / negative causal effects marked on the causal direction arrows are calculated by the nonlinear causal effect estimator The change in output given an input ranging from 0 to 1 (normalized value) reflects the causal effect on the target variable of a unit intervention in the dependent variable.

[0073] from Figures 1 to 4It can be seen that in the three batteries B0005, B0006 and B0007, the discharge current integral shows a strong negative causal effect on the health status, which are -0.56, -0.58 and -0.53 respectively (-0.53 is an indirect effect), which indicates that the gradual decrease in cumulative charge transfer during the discharge process is a common mechanism leading to battery capacity degradation. It should be pointed out that the discharge current data in the NASA lithium-ion battery dataset used in this embodiment is presented as negative values, so the discharge current integral shows a negative causal effect on the health status. At the same time, all four groups of batteries show that the mutual influence between the health factor and the health status has an obvious time lag characteristic (ranging from t-1 to t-11), indicating that battery aging is a time-series accumulation process, and the changes in the health factor or health status of the current cycle will gradually show its influence in subsequent cycles. In addition, from Figures 1 to 4 It can also be seen that the causal relationship inferred for battery B0005 is the most complex of the four battery groups, containing 12 causal nodes and 11 causal paths, indicating that the degradation mechanism of this battery under its operating conditions is more complex than that of other operating conditions. Battery B0018 only exhibits a single causal path: "Healthy state (t) → Discharge current integral (t+1)", which may indicate that the battery is in a special aging stage, where changes in health state become the dominant factor guiding changes in health factors, rather than being influenced by health factors.

[0074] Take B0005 and B0007 batteries as an example. Figures 1 to 4The key causal paths in the battery were analyzed. The long-lag path for battery B0005, "minimum charge current (t-11) → single-discharge cycle duration (t-6) → integrated discharge current (t-1) → battery health status (t)," illustrates how abnormal charge current (possibly undercharging) can affect discharge duration, ultimately leading to a decrease in battery health status. The effect strength of this causal path is "+0.35 → -0.28 → -0.56," with a combined effect of approximately +0.055, indicating that insufficient charge current at an early stage can, through a complex pathway, ultimately slightly reduce battery health status. This long-span causal chain explains why some battery abnormalities may only manifest after a long time. The direct positive effect of the path for battery B0005, "single-discharge cycle duration (t-5) → battery health status (t)," is +0.26, indicating that appropriately extending discharge duration may benefit battery life, consistent with the battery usage recommendation of avoiding deep discharge. For the B0007 battery, the direct negative effect of the path "minimum charging voltage (t-3) → battery health state (t)" is -0.80. This strong negative effect indicates that excessively high charging starting voltage has a significant detrimental effect on battery life. This is consistent with the mechanism by which overcharging lithium-ion batteries can lead to problems such as electrolyte decomposition and lithium deposition. The causal path effect strength of the B0007 battery "health state (t) → charging current standard deviation (t+1)" is +0.34. This positive effect indicates that a decrease in health state leads to a decrease in the charging current standard deviation, possibly indicating that the battery loses its fast charging capability with aging, resulting in a flatter charging curve. As batteries age, due to the thickening of the solid electrolyte interface film and the loss of active materials, internal resistance and polarization are often observed, resulting in a flatter charging and discharging curve and reduced charging current fluctuations.

[0075] In summary, it can be seen that there is a significant causal cascade effect between battery health status and health factors. This complex causal structure with multiple nodes and multiple paths reveals the underlying physicochemical mechanism of lithium-ion battery degradation. The differences in causal structure under different battery operating conditions also indicate that relying solely on data correlation for health status estimation may not be able to effectively adapt to the battery degradation characteristics under different operating conditions. This phenomenon indicates that it is necessary to combine mechanism and data-driven methods to enhance feature distribution using causal structure, thereby improving the accuracy, generalization, and interpretability of battery health status estimation.

