Digital-analog fusion lithium battery state estimation method based on relaxation process

Abstract

The invention provides a relaxation process-based digital-analog fusion lithium battery state estimation method. The relaxation process-based digital-analog fusion lithium battery state estimation method comprises the following steps of: acquiring relaxation process voltage and temperature data in a battery end signal acquisition and data preprocessing module; estimating related physical parameters by using a second-order RC equivalent circuit model; inputting the processed physical characteristics, the prior relaxation voltage sequence and the temperature into a first-stage ATCN module to obtain an SOH estimated value; inputting the first-stage SOH estimated value, the physical characteristics, the relaxation voltage sequence and the temperature into a second-stage ATCN module, and outputting an SOC estimated value; and the final SOC and SOH are output by judging whether the variation amplitude of the two SOC and SOH estimation results converges or not. According to the relaxation process-based digital-analog fusion lithium battery state estimation method provided by the invention, the interpretability of a physical model and the nonlinear processing capability of a deep learning network model are combined, the SOC and SOH of the battery are accurately estimated, and the battery is accurately managed.

Description

Technical Field

[0001] This invention relates to lithium battery state estimation, and more specifically to a numerical-analog fusion lithium battery state estimation method based on relaxation processes. Background Technology

[0002] Lithium-ion batteries are widely used in portable electronic devices, electric vehicles, smart grid energy storage, and renewable energy integration due to their advantages such as high energy density, long cycle life, and low self-discharge rate. With the increasing prevalence of electric vehicles and energy storage systems, precise management of lithium-ion batteries has become crucial. State of Charge (SOC) and State of Health (SOH) are two core parameters in a Battery Management System (BMS) and are important bases for battery management. However, the complexity, nonlinearity, and time-varying characteristics of the internal electrochemical processes of batteries make accurate estimation of SOC and SOH a significant technical challenge.

[0003] Existing methods for estimating SOC and SOH mainly fall into three categories: measurement-based (open-circuit voltage method, coulomb counting, etc.), model-based (equivalent circuit model, electrochemical model, etc.), and data-driven (deep learning, statistical model, etc.). However, each of these methods has the following shortcomings: measurement methods are limited by conditions such as resting time and sensing errors; model-based methods exhibit significant parameter drift under aging, temperature, and rate coupling; and purely data-driven methods require a large amount of data and lack generalization and interpretability. Summary of the Invention

[0004] The purpose of this invention is to overcome the above-mentioned technical deficiencies and provide a numerical-analog fusion lithium battery state estimation method based on the relaxation process. It is a novel estimation method that integrates an improved equivalent circuit model and a temporal convolutional neural network (ATCN) with an integrated attention mechanism. By combining the advantages of both, it utilizes the battery relaxation process to achieve accurate estimation of the battery state, thereby improving the interpretability and robustness of the method.

[0005] To achieve the above-mentioned technical objectives, in a first aspect, the present invention provides a method for state estimation of lithium batteries based on a relaxation process using a fusion of numerical and analog data, comprising the following steps:

[0006] Real-time acquisition of terminal voltage and temperature data during the effective relaxation process of lithium batteries;

[0007] By introducing circuit elements for the charge transfer process and the diffusion process, an improved second-order RC equivalent circuit model of a lithium battery is constructed to model the relaxation process, and it is assumed that the charge transfer time constant and the diffusion time constant change linearly with the open-circuit relaxation time.

[0008] Based on the terminal voltage data and the temperature data, the key parameters of the second-order RC equivalent circuit model of the lithium battery are identified by the "alternating estimation" strategy and the least squares method, and the physical model feature parameters are extracted.

[0009] A two-stage iterative ATCN model is constructed. During actual estimation, the voltage sequence and temperature data of the cell terminal relaxation stage are collected in real time. The physical model characteristic parameters and prior data calculated by ampere-hour integration are then used. Voltage sequence and temperature data are used as inputs and fed into the first stage ATCN module of the ATCN model to output the battery SOH estimate.

[0010] The SOH value estimated in the first stage, along with physical characteristic parameters, relaxation stage voltage sequence, and temperature data, are input to the second-stage ATCN module, which outputs the posterior SOC estimate of the battery.

