Lithium ion battery health state estimation method

Abstract

The invention discloses a lithium ion battery health state estimation method, and relates to the technical field of battery management. Comprising the following steps: acquiring time series data in the cyclic charging and discharging process of a lithium ion battery, constructing a second-order equivalent circuit model to extract dynamic characteristics, fitting an open-circuit voltage curve through a multi-layer perceptron to obtain static characteristics, and combining the dynamic characteristics and the static characteristics into physical flow characteristics; decoding the physical flow feature through a physical decoder to obtain a first battery health state; decoding the data stream characteristics through a data driving decoder to obtain a second battery health state; and determining a combined weight of the physical decoder and the data driving decoder according to the hidden space characteristics, and performing weighted fusion on the first battery health state and the second battery health state according to the combined weight to obtain a final battery health state of the lithium ion battery. According to the method, the problem that weight distribution of the PINN model between physical constraint and data-driven learning is difficult to coordinate is effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of battery management, and particularly relates to a method for estimating the state of health of a lithium-ion battery. Background Art

[0002] As one of the most widely used energy storage devices today, lithium-ion batteries are widely used in fields such as consumer electronics, electric vehicles, and renewable energy storage due to their high energy density, long cycle life, and no memory effect. With the rapid development of electric vehicles and energy storage systems, the demand for lithium-ion batteries has increased sharply, and the estimation of the state of health (SOH) of the battery has become a key technology to ensure the safe operation of the battery, optimize the battery management strategy, and extend the battery life. The battery SOH refers to the health degree of the battery relative to its initial state or rated state, usually expressed as the capacity retention rate or the percentage of health. Accurately estimating the SOH is crucial for the performance of the battery management system. The estimation of the SOH can help detect battery aging or faults in a timely manner, thereby avoiding potential safety hazards, optimizing the battery usage efficiency, and providing decision-making support for battery replacement or maintenance.

[0003] In recent years, with the rapid development of artificial intelligence and machine learning technologies, data-driven methods have gradually become an important research direction for SOH estimation. Deep learning, with its powerful non-linear modeling ability, can extract complex features from battery operation data and provide a new solution for SOH estimation. For example, deep learning models such as convolutional neural networks (CNNs), recurrent neural networks (RNNs), and long short-term memory networks (LSTMs) have been successfully applied to the SOH estimation of batteries. These methods can capture the long-term trends and non-linear dynamics of battery degradation by processing multi-dimensional data (such as voltage, current, temperature, SOC, etc.), thereby improving the estimation accuracy. However, deep learning methods also face some challenges, such as the dependence on high-quality labeled data, insufficient generalization ability of the model, and interpretability issues.

[0004] As an emerging deep learning framework, Physics-informed Neural Networks (PINN) combines the advantages of data-driven methods and physical models, providing new ideas for the State of Health (SOH) estimation of lithium-ion batteries. By embedding the physical laws of the battery, including electrochemical equations, thermodynamic equations, etc., into the loss function of the neural network, PINN can maintain the physical consistency of the model within a data-driven framework. This combination not only improves the accuracy and reliability of SOH estimation but also reduces the dependence on labeled data and enhances the generalization ability of the model.

[0005] However, in practical applications of PINN, the simple weighting of physical constraints and data-driven approaches leads to optimization difficulties. In complex systems, a single decoder cannot flexibly handle different regions, lacks the ability to adapt to dynamic changes in physical parameters, has insufficient interpretability, and it is difficult to analyze the basis for model decisions. Summary of the Invention

[0006] Based on this, it is necessary to provide a method for estimating the State of Health of lithium-ion batteries to address the above technical problems.

