Lithium battery soc estimation method based on multi-scale physical information neural network

Abstract

The present application relates to a kind of lithium battery SOC estimation method based on multiscale physical information neural network, belong to battery state of charge estimation technical field.The method includes: S1: the dynamic characteristics of battery is described using first-order RC equivalent circuit model, and the electrical parameters in it are used as the physical knowledge of Mamba model output;S2: voltage characteristics, current characteristics, time characteristics and edge features are fused in multiple scales, so as to obtain new features as the input of Mamba model;S3: construct Mamba model, including SSM, convolution layer, linear layer and element operator;SSM represents selective state space model, which is formed by combining structured state space sequence model with selection mechanism and scanning module;S4: lithium battery SOC estimation is carried out using Mamba model, and SPKF adaptive weight is used to improve model training efficiency.The present application improves the accuracy and reliability of SOC prediction.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of battery state of charge estimation, and relates to a lithium battery SOC estimation method based on a multi-scale physical information neural network. BACKGROUND

[0002] In the industry chain of new energy vehicles, lithium batteries are widely used in electric vehicles (EV) and hybrid electric vehicles (HEV). However, lithium-ion batteries will face many challenges in practical application, such as aging, increased internal resistance, and temperature changes. These problems not only affect the performance and safety of the battery, but also can cause premature failure of the battery, thereby affecting the reliability and economy of the overall vehicle.

[0003] In order to solve these problems, the battery management system (BMS) as a key technology is particularly important. An efficient BMS can monitor various parameters of the battery in real time, accurately estimate the state of charge (SOC) of the battery, and ensure that the battery operates within a safe range. In order to ensure that the BMS can accurately monitor the working state of the lithium battery, it is particularly important to design an efficient, fast-converging and adaptable lithium-ion battery model. At present, a variety of mainstream battery models have been established, including electrochemical models, data-driven models and equivalent circuit models (ECM). The electrochemical model uses a series of complex partial differential equations to describe the physical and electrochemical processes inside the battery, which can provide a relatively accurate description, but the calculation process is complex and time-consuming, and it is difficult to apply to real-time applications that require fast response. In contrast, data-driven models rely on large-scale data sets to construct the nonlinear mapping relationship between battery characteristics and SOC and remaining useful life (RUL) through machine learning techniques. Although this method can achieve good results in some cases, it still has limitations in terms of physical interpretability and sensitivity to input noise. In order to solve this problem, the equivalent circuit model has become an ideal choice for real-time monitoring due to its simple structure based on circuit theory and high computational efficiency.

[0004] With the in-depth study of lithium battery mechanism model, many researchers begin to combine these mechanism models with deep learning technology. For example, the physical information neural network unifies the explanation of the behavior of the battery by introducing physical information, and improves the interpretability and accuracy of the model. In addition, some new models, such as differential equation-based neural network, embed the physical law of the battery into the training process of the network, so that the model can better learn the dynamic characteristics of the battery. The above methods improve the performance and interpretability of the model to some extent, but there are still three problems: first, the traditional neural network model is always a huge challenge in the nonlinear mode, and it is easy to forget the information of the early time steps when processing the long time series data of the battery system. Second, the training generalization ability in a single scale is not strong, and it cannot adapt to different data distributions well, and it lacks flexibility to respond to different complex tasks. Finally, the fixed weight cannot guarantee that the training reaches the optimal value, and cannot effectively change dynamically to enhance the robustness. SUMMARY

[0005] Therefore, the purpose of the present application is to provide a lithium battery SOC estimation method based on a multi-scale physical information neural network, which can improve the accuracy and reliability of SOC prediction.

[0006] To achieve the above purpose, the present application provides the following technical solutions:

[0007] A lithium battery SOC estimation method based on a multi-scale physical information neural network, specifically comprising the following steps:

[0008] S1: A first-order RC equivalent circuit model is used to describe the dynamic characteristics of the battery, and the internal electrical parameters are used as the physical knowledge of the Mamba model output;

[0009] S2: Multi-scale fusion is performed on the voltage features, current features, time features and edge features to obtain new features as the input of the Mamba model;

[0010] S3: A Mamba model is constructed, including SSM, convolution layer, linear layer and element operator; wherein SSM represents a selective state space model, which is formed by combining a structured state space sequence model with a selection mechanism and a scanning module;

[0011] S4: The Mamba model is used for lithium battery SOC estimation, and SPKF (nonlinear Sigma point Kalman filter algorithm) adaptive weight is used to improve the model training efficiency.

