Methods, equipment, media, and products for assessing the health status of lithium-ion batteries based on deconstructed physical information.

Abstract

This application discloses a method, device, medium, and product for assessing the state of health (SOH) of lithium-ion batteries based on deconstructed physical information. It relates to the field of lithium-ion battery SOH assessment technology, including the extraction of multi-dimensional health factors, inputting these factors into a deep physical information neural network that integrates a self-attention mechanism module and a Koopman neural operator module, and outputting an estimated SOH value. By extracting universal health factors applicable to multiple operating conditions, the problem of insufficient universality of health factors is solved; the self-attention mechanism is used to enhance features and reduce redundancy, improving the model's ability to perceive key information; the Koopman neural operator is used to deconstruct physical information, integrating the physical mechanisms of battery degradation into the model, enhancing the model's physical interpretability; the deep physical information neural network integrates multi-dimensional information for estimation, effectively addressing individual differences in different batteries and complex operating conditions, thereby achieving high-precision and robust SOH estimation of lithium-ion batteries.

Description

Technical Field

[0001] This application relates to the field of lithium-ion battery health status assessment technology, and in particular to a lithium-ion battery health status assessment method, device, medium and product based on deconstructed physical information. Background Technology

[0002] Lithium-ion batteries, with their high energy density, low self-discharge rate, long cycle life, and lack of memory effect, have been increasingly widely used in new energy vehicles and grid-side energy storage. However, during long-term operation, lithium batteries often face complex and variable operating conditions and external environmental conditions, leading to irreversible chemical and physical degradation within the battery, manifested as a continuous increase in battery impedance and sustained capacity decay. In extreme cases, this can even trigger internal short circuits, resulting in thermal runaway accidents, causing significant economic losses and serious safety incidents. Therefore, achieving high-precision and robust State of Health (SOH) estimation is crucial for building a reliable Battery Management System (BMS), effectively ensuring the safe and stable operation of the entire system.

[0003] However, current methods for estimating the state of health (SOH) of lithium-ion batteries cannot guarantee high accuracy and robustness in SOH assessment. Summary of the Invention

[0004] The purpose of this application is to provide a method, device, medium, and product for assessing the state of health (SOH) of lithium-ion batteries based on deconstructed physical information, which can achieve high-precision and robust SOH estimation of lithium-ion batteries.

[0005] To achieve the above objectives, this application provides the following solution:

[0006] Firstly, this application provides a method for assessing the health status of lithium-ion batteries based on deconstructed physical information, including:

[0007] Real-time acquisition of operational data during the charging phase of the target lithium-ion battery for each cycle;

[0008] Feature extraction is performed on the running data for each set of loops to obtain a multidimensional set of general health factors;

[0009] Using the multidimensional general health factor set as input, the trained lithium-ion battery health status assessment model outputs the SOH estimate for each cycle. The lithium-ion battery health status assessment model is a deep physical information neural network integrating a self-attention mechanism module and a Koopman neural operator module. The self-attention mechanism module enhances the features of the multidimensional general health factor set and outputs an enhanced feature set. The Koopman neural operator module deconstructs the enhanced feature set using physical information and outputs the physical decay rate. The deep physical information neural network outputs the SOH estimate based on the enhanced feature set and the physical decay rate.

[0010] In a second aspect, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the lithium-ion battery health status assessment method based on deconstructed physical information as described in the first aspect.

[0011] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the lithium-ion battery health status assessment method based on deconstructed physical information as described in the first aspect.

[0012] Fourthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the lithium-ion battery health status assessment method based on deconstructed physical information as described in the first aspect.

[0013] According to the specific embodiments provided in this application, this application has the following technical effects:

[0014] This application provides a method, device, medium, and product for assessing the health status of lithium-ion batteries based on deconstructed physical information. The method collects operational data of the target lithium-ion battery during each charging stage in real time, extracts a set of multidimensional general health factors from the data, and then inputs it into a deep physical information neural network (a trained lithium-ion battery health status assessment model) that integrates a self-attention mechanism module and a Koopman neural operator module. The self-attention mechanism module enhances the features of the multidimensional general health factor set to output an enhanced feature set, and the Koopman neural operator module deconstructs the enhanced feature set to output the physical decay rate. Finally, the deep physical information neural network combines the enhanced feature set and the physical decay rate to output the estimated SOH value for each cycle. This scheme addresses the issue of insufficient universality of health factors by extracting general health factors applicable to multiple operating conditions; it enhances the model's ability to perceive key information by using a self-attention mechanism to enhance features and reduce redundancy; it deconstructs physical information using the Koopman neural operator module to incorporate the physical mechanisms of battery degradation into the model, thereby enhancing the model's physical interpretability; and it uses a deep physical information neural network to fuse multi-dimensional information for estimation, effectively addressing individual differences in different batteries and complex operating conditions, thus achieving high-precision and robust SOH estimation for lithium-ion batteries. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating a lithium-ion battery health status assessment method based on deconstructed physical information according to Embodiment 1 of this application.