[0007] An embodiment of the present invention provides a method for estimating the State of Health of lithium-ion batteries, including: Obtain the time-series data during the cyclic charge and discharge process of the lithium-ion battery; Construct a second-order equivalent circuit model of the lithium-ion battery based on the time-series data, and extract the dynamic characteristics reflecting the instantaneous physical state of the lithium-ion battery through the second-order equivalent circuit model; fit the open-circuit voltage curve of the lithium-ion battery at different states of charge during the charge and discharge process through a multi-layer perceptron model to obtain the static characteristics reflecting the aging state of the lithium-ion battery; combine the dynamic characteristics and the static characteristics to obtain the physical flow characteristics of the lithium-ion battery; capture the time dependence of the time-series data through a recurrent neural network to obtain the data flow characteristics of the lithium-ion battery; Project the physical flow characteristics and the data flow characteristics into the same latent space through implicit projection to obtain the latent space characteristics of the lithium-ion battery; Input the physical flow characteristics, the data flow characteristics, and the latent space characteristics into a dynamic gated decoder, where the dynamic gated decoder includes a physical decoder and a data-driven decoder arranged in parallel; decode the physical flow characteristics through the physical decoder to obtain the first battery health state; decode the data flow characteristics through the data-driven decoder to obtain the second battery health state; determine the combined weights of the physical decoder and the data-driven decoder according to the latent space characteristics, and perform weighted fusion of the first battery health state and the second battery health state according to the combined weights to obtain the final battery health state of the lithium-ion battery.

[0008] Optionally, the second-order equivalent circuit model of the lithium-ion battery includes: a voltage source, the positive electrode of the voltage source is sequentially connected to a first-order RC network formed by series connection of an ohmic internal resistance, a first resistor, and a first capacitor, and a second resistor R and a second-order RC network formed by series connection of a second capacitor; Extract the dynamic characteristics reflecting the instantaneous physical state of the lithium-ion battery through the second-order equivalent circuit model based on the following formula: ; ; ; ; wherein, I is the loop current, R R1 is the first resistor, R R2 is the second resistor, U U1 is the voltage of the first resistor, U U2 is the voltage of the second resistor, η η is the discharge efficiency of the storage battery, Q N Cn is the rated capacity of the storage battery, U t U is the terminal voltage, R R0 is the ohmic internal resistance.

[0009] Optionally, fit the open-circuit voltage curve of the lithium-ion battery under different state of charge during the charge and discharge process through a multi-layer perceptron model to obtain the static characteristics reflecting the aging state of the lithium-ion battery, specifically including: The multi-layer perceptron model is composed of an input layer, multiple hidden layers and an output layer; Define that the target hidden layer contains multiple neurons, and determine the output value of the target neuron in the target hidden layer based on the following formula: ; wherein, ω ij W is the weight vector, Z (i-1) is the neuron vector of the previous hidden layer, b ij b is the bias, ACT(·) is the activation function, i l is the target hidden layer, j i is the target neuron in the target hidden layer; Determine the open-circuit voltage based on the following formula: ; wherein, y Uoc is the open-circuit voltage, ACT is the activation function, ω m Wl is the weight vector of the target hidden layer; Z(m-1) is the output value of the neurons in the previous hidden layer, b m is the bias; The open-circuit voltage is taken as the static feature.

[0010] Optionally, the time dependence of the time-series data is captured by a recurrent neural network to obtain the data flow features of the lithium-ion battery, specifically including: The time-series data is a sequence arranged in chronological order X =( x 1, x 2, …, x k ), k = 1, 2, …, T ; The recurrent neural network updates the hidden state at the current time according to the input and the hidden state at the previous time; the output and state of the recurrent neural network at each time are calculated by the following formula: ; ; where, x k is the time-series data, h k is the hidden layer state at the current time, h k-1 is the hidden layer state at the previous time, g (·) is the activation function, y k is the output of the recurrent neural network, W h and W x are the input weight matrices, b is the bias term, y k is the second battery health state; W y is the output weight matrix; The output of the recurrent neural network is taken as the data flow features of the lithium-ion battery.

[0011] Optionally, based on the following formula, the physical flow features and the data flow features are projected into the same latent space through implicit projection to obtain the latent space features of the lithium-ion battery: ; where, f ( t , x ) are the data flow features, u ( t , x) is a physical flow feature; During the implicit projection process, the loss function is calculated by the following formula: ; Among them, , and u i are u ( t , x ) initial and boundary training data, and are f ( t , x ) collocation points, N u is the total number of training data, N f is the total number of collocation points, MSE u is the error between the data flow feature and the latent space, MSE f is the implicit constraint of the physical equation, λ is the weight of the physical constraint.

[0012] Optionally, the combined weights of the physical decoder and the data-driven decoder are determined based on the following formula according to the latent space features: ; Based on the following formula, the first battery health state and the second battery health state are weighted and fused according to the combined weights to obtain the final battery health state of the lithium-ion battery:

[0013] Among them, α is the combined weight, g (·) is the activation function, W is the weight matrix of the gating network, b is the bias of the gating network, z is the latent space feature, SOH h is the first battery health state, SOH x is the second battery health state.