[0012] Further, in step S2, the multi-scale fusion of the voltage features, current features, time features and edge features is represented by the formula:

[0013]

[0014]

[0015]

[0016] where T1 represents the voltage feature, T2 represents the current feature, T3 represents the time feature, T1', T2', T3' are new voltage, current, time features after layer fusion features; Cat is a connection operation, which combines the same dimension features; Conv is a convolution operation; BN is a batch normalization operation, which is used to improve the stability of training and accelerate convergence, and reduce the internal covariant bias; is a same dimension addition operation.

[0017] Further, in step S3, the structure of SSM is: structured state space sequence model input sequence is mapped to through the implicit latent state The superscripts b, L, D and N represent batch size, input length, feature number and hidden channel number, respectively, which describe The discrete state equation of

[0018]

[0019] y t = Ch t

[0020] The continuous parameter discretization formula is as follows:

[0021] Δ t = τ Δ (Parameter)

[0022]

[0023]

[0024] where x t , h t , y t represent the input, hidden state and output of t = 1, 2,..., L, and t represents the discrete time length; represents the state transition matrix, and represent the input-to-state and state-to-output mapping matrices, respectively; is the discretization parameter of (Δ, A, B) using zero-order hold (ZOH); τ Δis a softplus activation function; I denotes an identity matrix; Parameter denotes a learnable folding parameter; the structured state space sequence model follows a convolution computation, and a convolution process thereof can be defined as K = (k0,..., k t ) is a convolution kernel, where The equivalence of convolution and recursion enables the sequence to be trained in quadratic complexity in parallel, and facilitates the regression mapping by efficiently decoupling the computation steps.

[0025] Further, in step S3, the selection mechanism is as follows formula:

[0026]

[0027] y t = C t h t

[0028] Δ t = τ Δ (Parameter + s Δ (x t )) = softplus (Linear (x t ))

[0029]

[0030] Where τ Δ is a softplus activation function; s Δ is a linear function; given input x t = [T1′ T2′ T3′] T In this case, each layer can selectively capture important features according to the dependence relationship before and after the sequence.

[0031] Further, in step S3, the information transmission of each layer of the Mamba model is defined as:

[0032] x′ t = SilU (Conv1D (Linear (x t ))), z t = SilU (Linear (x t ))

[0033]

[0034] y′ t = LayerNorm (y t +x)

[0035]

[0036] wherein, is an element-wise multiplication.

[0037] Further, in step S4, the lithium battery SOC estimation is performed using the Mamba model, specifically including: assuming that the ohmic resistance R0, the polarization resistance R1 and the polarization capacitance C1 at a specified temperature remain unchanged, discretizing the first-order RC equivalent circuit equation of the lithium battery EMC model, the following equation can be obtained:

[0038]

[0039] U L [k]=h(SOC[k])-R1·I1[k]-R0·I[k]

[0040] wherein I1 is the current passing through R1; η is the charging conversion efficiency; Q is the rated capacity of the battery; Δt is the sampling interval; h(SOC[k]) is the open circuit voltage; U L [k] represents the output voltage at time k; I[k] represents the input current at time k; k represents time discretization; τ1=R1C1 represents the RC time constant; the Mamba model establishes a dynamic model by identifying a specific PDE (partial differential equation) error, and the PDE error ζ PDE of SOC can be represented by the following formula:

[0041]

[0042] In order to reflect the influence on SOC according to the circuit parameters, a weight value is used to change the voltage loss function so as to affect the updated network. The weights are obtained by introducing the resistance and capacitance into the SPKF to establish the state equation and the observation equation:

[0043] x k+1 =A k x k +B k (u k +w k )

[0044] wherein x k =[SOC k U k ] is a state vector, SOC k represents the SOC value at time k, U k represents the output voltage value at time k; u k =I k is an input, I k represents the current at time k; w k is state noise; is a state transition matrix, and the elements on the diagonal respectively represent the linear evolution of SOC and OCV (open circuit voltage) over time; is the input matrix, which is used to map the input u k and the noise w k to the state x k+1 at the next time step;