[0011] The estimation steps of the first-stage ATCN module and the second-stage ATCN module are executed alternately and iteratively until the change in the current iteration's SOC estimate or the change in the previous iteration's SOH estimate is less than a preset convergence threshold. The iteration stops when the current iteration's SOC estimate and SOH estimate are output as the result.

[0012] Compared with the prior art, the beneficial effects of the present invention include:

[0013] 1. Combination of physical and deep learning models: This approach integrates the physical interpretability of the equivalent circuit model with the powerful nonlinear fitting capability of neural networks, achieving complementary advantages in features and improving the estimation accuracy of SOC and SOH. Simultaneously, it retains the physical parameters of the equivalent circuit model, ensuring that the estimation result is not only a numerical value but also includes intermediate parameters reflecting the internal electrochemical state of the battery, facilitating fault diagnosis and health tracing.

[0014] 2. Improved Equivalent Circuit Modeling: A second-order RC equivalent circuit model is used to simulate the charge transfer and diffusion processes of the battery relaxation voltage. Based on the traditional second-order RC model, an assumption is introduced that the time constant changes linearly with the relaxation time. The least squares method is used to estimate the characteristic parameters of the physical model, improving the model's ability to capture the nonlinear characteristics of the battery. This method allows for the rapid acquisition of the required model parameters without requiring a long resting period.

[0015] 3. ATCN architecture design: The self-attention mechanism is integrated into the temporal convolutional network to construct ATCN. The causal dilated convolution and attention mechanism of TCN are used to capture the long and short sequence features of the relaxation process in parallel, which improves the accuracy of battery state estimation.

[0016] 4. Two-stage iterative estimation: A two-stage architecture is adopted, first estimating SOH and then SOC, and the estimated values ​​of SOC and SOH are dynamically corrected through iterative updates, which enhances the stability and accuracy of the estimation.

[0017] 5. High application value: The method focuses on the common static operating conditions of vehicles and equipment, under which battery changes are relatively stable, and only the data of the first 10 minutes of the relaxation process is needed. No complicated additional tests are required, making it easy to deploy and apply in practice.

[0018] According to some embodiments of the present invention, real-time acquisition of terminal voltage and temperature data during the effective relaxation process of a lithium battery includes the following steps:

[0019] Set the sampling period, current threshold, and minimum resting time to continuously collect battery terminal voltage data, current data, and surface temperature data;

[0020] When the current data is detected to be less than the current threshold and the duration exceeds the minimum rest time, it is determined that an effective relaxation process is entered, and starting from the moment when the current drops below the threshold, voltage data and temperature data with a length of N sampling cycles are extracted as a data window.

[0021] According to some embodiments of the present invention, charge transfer process circuit elements and diffusion process circuit elements are introduced to construct an improved second-order RC equivalent circuit model of a lithium battery to model the relaxation process, including the following steps:

[0022] An improved second-order RC equivalent circuit model for a lithium battery is constructed. The second-order RC equivalent circuit model for a lithium battery includes charge transfer process circuit elements and diffusion process circuit elements. The charge transfer process circuit elements are electrically connected to the battery terminals. The charge transfer process circuit elements include a resistor R1 and a capacitor C1 connected in parallel. The diffusion process circuit elements are electrically connected to the battery terminals. The diffusion process circuit elements include a resistor R2 and a capacitor C2 connected in parallel.

[0023] The charge transfer time constant and diffusion time constant are set to vary linearly with the open-circuit relaxation time. The charge transfer time constant is calculated based on the charge transfer linear coefficient and the charge transfer constant term, and the diffusion time constant is calculated based on the diffusion linear coefficient and the diffusion constant term.

[0024] According to some embodiments of the present invention, the physical model characteristic parameters include:

[0025] The parameters include open-circuit voltage, charge transfer process parameters, and diffusion process parameters. The charge transfer process parameters include: charge transfer linearity coefficient, charge transfer constant, and initial value of charge transfer polarization. The diffusion process parameters include: diffusion linearity coefficient, diffusion constant term, and initial value of diffusion polarization.

[0026] According to some embodiments of the present invention, the two-stage iterative ATCN model includes: a first-stage ATCN module, a second-stage ATCN module, and a fully connected layer connected in series. Each ATCN module contains three causal convolutional structures: the kernel size is 3, and the hole ratios are 1, 2, and 4 respectively.