[0076] The present invention constructs a causal relationship enhancement model based on causal-observation fusion deduction and uses multi-time series nonlinear causal relationship to solve the multi-time series nonlinear causal enhancement feature of the battery. To verify the effectiveness of the causal enhancement feature, Figure 5 The comparison results of the actual observation values ​​of the B0005 battery health status and discharge current integral and the causal-observation fusion deduction results are shown in Figure 2. Figure 5It can be seen that the actual health status observation value of the B0005 battery shows a clear downward trend, gradually decreasing from the initial close to 1.0 to about 0.1, and contains more local capacity regeneration phenomena, which is in line with the general law of lithium-ion battery capacity degradation. The causal-observation fusion deduction results of the health status better retain the overall degradation trend of the actual health status, and to a certain extent restore the capacity regeneration phenomenon that appears in the observation value in the early stage (number of cycles <80), confirming the accuracy of the identified causal relationship. However, it is difficult to track the change pattern of the actual observation value in the middle and late stages (number of cycles >80). This may be because the battery has a degradation mechanism in the deep aging stage that is difficult to capture through the identified causal relationship. From Figure 5 It can also be seen that the actual discharge current integral observation value of the B0005 battery shows a clear upward trend. Since the actual value of the discharge current is negative, it actually reflects that the battery's dischargeable charge gradually decreases with the increase in cycles, which is also in line with the general law of lithium-ion battery capacity degradation. The causal-observation fusion deduction results of the discharge current integral can also well track the variation pattern of the actual observation value. Overall, although the causal-observation fusion deduction results are significantly different from the observed values ​​in some cycles, the differences between the health state and discharge current integral deduction results and the observed values ​​are similar in each charge and discharge cycle. In particular, around cycle number 80, the intersection point of the four curves in the fusion deduction results and the actual observation value almost coincides. This verifies the rationality of using the health state and health factor deduction results obtained by causal-observation deduction as causal enhancement features to estimate the battery health state.

[0077] To further analyze the distribution characteristics of causal enhancement features, Figure 6-Figure 7 The distribution of the original health factor features and causal enhancement features of different batteries after projection in the two-dimensional feature space is shown. This embodiment uses the t-distributed random neighbor embedding dimensionality reduction method to project the original health factor features and causal enhancement features into the two-dimensional space for visualization, and distinguish them by color according to the battery samples. Figure 6 、 Figure 7 It can be seen that compared with the original health factor features, the causal enhancement features show a more consistent distribution pattern across different batteries. Figure 6 In , there are obvious boundaries between the feature distributions of different batteries, indicating that the original features are more sensitive to individual differences of batteries. Figure 7 In the , this boundary is significantly weakened, and the distributions of different battery characteristics tend to cluster, indicating that the causal enhancement feature better captures the common health degradation mechanisms among different batteries. This shows that the causal enhancement feature is of great significance for improving the in-distribution generalization of the health state estimation model, helping to reduce the model's dependence on specific individual battery characteristics and enhance its ability to learn common degradation mechanisms.

[0078] The accuracy of battery health state estimation of the constructed causal enhancement-bidirectional gated recurrent unit model was evaluated. In order to verify the effectiveness of the battery health state estimation method based on the causal enhancement-bidirectional gated recurrent unit in the present invention, the following methods were selected for comparison: support vector regression, long short-term memory network, gated recurrent unit, bidirectional long short-term memory network, bidirectional gated recurrent unit and Granger causality-long short-term memory network. Among them, the Granger causality-long short-term memory network method screens health factors that have a direct causal effect on the health state through the Granger causality test, and then estimates the battery health state through the long short-term memory network. All models are evaluated using the root mean square error and mean absolute error as two indicators, and the average, minimum and maximum values ​​of each evaluation indicator in 29 groups of test batteries are calculated. Table 3 shows the overall performance comparison of the health state estimation of different methods on all test batteries. It can be seen from Table 3 that the causal enhancement-bidirectional gated recurrent unit method in the present invention is significantly better than the comparison method in all average and maximum evaluation indicators. Compared with the traditional machine learning method support vector regression, the Causal Enhancement-Bidirectional Gated Recurrent Unit (CA-BGRU) achieved mean and maximum root mean square error (RMS) reductions of 23.5% and 21.5%, respectively. Compared with the deep learning method BGRU based on correlation analysis, the reductions were 17.0% and 30.9%, respectively. Compared with the Granger Causality-Long Short-Term Memory (LSTM) network based on causality analysis, the reductions were 36.5% and 3.6%, respectively. This demonstrates that the CA-BGRU effectively improves the accuracy of health state estimation by integrating multi-temporal nonlinear causal enhancement features, a BGRU network, and invariant risk minimization theory. Notably, the CA-BGRU achieved a maximum RMS error of 0.7949 and a maximum mean absolute error of 0.7085, significantly lower than those of other methods. This is particularly important for capacity anomaly detection and safety early warning in lithium-ion battery management systems.