[0045] y k = C k x k + D k u k + v k

[0046] where y k = U k is the observation; v k is the observation noise; is the observation matrix, which reflects the degree of influence of the state variables SOC and OCV on the observation y k ; D k = -R0 is the input matrix;

[0047] The adaptive weight is calculated using a nonlinear mapping according to the observation value, and the Sigmoid function is used for nonlinear mapping:

[0048]

[0049] where k is a parameter that controls the steepness of the function, and y mid is an observation intermediate value; once the weight w(y) is calculated, the weighted MSE can be calculated, as follows:

[0050]

[0051] where L is the loss function of the voltage, y i is the observation prediction value, and y true is the true value of the voltage; the gradient optimization differential equation is L

[0052]

[0053] where SOC k and SOC k+1 are the estimated values of SOC at the previous and next time steps, respectively; the final total loss function L

[0054]

[0055] where N represents the number of samples; M represents the differential quantity; and a is the constraint factor of the voltage; the gradient of the loss function with respect to Θ is calculated using the Adam optimizer, which can be obtained by the following formula:

[0056]

[0057] Loss function g t The incoming Adam first moment estimate and second moment estimate are used for bias correction, and the reverse gradient updates the weights W between the output layer and the hidden layer t And the bias b t ; for the propagation gradient, the update formula of W t , b t is shown as follows:

[0058]

[0059]

[0060]

[0061]

[0062] Where, β1, β2 are the decay rates of the first moment estimate m t , v t and the second moment estimate respectively; m t , v t Bias correction; ξ is the network learning rate; ∈ is a small constant to prevent the algorithm from dividing by zero.

[0063] The beneficial effects of the present application are that: firstly, the present application combines physical laws with deep learning, fully utilizes the mechanism model of lithium batteries, and effectively improves the accuracy of state estimation. By introducing the electrochemical characteristics and physical behavior of the battery, the model can more accurately reflect the true state of the battery under different working conditions, thereby improving the reliability of the prediction. Secondly, the Mamba model uses multi-scale combined features to improve the estimation accuracy. In addition, Mamba significantly enhances the ability to explain the internal state of lithium batteries by embedding physical constraints. At the same time, the output end of the network adaptively adjusts the weight of the voltage loss according to the SPKF algorithm to achieve the best training effect. Mamba can effectively handle the time-varying characteristics of lithium battery systems and the mutual influence between multiple variables, thereby providing more accurate state estimation in practical applications. Finally, the present application has good practicability and operability, and can be easily integrated into existing battery management systems to provide strong support for real-time monitoring and prediction.

[0064] In summary, the method of the present application has significant advantages in lithium battery state estimation tasks and has good application prospects, providing important technical support for improving the safety and reliability of lithium batteries.

[0065] Additional advantages, objects, and features of the application will be apparent to those skilled in the art upon examination of the following specification. It is intended that the application not be limited by any of the details of the specification. Instead, such details are intended to be illustrative of the general nature of the application. Other embodiments will be apparent to those skilled in the art from consideration of the specification and practice of the application disclosed herein. BRIEF DESCRIPTION OF DRAWINGS

[0066] In order to make the objects, technical solutions and advantages of the present application clearer, the preferred embodiments of the present application will be described in detail below with reference to the drawings, in which:

[0067] Fig. 1 is a first-order RC equivalent circuit model schematic diagram;

[0068] Fig. 2 is a Mamba model structure framework schematic diagram;

[0069] Fig. 3 is a lithium battery SOC estimation method flowchart based on the Mamba model. DETAILED DESCRIPTION

[0070] The embodiments of the present application will be described in detail below with reference to specific examples. Those skilled in the art can easily understand other advantages and effects of the present application from the content disclosed in the specification. The present application can also be implemented or applied in different specific embodiments, and the details in the specification can be modified or changed in various ways based on different views and applications without departing from the spirit of the present application. It should be noted that the diagrams provided in the following examples only illustrate the basic concept of the present application in a schematic manner, and the following examples and features in the examples can be combined with each other without conflict.