[0017] Figure 2 This is a schematic diagram of the overall KNO-SA-DPINN estimation model in Embodiment 1 of this application;

[0018] Figure 3 The current-voltage curves of the CCCV stage under the aging cycle in Embodiment 1 of this application are shown.

[0019] Figure 4 This is the feature correlation matrix diagram in Embodiment 1 of this application, wherein, Figure 4 (a) in the diagram is the feature correlation matrix diagram corresponding to HI1-HI4. Figure 4 (b) in the diagram is the feature correlation matrix diagram corresponding to HI5-HI8. Figure 4(c) in the diagram represents the feature correlation matrix corresponding to HI9-HI12. Figure 4 (d) in the figure represents the feature correlation matrix diagram corresponding to HI13-HI15;

[0020] Figure 5 This is a schematic diagram of the self-attention mechanism structure in Embodiment 1 of this application;

[0021] Figure 6 This is a schematic diagram of the KNO structure in Embodiment 1 of this application;

[0022] Figure 7 This is a schematic diagram of the capacity decay trajectory in Embodiment 1 of this application, wherein, Figure 7 (a) in the figure is a schematic diagram of the capacity decay trajectory corresponding to the NEEPU dataset. Figure 7 (b) in the figure is a schematic diagram of the capacity decay trajectory corresponding to the XJTU dataset;

[0023] Figure 8 This is a schematic diagram of the SOH estimation results of the method proposed in Embodiment 1 of this application under simulated vehicle operating conditions;

[0024] Figure 9 This is a schematic diagram of the SOH estimation results of the method proposed under conventional discharge conditions in Embodiment 1 of this application;

[0025] Figure 10 This is a schematic diagram of the SOH estimation results of the method proposed under random discharge conditions in Embodiment 1 of this application. Detailed Implementation

[0026] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0027] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, this application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0028] Example 1

[0029] Research has revealed shortcomings in current data-driven SOH estimation methods: Mathematically driven models are mostly "black box" models, lacking in-depth research into aging mechanisms and thus lacking physical interpretability; extracted health indicators (HI) are often limited to specific operating conditions, failing to apply to diverse scenarios and lacking universality. Extracting multiple features often leads to feature redundancy, impacting the model's overall estimation performance; in practical applications, significant inconsistencies exist between different batteries, and operating conditions are complex and variable. This complexity makes effective model coupling difficult, greatly affecting estimation performance and compromising the accuracy and robustness of lithium-ion battery SOH assessment.

[0030] To address the issues of poor cross-cell estimation accuracy and weak robustness in current lithium-ion battery SOH estimation models when facing complex operating conditions due to the lack of general applicability of HI and poor physical interpretability of the models, such as... Figure 1 As shown, this embodiment provides a method for assessing the health status of lithium-ion batteries based on deconstructed physical information, including:

[0031] SA: Real-time acquisition of operational data during the charging phase of the target lithium-ion battery for each cycle;

[0032] SB: Extract features from the running data for each set of loops to obtain a multidimensional set of general health factors;

[0033] SC: Taking the multidimensional general health factor set as input, the trained lithium-ion battery health status assessment model outputs the SOH estimate for each cycle number. The lithium-ion battery health status assessment model is a deep physical information neural network integrating a self-attention mechanism module and a Koopman neural operator module. The self-attention mechanism module enhances the features of the multidimensional general health factor set and outputs an enhanced feature set. The Koopman neural operator module deconstructs the enhanced feature set using physical information and outputs the physical decay rate. The deep physical information neural network outputs the SOH estimate based on the enhanced feature set and the physical decay rate.

[0034] By extracting universal health factors applicable to multiple operating conditions, the problem of insufficient universality of health factors is solved; by using a self-attention mechanism to enhance features and reduce redundancy, the model's ability to perceive key information is enhanced; by using the Koopman neural operator to deconstruct physical information, the physical mechanism of battery degradation is incorporated into the model, enhancing the physical interpretability of the model; and by using a deep physical information neural network to fuse multi-dimensional information for estimation, the model effectively addresses individual differences in different batteries and complex operating conditions, thereby achieving high-precision and robust SOH estimation of lithium-ion batteries.