[0014] The above-mentioned method for estimating the health state of a lithium-ion battery provided by the embodiments of the present invention has the following beneficial effects compared with the prior art: The present invention realizes the deep coordination of physical constraints and data-driven through a dynamic gating decoder, adaptively adjusts the weights of the dual decoders with the latent space features as the medium, replaces the traditional simple weighting method, and effectively solves the problem that it is difficult to coordinate the weight distribution between the physical constraints and data-driven learning in the PINN model; the physical flow features explicitly encode the battery aging mechanism, and the data flow features implicitly mine the time series correlation law. The alignment mapping of the two in the latent space not only significantly reduces the optimization conflict and improves the adaptability to complex working conditions, but also enables the health state assessment to have both physical traceability and data-driven accuracy, strengthening the model's response ability to physical parameter changes. Description of the Drawings

[0015] Figure 1 It is a model framework diagram of a method for estimating the health state of a lithium-ion battery provided in an embodiment; Figure 2 It is a network structure diagram of a method for estimating the health state of a lithium-ion battery provided in an embodiment; Figure 3 It is a second-order RC equivalent circuit diagram of a method for estimating the health state of a lithium-ion battery provided in an embodiment. Detailed Embodiments

[0016] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0017] In one embodiment, a method for estimating the health state of a lithium-ion battery is provided, as Figure 1 shown, and the method includes: Obtain the time series data during the cyclic charge and discharge process of the lithium-ion battery.

[0018] Construct a second-order equivalent circuit model of the lithium-ion battery according to the time series data, and extract the dynamic features reflecting the instantaneous physical state of the lithium-ion battery through the second-order equivalent circuit model. Fit the open circuit voltage curve of the lithium-ion battery under different state of charge during the charge and discharge process through a multi-layer perceptron model to obtain the static features reflecting the aging state of the lithium-ion battery. Combine the dynamic features and the static features to obtain the physical flow features of the lithium-ion battery. Capture the time dependence of the time series data through a recurrent neural network to obtain the data flow features of the lithium-ion battery.

[0019] Project the physical flow features and the data flow features into the same latent space through implicit projection to obtain the latent space features of the lithium-ion battery.

[0020] Input the physical flow characteristics, data flow characteristics, and latent space characteristics into the dynamic gated decoder, which includes a physically-based decoder and a data-driven decoder arranged in parallel. Decode the physical flow characteristics through the physically-based decoder to obtain the first battery health state. Decode the data flow characteristics through the data-driven decoder to obtain the second battery health state. Determine the combined weights of the physically-based decoder and the data-driven decoder according to the latent space characteristics, and perform weighted fusion on the first battery health state and the second battery health state according to the combined weights to obtain the final battery health state of the lithium-ion battery.

[0021] As Figure 2 shown, the implementation process includes: (1)Equivalent circuit model.

[0022] The equivalent-circuit model (ECM) consists of three parts: a voltage source, an ohmic internal resistance, and an RC network. The ideal voltage source represents the open-circuit voltage of the battery U oc , and there is a non-linear relationship between it and the battery SOC. n The nth-order RC network characterizes the dynamic performance of the battery. As n n increases, the accuracy of the battery behavior characterization improves significantly, but the parameter identification difficulty increases exponentially. The second-order ECM achieves a good balance between complexity and accuracy. Therefore, the present invention selects the second-order ECM as shown in Figure 3 to construct the model. The second-order equivalent circuit model of the lithium-ion battery includes: a voltage source U oc , the positive electrode of the voltage source U oc is sequentially connected to the ohmic internal resistance R 0, the first resistor R 1 and the first capacitor C 1 in series to obtain a first-order RC network and the second resistor R 2 and the second capacitor C 2 in series to obtain a second-order RC network.