[0071] Referring to Figs. 1-3 The present application provides a lithium battery SOC estimation method based on a multi-scale physical information neural network. First, a lithium ion battery modeling method is proposed. The first-order RC equivalent circuit model describes the dynamic characteristics of the battery through a simple circuit structure, and the internal electrical parameters are used as the neural network output to serve as physical knowledge. Second, the model features are combined using multi-scale, and the application of state space equation in deep learning is analyzed, especially the Mamba model derived from the state space. In addition, SPKF adaptive weight is proposed to improve the network training efficiency. Finally, according to the selective attention mechanism of the Mamba model and its characteristics, as well as the battery mechanism model, a method for estimating the internal state of the lithium battery is proposed. The data input current, voltage and time are input, and the dynamic state space and mechanism constraints are used to more accurately capture the changes of the lithium battery SOC.

[0072] 1) First-order RC electrochemical equivalent circuit model

[0073] The first-order RC equivalent circuit model is mainly composed of a voltage source, an ohmic internal resistance and a first-order RC network. The output of the voltage source can be approximated as an open-circuit voltage and there is a nonlinear relationship between the SOC and the voltage source. The ohmic internal resistance reflects the transient response capability of the battery and characterizes the voltage fluctuation of the battery under different loads. The first-order RC network is used to capture the dynamic characteristics of the battery and help analyze the performance of the battery under different working conditions. The partial differential equation of the simplified EMC of the lithium battery can be defined as follows:

[0074]

[0075] where η is the charge conversion efficiency; I(t) is the applied current of the battery, the discharge current is positive and the charging current is negative; Q is the rated capacity of the battery; R0 is the ohmic internal resistance; R1 is the polarization resistance; C1 is the polarization capacitance; I1(t), U1(t) are the applied current and voltage of R1; h(*) is the OCV-SOC curve function.

[0076] 2) Multi-scale feature fusion

[0077] The voltage, circuit and time information extracted from the database are often separated from each other. During the preliminary data processing, the underlying data features are not related to each other and contain more detailed information such as time sequence. In order to ensure the accuracy of the features of each layer and the target subject, multi-scale feature fusion can be used to combine the features of the first, second and third layers and the edge features to improve the clarity of the significant SOC target.

[0078] The input end contains three main features, namely voltage, current and time, which are represented as T i {i = 1, 2, 3}, where T1 represents the voltage feature, T2 represents the current feature, and T3 represents the time feature. The edge random feature E is fused with other features to obtain new features, and the formula flow is represented by the formula as follows:

[0079]

[0080]

[0081]

[0082] where T1', T2', T3' are new voltage, current and time features after layer-by-layer fusion of features; Cat is a connection operation that combines features of the same dimension; Conv is a convolution operation; BN improves the stability of training and accelerates convergence, and reduces the internal covariant bias. The BN formula can be represented as follows:

[0083]

[0084]

[0085] The input is normalized using the computed mean and variance:

[0086]

[0087] The normalized output is linearly transformed to get the output:

[0088]

[0089] where m is the batch size, x i is the i-th sample in the current batch. μ and σ are the mean and variance, respectively. ∈ is a small constant to prevent division by zero. γ and β are learnable parameters for scaling and shifting, respectively.

[0090] 3) Selective State Space Model (SSM)

[0091] Input sequence of structured state space sequence model (S4) through an implicit latent state is mapped to described The discrete state equation of

[0092]

[0093] y t = Ch t (10)

[0094] The continuous parameters are discretized as follows:

[0095] Δ t = τ Δ (Parameter) (11)

[0096]

[0097]

[0098] where x t , h t , y t denote the input, hidden state, and output at t = 1, 2,..., L; denotes the state transition matrix, and denote the input-to-state and state-to-output mapping matrices, respectively; is the discretization parameter of (Δ, A, B) using zero-order hold (ZOH); is the step size; Parameter denotes the learnable folding parameters. Structured state space sequence model follows the convolution computation, whose convolution process can be defined as K = (k0,..., k t ) is the convolution kernel, where The equivalence of convolution and recursion enables the sequence to be trained in quadratic complexity in parallel, facilitating the regression mapping by efficiently decoupling the computation steps.