[0035] The following section elaborates on the lithium-ion battery health status assessment method based on deconstructed physical information in this embodiment from the perspective of training the lithium-ion battery health status assessment model.

[0036] This embodiment proposes a KNO-SA-DPINN model (i.e., a lithium-ion battery health state assessment model) for SOH estimation of lithium-ion batteries. First, 15 general features highly correlated with battery capacity are extracted based on battery aging data, and a self-attention mechanism is introduced for secondary screening. This process not only effectively overcomes the feature redundancy problem but also significantly improves the universality of the features, providing high-quality feature input for subsequent modeling. Second, a generalized dynamic degradation equation for the battery is constructed to characterize the complex physical degradation characteristics of the battery under the coupling of multiple factors, more accurately reflecting the aging law of the battery. The Koopman Neural Operator (KNO) module efficiently solves the generalized dynamic degradation equation, deeply mining the physical information of the battery and enhancing the estimation accuracy and interpretability of the model. Finally, combining the self-attention mechanism module and the KNO module, the KNO-SA-DPINN model is built. By deeply integrating domain knowledge with neural network training, the shortcomings of traditional Physics-Informed Neural Networks (PINNs) in terms of physical constraint capabilities are effectively overcome, achieving high-precision and robust SOH estimation. The flowchart for SOH estimation using the KNO-SA-DPINN model is shown below. Figure 2 As shown, the specific steps are as follows:

[0037] S1: Acquire battery's full lifecycle operating data and preprocess the aging data.

[0038] Online data collection of full-cycle battery operation data from vehicle-mounted or grid-side energy storage terminals was performed. Preliminary data cleaning was conducted, including Gaussian noise filtering and filling in missing and outlier values ​​based on the 3σ principle. The cleaned data was then grouped according to the number of cycles and labeled with SOH values ​​according to formula (1) to construct time-series data. This helps the model effectively capture the time-dependent characteristics of battery degradation. The number of cycles refers to the cumulative count of a lithium-ion battery completing one full charge-discharge cycle, which is a key time variable for quantifying battery aging. Finally, the data was divided into training, validation, and test sets for subsequent model performance verification.

[0039]

[0040] Among them, C i C0 represents the maximum available capacity value in the i-th cycle, and C0 represents the initial capacity value.

[0041] S2: Extract general features based on charging data and perform correlation analysis

[0042] Considering that the data obtained in practical applications are generally charge-discharge data under non-constant operating conditions, with random discharge phases and relatively fixed charging phases, we extract general features strongly correlated with capacity from specific intervals of the constant current (CC) and constant voltage (CV) charging phases to characterize the battery degradation process. The voltage-current curves of the CCCV phases under long-term aging cycles are shown below. Figure 3 As shown, voltage curves with a voltage window of 0.4V (i.e., [3.8V, 4.2V]) were selected for extraction of statistically based general features (voltage mean: V-mean, voltage standard deviation: V-std, voltage kurtosis: V-kurt, voltage skewness: V-skew), namely HI1-HI4. Furthermore, from... Figure 3 It can be seen that when the voltage curve for a specific voltage window is linearly fitted, the angle α (V-LFA, i.e., the voltage linear fitting angle) formed with the horizontal line decreases with the increase of the number of cycles. The incremental capacity (IC) change on continuous voltage steps also contains many characteristics of battery aging, and these characteristic changes further reflect the aging mechanism of the battery. Therefore, IC analysis is performed on the voltage of a specific voltage window, and the position of the IC peak (P-place), the height difference between the two ends of the peak (AD), the area corresponding to the peak (P-area), and V-LFA are taken as HI5-HI8.

[0043] HI extraction was performed on the current curve with a current window of 1A (i.e., [0A, 1A]). Similar to the CC stage, statistically based features (current mean: I-mean, current standard deviation: I-std, current kurtosis: I-kurt, and current skewness: I-skew) and the current linear fitting angle β (I-LFA) were extracted. Furthermore, to more comprehensively characterize the characteristics of the CV stage, the area enclosed by the current curve and the x-axis (I-area) and the sampling time corresponding to [0A, 1A] (T-CV), i.e., HI9-HI15, were extracted. HI was then subjected to Min-Max normalization to eliminate the potential impact of differences in feature dimensions on subsequent analysis. The correlation matrix between HI and capacity was calculated based on Pearson correlation coefficients (PCCs) to quantitatively characterize the linear correlation between variables. The feature correlation matrix diagram is shown below. Figure 4 As shown, the absolute values ​​of the correlation coefficients of HI1-HI15 averaged 0.93, all exhibiting a strong linear correlation with SOH.