[0023] According to Kirchhoff's law and the current integration method, the mathematical expression for extracting the dynamic characteristics reflecting the instantaneous physical state of the lithium-ion battery through the second-order equivalent circuit model is expressed as: (1) (2) (3) (4) Among them, I is the loop current, R 1 is the first resistor,R 2 is the second resistor, U 1 is the voltage of the first resistor, U 2 is the voltage of the second resistor, η is the discharge efficiency of the storage battery, Q N is the rated capacity of the storage battery, U t is the terminal voltage, R 0 is the ohmic internal resistance. Under the condition that the battery parameters and the working current remain unchanged in each sampling period, the discretized battery model can be expressed as: (5) (6) Among them, k represents time discretization; Δ t is the sampling interval; τ 1 = R 1 C 1, τ2 = R 2 C 2.

[0024] (2) Open-circuit voltage modeling based on the multi-layer perceptron model.

[0025] Since the neural network has the ability to approximate any continuous function and can estimate the complex non-linear characteristics of the system, compared with the traditional predefined look-up table method and polynomial fitting method, this method shows stronger flexibility in capturing the dynamic characteristics of the battery. In view of U oc- the SOC curve will shift with the aging degree and environmental factors, the present invention comprehensively considers the problem complexity and calculation efficiency, and uses the multi-layer perceptron model (Multilayer Perceptron, MLP) to establish an open-circuit voltage model.

[0026] (7) Among them, U oc is the open-circuit voltage; SOC is the input value of the MLP; f MLP is the trained MLP network model.

[0027] The MLP consists of an input layer, m hidden layers and an output layer. Define that the i th hidden layer contains a total of d neurons, then the output value calculation formula of the j th neuron in this layer is: (8) Among them, ωij is the weight vector, Z (i-1) is the neuron vector of the previous hidden layer, b ij is the bias, and ACT(·) is the activation function, m is the total number of hidden layers, d is the total number of neurons in the target hidden layer, i is the target hidden layer, j is the target neuron in the target hidden layer.

[0028] The open-circuit voltage is: (9) where, y is the output value of the open-circuit voltage model, and ACT is the activation function, ω m is the weight vector of the target hidden layer; Z (m-1) is the neuron output value of the previous hidden layer, b m is the bias; the open-circuit voltage is used as a static feature.

[0029] As the network depth m increases, the non-linear representation ability of the model is enhanced, but the geometric growth of the parameter scale will lead to a significant increase in computational complexity. Therefore, it is necessary to determine the optimal network depth through strict complexity analysis and experimental verification.

[0030] Combine the dynamic features with the static features to obtain the physical flow features of the lithium-ion battery u ( t , x )

[0031] (3) Temporal feature extraction based on RNN.

[0032] As a classic deep learning model, the Recurrent Neural Network (RNN) can capture the time dependence of temporal data. Its structure is simple and the computational complexity is low. Therefore, the present invention adopts the RNN structure to implement the extraction of temporal features.

[0033] RNN accepts experimental data as input. Assume that the experimental data is a sequence X =( x 1, x 2,..., x k ), k = 1, 2,..., T. The RNN updates its hidden state at each time step based on the input and the previous hidden state. The layers of the RNN are connected in a cyclic manner, and the output and state at each time step are generated based on the input and state of the previous step, as shown below: (10) (11) Where, x k is the time-series data, h k is the hidden layer state at the current time step, h k-1 is the hidden layer state at the previous time step, g (·) is the activation function, y k is the output of the recurrent neural network, W h and W x are the input weight matrices, b is the bias term, y k is the second battery health state; W y is the output weight matrix; the output of the recurrent neural network is used as the data stream feature of the lithium-ion battery f ( t , x ).

[0034] (4) Fusion of physical flow features and data stream features.

[0035] The data stream feature f ( t , x ) and the physical flow feature u ( t , x ) are projected into a shared latent space through the implicit projection layer: (12) The data stream feature and the physical flow feature are combined in a latent space through implicit projection to ensure that they can be jointly optimized and satisfy physical constraints. To enhance the combination of the data stream feature and the physical flow feature, the loss function is defined as: (13) Where, , and u i are u ( t , x )'s initial and boundary training data, and is f ( t , x ) is the matching point, N u is the total number of training data, N f is the total number of configuration points, MSE u is the error between the data stream feature and the latent space; while MSE f is the implicit constraint of the physical equation; λ is the weight of the physical constraint, used to balance the relationship between the data stream feature and the physical constraint.

[0036] (5) Dynamic gating mechanism.