[0079] Figures 8 to 15 The comparison results of the health state estimation curves of the causal enhancement-bidirectional gated recurrent unit and the comparison method in 29 groups of test batteries are shown. Here, the test battery B0042 is used as an example to compare and analyze the health state estimation curves of different methods. Figure 16 According to the operating parameters of each battery group listed in Table 1, the B0042 battery operates under low temperature conditions, and the discharge current switches between 1A and 4A. Figure 16 It can be seen from the actual degradation curve of B0042 battery that there is a significant capacity drop and capacity regeneration phenomenon. Figure 16It can be seen that the causal enhancement-bidirectional gated recurrent unit method accurately tracks the health status of the B0042 battery throughout its entire battery life cycle. In particular, its estimation accuracy for the mid-term capacity drop (between 40 and 85 cycles) and the late-term capacity regeneration (after 85 cycles) is significantly superior to other methods. Health status estimation methods based on correlation analysis, such as the bidirectional gated recurrent unit, fail to identify the battery capacity drop in the mid-term. The Granger causality-long short-term memory network method based on causality analysis, while identifying the battery capacity drop in the mid-term, is less accurate than the causal enhancement-bidirectional gated recurrent unit for the capacity regeneration in the late-term. This demonstrates that the causal enhancement-bidirectional gated recurrent unit method not only outperforms the comparison methods in overall accuracy but also maintains stable estimation accuracy over long timescales, which is of great significance for long-term battery health status monitoring.

[0080] To verify the generalization of the proposed method under different operating conditions, the test batteries were divided into three categories based on the similarity between the operating conditions and the training set: similar operating conditions (ambient temperature consistent with the training set, discharge using a square wave current, but the average current consistent with the training set, including B0025, B0026, B0027, and B0028); moderately different operating conditions (ambient temperature consistent with the training set, but with significantly different discharge currents, or discharge current consistent with the training set, but with significantly different ambient temperatures, including 10 battery groups, B0033, B0034, B0036, B0049, B0050, B0051, B0053, B0054, B0055, and B0056); and significantly different operating conditions (ambient temperature of high, low, or alternating high and low temperatures, and discharge current different from the training set, including all remaining 15 battery groups in the test set). Table 4 shows the accuracy performance of different methods under these three operating conditions. Table 4 shows that all methods perform best under similar conditions and worst under conditions with moderate or significant differences, which is consistent with the generalization ability of models. However, the Causal Enhancement-Bidirectional Gated Recurrent Unit method exhibits the smallest performance degradation across different conditions, with the root mean square error increasing by only 0.2074 from similar to significant differences, while the Granger Causal-LSTM, Bidirectional Gated Recurrent Unit, and Support Vector Regression methods increase by 0.3709, 0.2908, and 0.3329, respectively. Notably, the Causal Enhancement-Bidirectional Gated Recurrent Unit outperforms all compared methods in both moderate and significant differences (0.3585 and 0.2476), fully demonstrating its superiority in improving cross-condition generalization.

[0081] Figures 17 to 19 The comparison of the health status estimation results of the causal enhancement-bidirectional gated recurrent unit and the comparative method on three typical working conditions of batteries is shown. Figure 17The battery in the middle is B0028 with similar working conditions. Figure 18 The middle one is the medium difference working condition battery B0056, Figure 19 The figure shows battery B0045 under significantly different operating conditions. As can be seen from the figure, on battery B0028 under similar operating conditions, both the causal enhancement-bidirectional gated recurrent unit (CA-BGRU) and other methods can effectively estimate health state changes, but the CA-BGRU has relatively better estimation accuracy. On battery B0056 under moderately different operating conditions, the comparison methods exhibit significant deviations in the middle and late stages (number of cycles > 50), while the CA-BGRU maintains high accuracy. On battery B0045 under significantly different operating conditions, the comparison methods exhibit large fluctuations and systematic deviations, while the CA-BGRU still accurately captures health state change trends. This demonstrates that under unknown or extreme operating conditions, the CA-BGRU learns more robust and generalizable battery degradation patterns by combining multi-series nonlinear causal enhancement features with invariant risk minimization theory.