[0044] S3: Feature enhancement of the extracted multidimensional universal health factor set based on a self-attention mechanism with residuals.

[0045] Among the 15 general features extracted by S2, combined with Figure 4 It can be seen that the PCCs between some HIs are as high as 1, indicating significant redundancy in the feature set, which easily leads to model overfitting and weakens generalization ability. Therefore, a self-attention mechanism that integrates residuals and layer normalization is constructed. By dynamically allocating weights, multi-dimensional general feature information is integrated to further enhance the expressive power of features, effectively avoid the feature redundancy problem, and more accurately map the inherent law of SOH decay. Introducing residual connections helps to alleviate the gradient vanishing problem in deep neural networks, and adding normalization operations can ensure the scale consistency of input data, thereby maintaining the stability of the model training process. The specific structure is as follows: Figure 5 As shown. The specific implementation steps are as follows:

[0046] S301: Assume the input time series is X = {x1, x2, ..., x...} T}, where x i Given a d-dimensional vector (d = 15), where T is the sequence length, perform a linear transformation on the input sequence to generate the query, key, and value respectively.

[0047] Q = XW Q K = XW K V = XW V (2)

[0048] Among them, W Q W K W V The dimensions are T×d respectively. q 、Τ×d k 、Τ×d v d q =d k =d v .

[0049] S302: Calculate the attention score, introduce a scaling factor for scaling, and perform normalization to obtain the attention weight matrix A:

[0050]

[0051] Where, d k is the dimension of the key vector.

[0052] S303: Weighted sum of A and V is performed to generate a weighted feature set Z. A residual is introduced and connected to Z, and normalization is applied to it to obtain the final HI set Z′. The feature set Z′ serves as the input data for subsequent neural operator solutions, providing a solid data foundation for efficient model training and accurate estimation.

[0053] S4: Solving the generalized dynamic equations of a battery based on the Koopman neural operator, achieving deep deconstruction of physical information.

[0054] To better characterize the complex degradation process of batteries, a generalized dynamic equation is constructed. The specific steps are as follows:

[0055] S401: Construct a multivariate capacity characterization function that accurately describes the complex degradation characteristics of lithium batteries, specifically defined as follows:

[0056] u=f(t,X)(4)

[0057] Where t represents time (in this embodiment, the number of cycles), and X represents a feature set consisting of various influencing factors such as state of charge, depth of discharge, temperature, charge / discharge rate, and health indicators. In this embodiment, X is the feature set Z′.

[0058] Without loss of generality, to better characterize the dynamic degradation mechanism of the battery, the decay rate of SOH is defined as follows:

[0059]

[0060] S402: Equation (5) is an explicit partial differential equation (PDE) parameterized by the set θ, where h(θ) represents a nonlinear function of t, X, and u. Changing h(θ) can represent various forms of degradation models. However, the specific parameters of explicit PDEs are difficult to obtain, and their forms lack fixedness, making them difficult to directly apply to practical scenarios of lithium battery degradation modeling. In view of the above limitations, this paper defines a more generalized dynamic degradation equation with parameter Θ to extract the degradation mechanism of specific HI and aging time data, defined as follows:

[0061] u t =H(t,Z′,u,u) t ,u Z′ ,u Z′Z′ ,…;Θ) (6)

[0062] Among them, u t The decay rate of SOH, Let θ be the first-order partial derivative of u with respect to the set Z′, H() be the explicit characterization function of the generalized dynamic degradation equation of the battery, u be the capacity, and Θ be the parameter set of the generalized equation.

[0063] KNO, as a novel neural operator solver, maps nonlinear dynamics to a linear space and efficiently learns solutions to nonlinear partial differential equations through frequency division. While ensuring prediction accuracy, KNO can accurately solve the generalized dynamic degradation equations of batteries, effectively addressing the challenge of predicting the long-term complex dynamic behavior of batteries. Therefore, KNO, as a solver of the generalized dynamic equations of batteries, efficiently mines the degradation physical information hidden in the data. To improve solution efficiency and reduce computational complexity, equation (6) can be simplified to:

[0064] u t =H(t,Z′,u,u) t ,u z′ ;Θ) (7)

[0065] S403: Assume the input of the PDE in equation (7) Where D is (t, Z′, u, u) t ,u Z′ The set of ). The Encoder, as the observation function g(·), will Mapping to space Transform into And perform a Fast Fourier Transform:

[0066]

[0067] The Fourier series is truncated at the maximum number f in the frequency model, dividing the system into low-frequency and high-frequency components. The low-frequency component learns the Koopman operator, while the high-frequency component extracts high-frequency fluctuation information through a one-dimensional convolutional layer.