[0037] In order to determine the type of decoder to be adopted at each time point, the present invention designs a gating mechanism. This mechanism dynamically adjusts the combined weights of the data-driven decoder and the physical decoder through the latent space features. Specifically, the combined weights are calculated through a neural network, and the combined weights are used to adjust the weighting coefficients of the data-driven decoder and the physical decoder. The formula is as follows: (14) where, α is the combined weight; g (·) is the activation function, which limits the output within the range of [0, 1]; W is the weight matrix of the gating network, b is the bias of the gating network.

[0038] The neural network includes but is not limited to: RNN, Long Short-Term Memory (LSTM), or Gated Recurrent Unit (GRU).

[0039] (6) SOH estimation.

[0040] After the data-driven decoder and the physical decoder respectively output the SOH, a weighting mechanism is adopted to fuse the two to obtain the final SOH. The specific implementation method is as follows: (15) where, SOH h is the first state of health of the battery, SOH x is the second state of health of the battery.

[0041] Provide the specific implementation process of the present invention: Step 1: Data acquisition and preprocessing First, collect the experimental data of the lithium-ion battery during cyclic charge and discharge, including voltage, current, temperature, time, capacity, etc.

[0042] The experimental data is normalized using min - max normalization, which is achieved by mapping the data proportionally to a specified interval. For each feature x , the normalized value x norm is calculated as follows: (15) where, x min and x max are the minimum and maximum values of the data respectively. In this way, all features will be mapped to the interval range of [0, 1].

[0043] Step 2: Equivalent circuit model The ECM consists of three parts: a voltage source, an ohmic internal resistance, and an RC network. The ideal voltage source represents the open - circuit voltage of the battery U oc , and there is a non - linear relationship between it and the battery SOC. n The n - th order RC network characterizes the dynamic performance of the battery. As n increases, the accuracy of battery behavior characterization improves significantly, but the difficulty of parameter identification increases exponentially. The second - order ECM achieves a good balance between complexity and accuracy. Therefore, the present invention selects the second - order ECM shown in Figure 3 to build the model.

[0044] According to Kirchhoff's law and the current integration method, the mathematical expression of this model is: (16) (17) (18) (19) where, I is the loop current, and it is stipulated that I is positive during discharge; U 1, U 2 are the voltages of R 1, R 2 respectively; η is the discharge efficiency of the storage battery; Q N is the rated capacity of the storage battery; U t is the terminal voltage, R 0 is the ohmic resistance. Under the condition that the battery parameters and the working current remain unchanged at each sampling period, the discretized battery model can be expressed as: (20) (21) Among them, k represents time discretization; Δ t is the sampling interval; τ 1 = R 1 C 1, τ2 = R 2 C 2.

[0045] Step 3: Open-circuit voltage modeling based on a multi-layer perceptron neural network Since a neural network has the ability to approximate any continuous function and can estimate the complex nonlinear characteristics of a system, compared with traditional predefined look-up table methods and polynomial fitting methods, this method shows stronger flexibility in capturing the dynamic characteristics of a battery. In view of Uoc the -SOC curve will shift with the degree of aging and environmental factors, the present invention comprehensively considers the problem complexity and computational efficiency, and uses a multi-layer perceptron to establish an open-circuit voltage model.

[0046] This MLP network consists of an input layer, m hidden layers, and an output layer. Define the i th hidden layer contains a total of d neurons, then the output value calculation formula of the j th neuron in this layer is: (22) Among them, ω ij is the weight vector; Z (i-1) is the neuron vector of the i th - 1 layer; b ij is the bias; ACT is the activation function. The output value is: (23) As the network depth m increases, the nonlinear representation ability of the model is enhanced, but the geometric growth of the parameter scale will lead to a significant increase in computational complexity. Therefore, it is necessary to determine the optimal network depth through strict complexity analysis and experimental verification.

[0047] Step 4: Temporal feature extraction based on RNN As a classic deep learning model, RNN can capture the time dependence in sequential data. Its structure is simple and the computational complexity is low. Therefore, the present invention uses the RNN structure to achieve the extraction of temporal features.

[0048] RNN accepts sequential data as input. Assume the data is a sequence X =( x 1,x 2, …, x k ), k = 1, 2, …, T . The RNN updates its hidden state at each time step based on the input and the previous hidden state. The layers of the RNN are connected in a cyclic manner, and the output and state at each time step are generated based on the input and state of the previous step, as follows: (24) (25) Where, x k represents the input; h k represents the hidden layer state; g (·) is the activation function; W h and W x are the input weight matrices; b is the bias term; y k is the output; W y is the output weight matrix.