[0082] In order to further verify the stability of the proposed method and guide the selection of parameters for practical applications, sensitivity analysis of key parameters was performed, including the common causal threshold The weight used to weigh the causal relationship and the proportion of observation data in causal-observation fusion deduction .

[0083] Table 5 shows the different and The impact on the accuracy of the battery health state estimation method based on causal enhancement-bidirectional gated recurrent unit. It can be seen that when When it increases from 0.2 to 0.8, the causal enhancement-bidirectional gated recurrent unit has a stricter requirement on the proportion of the extracted common causal relationships in the four groups of batteries in the training set. At this time, the root mean square error under all working conditions shows a trend of first increasing and then decreasing, and the lowest point of the root mean square error is at =0.2. Since the training set of this embodiment includes 4 groups of batteries, the value of this parameter indicates that the common causal relationship extracted is the union of the causal relationships of the 4 groups of batteries, which shows that adequately accommodating multiple causal relationships helps to extract cross-condition invariant features and enhance cross-condition generalization capabilities. When it increases from 0.1 to 0.9, it means that the causal enhancement-bidirectional gated recurrent unit gradually increases its attention to causal relationships when performing causal-observation fusion deduction. At this time, the root mean square error under all working conditions shows a trend of first decreasing and then increasing. The lowest point of the root mean square error under the significant difference working condition is at This means that when the ratio of causal relationship and observation data in causal-observation fusion deduction is 1:1, the model has the best prediction accuracy and cross-condition generalization ability. Equal to 0.2 and Equal to 0.5 is a comprehensive balance point. Under this setting, the model can maintain its accuracy under similar working conditions and significantly improve its generalization under different working conditions.

[0084] To analyze the contributions of each module in the causal enhancement-bidirectional gated recurrent unit (CA-BGRU) approach, Table 6 shows the accuracy and generalization performance of model variants obtained by combining different modules. As shown in Table 6, replacing the original features with the causal enhancement features reduces the average root mean square error (RMS) of the BGRU from 0.2521 to 0.2381, a 5.6% decrease. Specifically, the RMS error under the significant variance condition decreases from 0.3281 to 0.2681, an 18.3% improvement, demonstrating the importance of causality in battery state-of-health estimation. Introducing the domain calibration batch normalization mechanism reduces the RMS error to 0.2173, an 8.7% decrease, for an overall improvement of 13.8%, validating the importance of feature distribution calibration under different operating conditions. Incorporating invariant risk minimization regularization to obtain the complete causal enhancement-bidirectional gated recurrent unit model reduces the RMS error to 0.2154, a 0.9% improvement. Specifically, the RMS error under the moderate variance condition decreases from 0.3631 to 0.3585, a 1.3% improvement, demonstrating the role of invariant risk minimization in improving the model's out-of-distribution generalization capability. These results show that each module of the causal enhancement-bidirectional gated recurrent unit significantly contributes to the improvement of model performance, and these improvements are more obvious when facing out-of-distribution data, which proves the rationality and effectiveness of the constructed causal enhancement-bidirectional gated recurrent unit model.

[0085] Table 1 Lithium-ion battery operating parameters

[0086]

[0087] Table 2 Causal Enhancement-Bidirectional Gated Recurrent Unit Model Parameter Settings, Search Space, and Determination Method

[0088]

[0089] Table 3 Comparison of health status estimation accuracy of different methods

[0090]

[0091] Table 4 Comparison of the root mean square error of health state estimation by different methods under different working conditions

[0092]

[0093] Table 5 Comparison of RMS error of battery health status estimation with different parameters of causal enhancement-bidirectional gated cycle unit

[0094]

[0095] Table 6 Comparison of RMS error of battery health state estimation for different module combination model variants

[0096]