[0068] S404: For the low-frequency range, the definition is... Hankel matrix where t∈ε + And train an o-dimensional linear layer to learn the Koopman operator. Used to predict the latest observables Future state: Mapped to via inverse Fourier transform

[0069] S405: For the high-frequency component, a one-dimensional convolutional layer is used to independently predict the high-frequency component forward.

[0070]

[0071] S406: Connect two future states and Weighted summation is used to train the Decoder to represent the observation function g. -1 (·) Analytic space The target state of the solution to the equation in

[0072]

[0073] Where λ∈[0,1] represents the relative weights controlling low-frequency and high-frequency information.

[0074] S5: Construct the KNO-SA-DPINN model and use the trained model for online SOH estimation of lithium batteries.

[0075] Based on the Deep Physics-Informed Neural Network (DPINN), combined with a self-attention mechanism with residuals and a KNO module, a comprehensive estimation model, KNO-SA-DPINN, is constructed. The overall framework of the model is as follows: Figure 6 As shown, the specific steps for model construction are as follows:

[0076] S501: To truly integrate physical information into the neural network training process, rather than merely using it as a constraint, the KAPINN used to fit the generalized dynamic degradation equation of the battery (Equation 7) is defined as follows:

[0077]

[0078] S502: Fitting u using a deep feedforward neural network Γ(·) t The final KAPINN equation is obtained:

[0079]

[0080] Where t represents time (in this embodiment, it refers to the number of cycles), X represents the multidimensional general feature set, and ψ is the parameter set of the neural network.

[0081] S503: KAPINN, combined with a feature-enhanced KNO network, integrates the mined physical information into the model training, effectively overcoming the limitations of traditional PINNs in lithium battery SOH estimation, such as poor interpretability and insufficient constraint capability. Specifically, the parameters ψ and Θ of KAPINN can be trained by minimizing the mean squared error loss, with the specific loss function as follows:

[0082] ζ PINN =ζ data +αζ momo +βζ PDE (14)

[0083] Where, ζ data For the loss of the data item, ζ momo For monotonicity loss, ζ PDEThe loss term is calculated for KNO. α and β are the loss weights. For the data term loss, the loss function is determined by calculating the mean square error between the estimated value and the target value:

[0084]

[0085] in, Let u be the capacity estimate of Γ(·). i is the actual capacity, and i is the number of loops.

[0086] For monotonic losses, the focus is on a comprehensive analysis of the overall downward trend in battery capacity and the capacity regeneration phenomenon. The ReLU function is used to determine the prediction difference. and the true difference (u i -u i+1 The product of ) is used to determine the monotonic loss function:

[0087]

[0088] The loss for KNO solution terms mainly includes prediction loss and data reconstruction loss:

[0089]

[0090] Where, λ p , λ r These represent the weights for the prediction and reconstruction processes in the control loss function.

[0091] S504: Following the principle of minimizing the loss function value, the Adam optimizer will be used to optimize the hyperparameters of the constructed estimation model. The optimal model will then be used for experimental validation, and its estimation performance will be verified using a test set. Four evaluation metrics will be selected to scientifically describe the experimental results: Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), and R². 2 The specific calculation formula for the indicator is as follows:

[0092]

[0093] Where N is the number of samples, u i , These are the actual capacity value and the estimated capacity value for the i-th iteration, respectively.

[0094] This embodiment proposes a deep physical information neural network framework integrating KNO and self-attention mechanisms, effectively addressing the limitations of existing methods in cross-battery SOH estimation tasks. First, by constructing a multi-dimensional universal feature set, the model can capture the nonlinear dynamic characteristics of battery degradation, significantly improving its universality under complex operating conditions. Second, the introduction of a self-attention mechanism enables dynamic adjustment of feature weights, effectively overcoming the inconsistency problem caused by individual battery differences. Then, KNO linearizes the nonlinear dynamic system, deeply mining hidden physical information and significantly enhancing the model's interpretability. Finally, simulation examples verify the accuracy and robustness of the proposed model.