[0049] Step Five: Fusion of Physical Flow Features and Data Flow Features Project the data flow features f ( t , x ) and the physical flow features u ( t , x ) into a shared latent space through an implicit projection layer: (26) Combine the data flow features and the physical flow features in a latent space through an implicit projection to ensure that they can be jointly optimized and satisfy physical constraints. To enhance the combination of the data flow features and the physical flow features, define the loss function as: (27) Where, , and u i are the initial and boundary training data of u ( t , x ); and are the collocation points of f ( t , x ); N u andN f is the total number of training data and configuration points; MSE u is the error between the data stream feature and the latent space; and MSE f is the implicit constraint of the physical equation; λ is the weight of the physical constraint, used to balance the relationship between the data stream feature and the physical constraint.

[0050] Step Six: Dynamic Gating Mechanism To determine the type of decoder to be adopted at each time point, the present invention designs a gating mechanism. This mechanism dynamically adjusts the combined weights of the data-driven decoder and the physical decoder through the latent space features. Specifically, the combined weights are calculated through a neural network α , and the combined weights are used to adjust the weighting coefficients of the data-driven decoder and the physical decoder. The formula is as follows: (28) where α represents the combined weights; g (·) is the activation function, which limits the output within the range of [0, 1]; W is the weight matrix of the gating network; b is the bias of the gating network.

[0051] Step Seven: SOH Estimation After the data-driven decoder and the physical decoder respectively output the SOH, a weighting mechanism is adopted to fuse the two to obtain the final SOH. The specific implementation method is as follows: (29) Step Eight: SOH Evaluation To evaluate the performance of transfer learning in transferring the previously learned knowledge to the target task, four different regression evaluation metrics are used in this paper to evaluate the performance of the model in different scenarios. Including R 2 , Mean Absolute Error (MAE), Mean Bias Error (MBE), and Root Mean Square Error (RMSE).

[0052] (1) R -Square (30) R 2 Measures the goodness of fit of the model, indicating the proportion of the data variation explained by the model. Its value ranges between 0 and 1, and the closer it is to 1, the better the model fits. R 2The larger it is, the stronger the prediction ability of the model.

[0053] (2)Mean Absolute Error (31) MAE is the average of the differences between the predicted values and the true values, and is used to measure the mean absolute error between the predicted values and the true values. The smaller the MAE value, the more accurate the prediction of the model.

[0054] (3)Mean Bias Error (32) MBE is the average of the differences between the predicted values and the true values. If MBE is positive, it means the model tends to overestimate; if it is negative, it means the model tends to underestimate.

[0055] (4)Root Mean Square Error (33) RMSE is the average of the square roots of the errors, which can punish larger errors. The smaller the RMSE, the more accurate the prediction of the model.

[0056] The above-described embodiments merely represent several implementation manners of the present invention. Their descriptions are relatively specific and detailed, but should not be construed as limiting the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention.

Claims

1. A method for estimating the state of health of a lithium-ion battery, characterized in that, Including: Obtaining the time-series data during the cyclic charge and discharge process of a lithium-ion battery; Constructing a second-order equivalent circuit model of the lithium-ion battery according to the time-series data, and extracting the dynamic characteristics reflecting the instantaneous physical state of the lithium-ion battery through the second-order equivalent circuit model; fitting the open-circuit voltage curve of the lithium-ion battery under different state of charge during the charge and discharge process through a multi-layer perceptron model to obtain the static characteristics reflecting the aging state of the lithium-ion battery; combining the dynamic characteristics and the static characteristics to obtain the physical flow characteristics of the lithium-ion battery; Capturing the time-dependence of the time-series data through a recurrent neural network to obtain the data flow characteristics of the lithium-ion battery; Projecting the physical flow characteristics and the data flow characteristics into the same latent space through implicit projection to obtain the latent space characteristics of the lithium-ion battery; Inputting the physical flow characteristics, the data flow characteristics and the latent space characteristics into a dynamic gated decoder, where the dynamic gated decoder includes a physical decoder and a data-driven decoder arranged in parallel; decoding the physical flow characteristics through the physical decoder to obtain the first battery health state; Decoding the data flow characteristics through the data-driven decoder to obtain the second battery health state; Determining the combined weight of the physical decoder and the data-driven decoder according to the latent space characteristics, and performing weighted fusion on the first battery health state and the second battery health state according to the combined weight to obtain the final battery health state of the lithium-ion battery.