[0099] However, the state matrix of structured state space sequence model does not change with the input, and cannot selectively focus on the importance of different features (tokens). Mamba model combines structured state space sequence model with selection mechanism and scanning module to form selective state space model (SSM), which realizes the selection of state space for time series input data. The selection mechanism is as follows:

[0100]

[0101] y t = C t h t (15)

[0102] Δ t = τ Δ (Parameter + s Δ (x t )) = softplus (Linear (x t )) (16)

[0103]

[0104] where τ Δ is the softplus activation function; s Δ is a linear function. The state equation of Mamba model has the same structure as formula (9), (10), only the matrix dimension changes, where Given the input x t = [T1′ T2′ T3′] T , each layer can selectively capture important features according to the dependence relationship before and after the sequence. The whole neural network (Mamba model) contains SSM block, convolution layer, linear layer and element operator, and the information transmission of each layer can be defined as:

[0105] x′ t = SilU (Conv1D (Linear (x t ))), z t = SilU (Linear (x t )) (18)

[0106]

[0107] y′ t = LayerNorm(y t +x) (20)

[0108] where, is element-wise multiplication. Meanwhile, since the state matrix becomes dynamic, the space cannot be calculated using the assumed fixed core, and only recursive representation can be used. Therefore, Mamba gives up the convolution operation and uses a new operation process of formula (21), which satisfies the commutative law & associative law.

[0109]

[0110] 4) Lithium battery SOC estimation based on Mamba model

[0111] Assuming that R0, R1 and C1 remain unchanged at a specified temperature, the first-order RC equivalent circuit equation of the lithium battery EMC model can be discretized to obtain the following equation:

[0112]

[0113] U L [k] = h(SOC[k]) - R1 · I1[k] - R0 · I[k] (23)

[0114] where I1 is the current passing through R1; Δt is the sampling interval; h(SOC[k]) is the open circuit voltage; k represents time discretization; τ1 = R1C1 represents the RC time constant. Mamba establishes a dynamic model by identifying a specific PDE error, and the PDE error of SOC can be represented by the following formula:

[0115]

[0116] In order to reflect the influence on SOC according to the circuit parameters, the weight value is used to change the voltage loss function to affect the updated network. The weights are obtained by introducing the resistance and capacitance into the SPKF to establish the state equation and observation equation:

[0117] x k+1 = A k x k +B k (u k +w k ) (25)

[0118] where x k = [SOC k U k ] is the state vector; u k = I kis the input; w k is the state noise; is the state transition matrix, the elements on the diagonal represent the linear evolution of SOC and V over time, respectively; is the input matrix, which is used to map the input u k and the noise w k to the state x k+1 at the next time step.

[0119] y k = C k x k + D k u k + v k (26)

[0120] where y k = U k is the observation; v k is the observation noise; is the observation matrix, which reflects the influence of the state variables SOC and V on the observation y k ; D k = -R0 is the input matrix.

[0121] The adaptive weight is calculated using a nonlinear mapping based on the observation, and the Sigmoid function is used for the nonlinear mapping:

[0122]

[0123] where k is a parameter that controls the steepness of the function, and y mid is an observation intermediate value. Once the weight w(y) is calculated, the weighted MSE can be calculated, as follows:

[0124]

[0125] where y i is the observation prediction value, and y true is the true value of the voltage. The gradient optimization differential equation, and the SOC loss function can be expressed as:

[0126]

[0127] where SOC k and SOC k+1 are the estimated values of SOC at the previous and current time steps, respectively. The final total loss function is obtained by weighting the values of the above two loss functions, i.e.,

[0128]

[0129] Wherein, N represents the number of samples; M represents the differential amount; a is the constraint factor of voltage. The gradient of the loss function with respect to Θ calculated using the Adam optimizer can be obtained by the following formula:

[0130]

[0131] Loss function g t The incoming Adam first moment estimate and second moment estimate are subjected to bias correction, and the reverse gradient updates the weights W between the output layer and the hidden layer t and the bias b t . For the propagation gradient, the update formula of W t , b t is shown as follows:

[0132]

[0133]

[0134]

[0135]

[0136] Wherein, β1, β2 are the decay rates of the first moment estimate m t , v t and the second moment estimate; m t , v t bias correction; ξ is the network learning rate; ∈ is a small constant to prevent the algorithm from dividing by zero.

[0137] Finally, it should be pointed out that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit, although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the purpose and scope of the technical solutions, which should be covered in the scope of the claims of the present application.