[0095] To verify the proposed lithium-ion battery health state estimation method based on the KNO-SA-DPINN model, which addresses the issues of low cross-battery estimation accuracy and weak robustness caused by the weak universality of health factors and poor physical interpretability of current SOH estimation models under complex dynamic conditions, simulation examples were conducted using a self-built on-board simulated lithium battery aging dataset (NEEPU) and the lithium-ion battery aging dataset from Xi'an Jiaotong University (XJTU). The specific implementation steps are as follows:

[0096] (I) Analysis of the raw battery data in the battery aging dataset. The data includes voltage and current curves during the charging phase, as well as capacity degradation curves. This study uses the first batch of lithium-ion battery cycle aging datasets from NEEPU and the first, third, and fourth batches of lithium battery aging datasets from XJTU, namely N11-4, B11-4, B31-4, and B41-4, totaling 16 batteries, for simulation examples and verification. The battery capacity degradation curves are shown below. Figure 7 As shown in Table 1, the specific experimental details of the NEEPU dataset and the XJTU dataset are as follows.

[0097] Table 1. Specific experimental details for each batch of the dataset.

[0098]

[0099] (II): Based on the acquired raw voltage, current, and capacity data, multidimensional general health factors are extracted. The extracted HI1-HI15 values ​​are sequentially labeled with SOH according to the number of cycles to facilitate subsequent division of the training and test sets. The labeled group data are then subjected to Min-Max normalization, with the specific formula as follows:

[0100]

[0101] Where, x i For the original group data, y iFor the normalized group data, max(x) and min(x) are the corresponding maximum and minimum values ​​in the original group data, respectively.

[0102] (III) Divide the processed feature data into a training set, a validation set, and a test set in a 2:1:1 ratio, i.e., 2 batteries for training, 1 battery for validation, and 1 battery for test. The input battery feature data X = {x1, x2, ..., x...} T The specific steps for inputting this data into the KNO-SA-DPINN model are as follows:

[0103] (1) Perform a linear transformation on the input sequence to obtain the query, key, and value. Then calculate the attention score, introduce a scaling factor for scaling and normalization, and perform a weighted summation of the weight matrices A and V. Perform residual connection and layer normalization on the results to obtain the enhanced feature combination Z′. Z′ is used as one of the input features for subsequent KNO module solutions.

[0104] (2) While inputting the input sequence X into the SA module, it is also input into the Deep Multilayer Perceptron (DMLP) for estimation, where feedforward estimation is performed, and the estimation result is then processed. This serves as one of the input features for the KNO module to solve the generalized dynamic equations of a battery.

[0105] (3) Integrate the outputs of the SA module and the DMLP network for the KNO module to solve the battery degradation equation u. t =H(t,Z′,u,u) t ,u z′ KNO performs frequency division processing on the input data to mine the degradation physical information inside the battery from different dimensions, and performs dimensionality reduction processing on the data after fast Fourier transform in order to calculate the loss function.

[0106] (4) Train the entire deep physical information neural network according to the principle of minimizing the loss function value until the optimal model is obtained. Then, use the optimal model to test the test set, output the final test results and perform case analysis.

[0107] (iv): The pre-defined test set was used for performance verification of the KNO-SA-DPINN model. Cross-testing with four batteries under different operating conditions was conducted to verify the overall estimation performance of the proposed model and test whether it met the expected requirements. The specific steps are as follows:

[0108] (1) Define the performance evaluation metrics for the model. Select MAE, MAPE, RMSE, and R... 2 As evaluation criteria, the smaller the MAE, MAPE, and RMSE, the higher the estimation accuracy of the model; R2 The closer the value is to 1, the higher the model's fit and the better the estimation result.

[0109] (2) For the simulated vehicle operating condition (N-Batch 1), the SOH estimation results of the proposed method are as follows: Figure 8 As shown. From Figure 8 The results show that, among the four cells, the SOH values ​​estimated by the KNO-SA-DPINN model exhibit a significant linear correlation with the actual values, and the data points are concentrated. The average absolute error is less than 0.0058, confirming that the proposed method has high accuracy and certain universality in cross-cell SOH estimation. Specifically, the mean MAE, MAPE, and RMSE of the KNO-SA-DPINN model are 0.57%, 0.64%, and 0.71%, respectively, while R... 2 The average reached 97.28%.