2. The method for estimating the state of health of a lithium-ion battery according to claim 1, wherein The second-order equivalent circuit model of the lithium-ion battery includes: a voltage source, the positive electrode of the voltage source is sequentially connected to a first-order RC network obtained by connecting an ohmic internal resistance, a first resistor, and a first capacitor in series, and a second resistor R 2 and a second-order RC network obtained by connecting a second capacitor in series; Extracting the dynamic characteristics reflecting the instantaneous physical state of the lithium-ion battery through the second-order equivalent circuit model based on the following formula: ； ； ； ； Among them, I is the loop current, R 1 is the first resistor, R 2 is the second resistor, U 1 is the voltage across the first resistor, U 2 is the voltage across the second resistor, η is the discharge efficiency of the storage battery, Q N is the rated capacity of the storage battery, U t is the terminal voltage, R 0 is the ohmic internal resistance.

3. The method for estimating the state of health of a lithium-ion battery according to claim 2, characterized in that, The step of fitting the open-circuit voltage curve of the lithium-ion battery under different state of charge during the charge and discharge process through the multi-layer perceptron model to obtain the static characteristics reflecting the aging state of the lithium-ion battery specifically includes: The multi-layer perceptron model is composed of an input layer, multiple hidden layers and an output layer; Defining that the target hidden layer contains multiple neurons, and determining the output value of the target neuron in the target hidden layer based on the following formula: ； Among them, ω ij is the weight vector, Z (i-1) is the neuron vector of the previous hidden layer, b ij is the bias, ACT(·) is the activation function, i is the target hidden layer, j is the target neuron in the target hidden layer; Determining the open-circuit voltage based on the following formula: ； Among them, y is the open-circuit voltage, and ACT is the activation function. ω m is the weight vector of the target hidden layer; Z (m-1) is the neuron output value of the previous hidden layer, b m is the bias; Taking the open-circuit voltage as the static characteristic.

4. The method for estimating the state of health of a lithium-ion battery according to claim 1, characterized in that, The step of capturing the time-dependence of the time-series data through the recurrent neural network to obtain the data flow characteristics of the lithium-ion battery specifically includes: Time series data is a sequence arranged in chronological order X =( x 1, x 2, …, x k ), k = 1, 2, …, T ; The recurrent neural network updates the hidden state at the current moment according to the input and the hidden state at the previous moment at each moment; calculating the output and state of the recurrent neural network at each moment through the following formula: ； ； Among them, x k is the time series data, h k is the hidden layer state at the current moment, h k-1 is the hidden layer state at the previous moment, g (·) is the activation function, y k is the output of the recurrent neural network, W h and W x are the input weight matrices, b is the bias term, y k is the second battery health state; W y is the output weight matrix; Taking the output of the recurrent neural network as the data flow characteristics of the lithium-ion battery.

5. The method for estimating the state of health of a lithium-ion battery according to claim 1, wherein, Projecting the physical flow characteristics and the data flow characteristics into the same latent space through implicit projection based on the following formula to obtain the latent space characteristics of the lithium-ion battery: ； Among them, f ( t , x ) is the data stream feature, u ( t , x ) is the physical stream feature; During the implicit projection process, calculating the loss function through the following formula: ； Among them, , and u i are u the initial and boundary training data of t , x ), and are f the collocation points of t , x ), N u is the total number of training data, N f is the total number of configuration points, MSE u is the error between the data stream feature and the latent space, MSE f is the implicit constraint of the physical equation, λ is the weight of the physical constraint.

6. The method for estimating the state of health of a lithium-ion battery according to claim 1, wherein Determining the combined weight of the physical decoder and the data-driven decoder according to the latent space characteristics based on the following formula: ； Performing weighted fusion on the first battery health state and the second battery health state according to the combined weight based on the following formula to obtain the final battery health state of the lithium-ion battery: Among them, α is the combined weight, g (·) is the activation function, W is the weight matrix of the gating network, b is the bias of the gating network, z is the latent space feature, SOH h is the first state of health of the battery, SOH x is the second state of health of the battery.

Images