Claims

1. A lithium battery SOC estimation method based on a multi-scale physical information neural network, characterized by, The method specifically comprises the following steps: S1: The dynamic characteristics of the battery are described by using a first-order RC equivalent circuit model, and the internal electrical parameters are used as the physical knowledge of the Mamba model output; S2: Multi-scale fusion is performed on the voltage characteristics, current characteristics, time characteristics and edge characteristics to obtain new features as the input of the Mamba model; S3: The Mamba model is constructed, including an SSM, a convolution layer, a linear layer and an element operator; wherein the SSM represents a selective state space model, which is formed by combining a structured state space sequence model with a selection mechanism and a scanning module; S4: The Mamba model is used for lithium battery SOC estimation, and the SPKF adaptive weight is used to improve the model training efficiency; SOC estimation of lithium batteries using the Mamba model, including: assuming that the ohmic internal resistance , polarization resistance and polarization capacitance at a specified temperature remain constant, discretize the first-order RC equivalent circuit equation of the lithium battery EMC model to obtain the following equation: wherein, is the current passing through; is the charging conversion efficiency; is the rated capacity of the battery; is the sampling interval; is the open circuit voltage; denotes the output voltage at time k; denotes the input current at time k; denotes the time discretization; denotes the time constant; the Mamba model builds a dynamic model by identifying a specific PDE error for the SOC, the PDE error for the SOC is expressed by the following equation: The weights are obtained by introducing the resistance and the capacitance into the SPKF to establish the state equation and the observation equation: wherein, is a state vector, denotes the SOC value at time k, denotes the output voltage value at time k; is an input, denotes the current at time k; is a state noise; is a state transition matrix, the elements on the diagonal respectively denote the linear evolution of SOC and OCV over time; is an input matrix for mapping the input and the noise to the state at the next time instant; wherein is the observation value; is the observation noise; is an observation matrix which reflects the degree of influence of the state variables SOC and OCV on the observation ; and is an input matrix; The adaptive weight is calculated using a nonlinear mapping according to the observation value, and a Sigmoid function is used for nonlinear mapping: where is a parameter that controls the steepness of the function, is the observed median; once the weights are calculated, the weighted MSE can be calculated as follows: wherein, is a loss function for voltage, is an observed prediction value, is a true value for voltage; the gradient optimization differential equation is represented as: wherein, and are the estimated values of the SOC estimation upper and lower time instants respectively; the final total loss function is obtained by weighting the values of the two loss functions described above : wherein, represents the number of samples; represents the differential; is a constraint factor for the voltage; the gradient of the loss function with respect to is calculated using the Adam optimizer by the following formula: Loss function The incoming Adam first moment estimate and second moment estimate are passed through a bias correction, and the backward gradient updates the weights between the output layer and the hidden layer and bias ; for the propagated gradient, The update formula is represented as follows: wherein, are the first and second moment estimates respectively, of the decay rate; denotes a bias correction; is the network learning rate; is a small constant, t denotes the discrete time length.

2. The lithium battery SOC estimation method according to claim 1, characterized by, In step S2, the multi-scale fusion of the voltage characteristics, current characteristics, time characteristics and edge characteristics is expressed by the formula: wherein, represents a voltage feature, represents a current feature, represents a time feature, E represents an edge feature, is a new voltage, current, time feature after layer fusion feature; is a connection operation, combining the same dimension features; is a convolution operation; BN is a batch normalization operation; is a same dimension addition operation.

3. The lithium battery SOC estimation method according to claim 1, characterized by, In step S3, the structure of SSM is: structured state space sequence model input sequence By implicit latent state mapped to , superscript b , L , D and respectively represent batch size, input length, feature number and hidden channel number, describe The discrete state equation of is written as: The continuous parameter discretization formula is as follows: wherein, denotes the input, hidden state and output of t denotes a discrete time length; denotes a state transition matrix, and denote input-to-state and state-to-output mapping matrices, respectively; is a discretization parameter of is a step size; is a softplus activation function; I denotes an identity matrix; Parameter denotes a learnable folding parameter; the structured state space sequence model follows a convolution computation, whose convolution process is defined as , is a convolution kernel, where ; the equivalence of convolution and recursion enables the sequence to be trained in quadratic complexity in parallel, facilitating the regression mapping by efficiently decoupling the computation steps.

4. The lithium battery SOC estimation method according to claim 3, characterized by, In step S3, the selection mechanism is as follows: wherein, is a linear function; given input In this case, each layer selectively captures important features according to the dependency relationship before and after the sequence.

5. The lithium battery SOC estimation method according to claim 4, characterized by, In step S3, the information transmission of each layer of the Mamba model is defined as: wherein is an element-wise multiplication.

Images