[0110] (3) For the conventional discharge condition (X-Batch 1), the SOH estimation results of the proposed method are as follows: Figure 9 As shown. From Figure 9 As can be seen, in each battery, the estimated results of the proposed method gradually converge towards the diagonal, exhibiting a denser clustering pattern, with an average absolute error of less than 0.0057. This fully demonstrates that the KNO-SA-DPINN model can fully leverage its algorithmic advantages to uncover deeper data values ​​when handling different operating conditions, significantly showcasing the method's good estimation accuracy and robustness.

[0111] (4) For random discharge conditions (X-Batch 3 and X-Batch 4), the SOH estimation results of the proposed method are as follows: Figure 10 As shown. From Figure 10 As can be seen, KNO-SA-DPINN demonstrated good estimation performance for different random discharge strategies. The majority of data points are distributed near the diagonal red line, and the average absolute error is less than 0.0061, proving the accuracy and robustness of the model's SOH estimation across cells under different operating conditions. Specifically, in the estimation results for four cells, the proposed method achieved mean MAE, MAPE, and RMSE of 0.61%, 0.66%, and 0.82%, respectively, demonstrating excellent estimation accuracy.

[0112] This embodiment proposes a lithium-ion battery health state estimation method based on the KNO-SA-DPINN model. First, general features are constructed. General features applicable to various complex operating conditions are extracted from battery aging data, providing high-quality input features for subsequent modeling. Second, multi-dimensional features are enhanced. The extracted multi-dimensional features are enhanced to avoid redundancy caused by excessive features. A self-attention mechanism with residuals is used to reconstruct the feature set, further amplifying the implicit physical information and improving the universality of the HI (Health State Estimation). Finally, the physical information is deeply deconstructed. Neural operators are used to solve the nonlinear partial differential equations characterizing battery degradation, deeply mining the domain knowledge hidden in the data. Physical information is truly integrated into the model training process, accurately depicting the evolution of battery state, effectively overcoming the shortcomings of traditional PINN in terms of physical constraint capabilities, thus achieving high-precision and robust SOH estimation.

[0113] The key to this embodiment lies in introducing KNO to solve the generalized dynamic degradation equation of the battery, achieving a deep fusion of physical information and the estimation model. This fusion provides the model with effective physical mechanism guidance, overcoming the problem of physical information imbalance in the cross-battery estimation process, thus ensuring the model's generalization ability under different battery samples and operating conditions. Furthermore, the self-attention mechanism integrated with KNO-SA-DPINN further amplifies the universality of multi-dimensional general features, more flexibly addressing the SOH estimation task of unknown lithium batteries. The proposed method, by combining physical information and a data-driven comprehensive estimation framework, provides an advanced solution for lithium battery SOH estimation. The superior performance demonstrated by this technology in simulation examples fully reflects its advantages in handling complex battery degradation problems, providing new research ideas and application schemes for the field of lithium battery state estimation, and offering an efficient and intelligent decision-making support tool for battery management systems. The application of this technology can not only accurately assess the health status of batteries and reduce maintenance costs, but also prevent safety accidents caused by battery failures, possessing significant social and economic value.

[0114] Example 2

[0115] This embodiment provides a computer device, which can be a server or a terminal. The computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database is used for data related to the lithium-ion battery health status assessment method based on deconstructed physical information. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the lithium-ion battery health status assessment method based on deconstructed physical information as described in Embodiment 1.