[0097] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A battery health status assessment method considering enhanced causal relationships, characterized in that: The following steps are involved: Obtain battery measurement data, obtain the multi-time series nonlinear causal relationship between battery health status and health factors based on causal inference methods, and construct a common battery multi-time series nonlinear causal relationship directed graph; Based on the directed graph of multi-time series nonlinear causal relationships of batteries and the battery health factor and health state deduction method based on causal cumulative response, a causal relationship enhancement model based on causal-observation fusion deduction is established to obtain the multi-time series nonlinear causal enhancement characteristics of batteries; The causal enhancement features are taken as input and the causal enhancement-bidirectional gated recurrent unit model is used to output the battery health status.

2. The battery health status assessment method considering enhanced causal relationships according to claim 1, characterized in that: The battery measurement data includes the mean, minimum, maximum, standard deviation and integral of the voltage, current and temperature measurement data in each charge and discharge cycle, as well as the time taken for each charge and discharge cycle.

3. The battery health status assessment method considering enhanced causal relationships according to claim 1, characterized in that: The causal inference method is used to obtain the multi-time series nonlinear causal relationship between the battery health state and the health factor, specifically including: using the potential Peter-Clark instantaneous conditional independence algorithm to discover the causal graph between the battery health state and the health factor; using the optimal adjustment set theory based on the causal graph to estimate the causal effect of the nonlinearity between the battery health state and the health factor, and combining the causal graph discovery results and the causal effect estimation results to obtain the multi-time series nonlinear causal relationship between the battery health state and the health factor.

4. The battery health status assessment method considering enhanced causal relationships according to claim 1, characterized in that: The common multi-time-series nonlinear causal relationship directed graph of batteries includes a node set and an edge set, wherein the node set is composed of a health factor set and a battery health status indicator with common causality, and the edge set describes the multi-time-series nonlinear common causal relationship between the battery health status and the health factors.

5. The battery health status assessment method considering enhanced causal relationships according to claim 1, characterized in that: Based on the directed graph of multi-time series nonlinear causal relationships of batteries and the battery health factor and health state deduction method based on causal cumulative response, a causal relationship enhancement model based on causal-observation fusion deduction is established to obtain the multi-time series nonlinear causal enhancement characteristics of batteries. The specific steps include: Establishing a battery state of health and health factor deduction model based on causal cumulative response according to a directed graph of multi-time series nonlinear causality of the battery, and using the deduction model to obtain the state of health and health factor deduction values ​​within each charge and discharge cycle based on historical state of health and health factor observations; Aiming to minimize the difference between the inferred value of health factors and the observed value of health factors, the optimal health status proxy value is introduced to replace the unknown health status observation value, and a causal relationship enhancement model based on causal-observation fusion deduction is constructed; By solving the causal enhancement model, the deduced values ​​of the battery health state and health factor are obtained, which together constitute the multi-time series nonlinear causal enhancement characteristics of the battery.

6. The battery health status assessment method considering enhanced causal relationships according to claim 5, characterized in that: The inference model starts with the observed values ​​of the health state and health factor of the first charge-discharge cycle, and calculates the average causal response of each health state and health factor in each charge-discharge cycle in chronological order under the influence of all other health states or health factors at all historical moments.

7. The battery health status assessment method considering enhanced causal relationships according to claim 1, characterized in that: The causal enhancement-bidirectional gated recurrent unit model training process adopts a loss function based on the invariant risk minimization theory.

8. The battery health status assessment method considering enhanced causal relationships according to claim 1, characterized in that: Before the causal enhancement features are input, they need to undergo domain calibration batch normalization processing, and each different operating condition has independent batch normalization parameters.

9. The battery health status assessment method considering enhanced causal relationships according to claim 8, characterized in that: For battery samples with known operating conditions, the batch normalization parameters corresponding to the operating conditions are directly applied for processing; For battery samples with unknown operating conditions, the similarity distance between them and all known operating conditions is calculated, and the batch normalization parameters corresponding to the operating condition with the smallest distance are selected for processing.

10. A computer device, characterized in that: The computer device comprises: one or more processors; a memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the battery health status assessment method considering enhanced causal relationships as described in any one of claims 1 to 9.

Images