[0116] Example 3

[0117] This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0118] Example 4

[0119] This embodiment provides a computer program product, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0120] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0121] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0122] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0123] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0124] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for lithium-ion battery state-of-health estimation based on deconstructed physical information, characterized by, The lithium-ion battery health state evaluation method based on deconstruction of physical information comprises: Real-time acquisition of operation data of the target lithium-ion battery in the charging phase under each cycle number; Feature extraction of the operation data under each cycle number to obtain a set of multi-dimensional general health factors; Using the set of multi-dimensional general health factors as input, the trained lithium-ion battery health state evaluation model outputs the SOH estimation value under each cycle number, wherein the lithium-ion battery health state evaluation model is a deep physical information neural network integrating a self-attention mechanism module and a Koopman neural operator module, the self-attention mechanism module is used to enhance the features of the set of multi-dimensional general health factors and output a set of enhanced features, the Koopman neural operator module is used to deconstruct the physical information of the set of enhanced features and output a physical decay rate, and the deep physical information neural network is used to output the SOH estimation value according to the set of enhanced features and the physical decay rate; The training process of the lithium-ion battery health state evaluation model comprises: Obtaining lithium-ion battery full-life cycle operation data; Pretreating the lithium-ion battery full-life cycle operation data to obtain time series data with SOH labels indexed by cycle numbers; Feature extraction of the charging data in the time series data to obtain a set of sample multi-dimensional general health factors, wherein the charging data comprises constant-current phase charging data and constant-voltage phase charging data; Feature enhancement of the set of sample multi-dimensional general health factors based on a self-attention mechanism with residual connection to obtain a set of sample enhanced features; Physical information deconstruction of the set of sample enhanced features based on a Koopman neural operator to output a sample physical decay rate; Using the set of sample enhanced features and the sample physical decay rate as input, training a deep physical information neural network by minimizing a loss function to output the SOH estimation value of the lithium-ion battery under each cycle number; Physical information deconstruction of the set of sample enhanced features based on a Koopman neural operator to output a sample physical decay rate, specifically comprising: Constructing a battery generalized dynamic degradation equation based on the set of sample enhanced features and cycle numbers; Solving the battery generalized dynamic degradation equation based on a Koopman neural operator to obtain a sample physical decay rate; Solving the battery generalized dynamic degradation equation based on a Koopman neural operator to obtain a sample physical decay rate, specifically comprising: Mapping the set of sample enhanced features and the cycle numbers to an observation space through an observation function and performing Fourier transform to obtain frequency domain data; Dividing the frequency domain data into a low-frequency part and a high-frequency part; Constructing a Hankel matrix according to the low-frequency part; Learning a Koopman operator from the Hankel matrix through a training linear layer and performing linear transformation of the latest segment of the matrix using the Koopman operator to predict a low-frequency prediction result under future cycle numbers; The high-frequency part is extracted by a one-dimensional convolution layer to obtain a high-frequency prediction result of a future cycle number; The low-frequency prediction result and the high-frequency prediction result are fused to obtain the sample physical attenuation rate.

2. The method for evaluating the state of health of a lithium-ion battery based on deconstructed physical information according to claim 1, characterized in that, The operation data includes current data and voltage data of the constant current stage and the constant voltage stage; Feature extraction is performed on the operation data of each cycle number to obtain a multi-dimensional general health factor set, specifically including: Feature extraction is performed on the voltage data of the specific voltage window of the constant current stage to obtain a voltage average, a voltage standard deviation, a voltage kurtosis, a voltage skewness, a voltage linear fitting angle, an incremental capacity peak position, an incremental capacity peak height difference on both sides, and an incremental capacity peak area; Feature extraction is performed on the current data of the specific current window of the constant voltage stage to obtain a current average, a current standard deviation, a current kurtosis, a current skewness, a current linear fitting angle, a current curve area, and a sampling time; The voltage average, the voltage standard deviation, the voltage kurtosis, the voltage skewness, the voltage linear fitting angle, the incremental capacity peak position, the incremental capacity peak height difference on both sides, the incremental capacity peak area, the current average, the current standard deviation, the current kurtosis, the current skewness, the current linear fitting angle, the current curve area, and the sampling time are normalized to obtain normalized factors; The normalized factors constitute the multi-dimensional general health factor set.

3. The method of claim 1, wherein, Feature enhancement is performed on the sample multi-dimensional general health factor set based on a self-attention mechanism with residual connection to obtain a sample enhanced feature set, specifically including: Linear transformation is performed on the sample multi-dimensional general health factor set to generate a query matrix, a key matrix, and a value matrix; An attention weight matrix is calculated according to the query matrix and the key matrix; Weighted summation is performed on the attention weight matrix and the value matrix to obtain a sample feature set; Residual connection is introduced to the sample feature set, and layer normalization processing is performed to obtain the sample enhanced feature set.

4. The method of claim 1, wherein, The expression of the battery generalized dynamic degradation equation is: ； wherein, is a physical rate of decay for the sample; is an explicit representation function of the battery generalized dynamic degradation equation; is a capacity; t is a cycle number; is a first order partial derivative of u with respect to a set of sample augmented features ; is a set of parameters of the generalized equation.

5. The method of claim 1, wherein, The loss function includes a data item loss, a monotonicity loss, and a KNO solving item loss.

6. A computer device comprising: A memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that the processor executes the computer program to implement the lithium ion battery health state evaluation method based on deconstructed physical information according to any one of claims 1-5.

7. A computer-readable storage medium having stored thereon a computer program, characterized in that The computer program is executed by the processor to implement the lithium ion battery health state evaluation method based on deconstructed physical information according to any one of claims 1-5.

8. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the lithium ion battery health state evaluation method based on deconstructed physical information according to any one of claims 1-5.

Images