[0027] The first-stage ATCN module and the second-stage ATCN module use residual connections to alleviate the gradient vanishing problem and accelerate convergence;

[0028] A residual block contains two convolutional layers, using the ReLU activation function, and WeightNorm and Dropout are added to each layer to regularize the network. A standard multi-head self-attention module is connected after each residual block to enhance the model's ability to capture temporal features and dynamically weight the importance of features at different time points.

[0029] The final outputs of the first-stage ATCN module and the second-stage ATCN module are mapped through a fully connected layer to obtain the battery SOH estimate and SOC estimate, respectively.

[0030] According to some embodiments of the present invention, the preset convergence threshold is 0.1%.

[0031] Secondly, the technical solution of the present invention provides a state estimation system for lithium batteries based on a relaxation process and a fusion of numerical and analog data, comprising:

[0032] The data acquisition module is used to collect terminal voltage and temperature data of the lithium battery in real time during the effective relaxation process.

[0033] The circuit model building module is used to introduce circuit elements for the charge transfer process and the diffusion process, and to build an improved second-order RC equivalent circuit model of the lithium battery to model the relaxation process. It is assumed that the charge transfer time constant and the diffusion time constant change linearly with the open-circuit relaxation time.

[0034] The feature parameter extraction module, based on the terminal voltage data and the temperature data, uses an "alternating estimation" strategy and the least squares method to identify key parameters of the second-order RC equivalent circuit model of the lithium battery and extracts the physical model feature parameters.

[0035] The battery SOH estimation module collects real-time voltage sequence and temperature data during the cell's terminal relaxation phase during actual estimation, and incorporates the physical model characteristic parameters and prior data calculated from the ampere-hour integral. Voltage sequence and temperature data are used as inputs and fed into the first stage ATCN module of the ATCN model to output the battery SOH estimate.

[0036] The SOC estimation module takes the SOH value estimated in the first stage, along with physical characteristic parameters, relaxation stage voltage sequence, and temperature data, as inputs and sends them to the second-stage ATCN module to output the posterior SOC estimate of the battery.

[0037] The results output module alternately and iteratively executes the estimation steps of the first-stage ATCN module and the second-stage ATCN module until the change in the current iteration's SOC estimate or the change in the previous iteration's SOH estimate is less than a preset convergence threshold. The iteration then stops and outputs the current iteration's SOC estimate and SOH estimate as the results.

[0038] Thirdly, the present invention provides a computer-readable storage medium storing computer-executable instructions for causing a computer to execute the state estimation method for a fusion of numerical and analog data based on a relaxation process as described in any one of the first aspects.

[0039] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0040] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, wherein the abstract drawings are to be completely consistent with one of the drawings in the specification:

[0041] Figure 1 A flowchart illustrating a numerical-analog fusion lithium battery state estimation method based on relaxation process provided in one embodiment of the present invention;

[0042] Figure 2 A schematic diagram of a second-order RC equivalent circuit model of the battery relaxation process provided in one embodiment of the present invention;

[0043] Figure 3 This is a flowchart of physical model parameter identification provided in one embodiment of the present invention;

[0044] Figure 4 This is a flowchart illustrating the execution of a method example provided in one embodiment of the present invention;

[0045] Figure 5 This is a schematic diagram of a temporal convolutional neural network structure with integrated attention mechanism provided in an embodiment of the present invention. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this invention clearer, the 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 merely illustrative and not intended to limit the invention.

[0047] It should be noted that although functional modules are divided in the system diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than the module division in the system or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and the aforementioned figures are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0048] Reference Figures 1 to 5 , Figure 1 A flowchart illustrating a numerical-analog fusion lithium battery state estimation method based on relaxation process provided in one embodiment of the present invention; Figure 2 A schematic diagram of a second-order RC equivalent circuit model of the battery relaxation process provided in one embodiment of the present invention; Figure 3 This is a flowchart of physical model parameter identification provided in one embodiment of the present invention; Figure 4 This is a flowchart illustrating the execution of a method example provided in one embodiment of the present invention; Figure 5 This is a schematic diagram of a temporal convolutional neural network structure with integrated attention mechanism provided in an embodiment of the present invention.

[0049] In one embodiment, a fusion-based numerical modeling method for lithium-ion battery state estimation based on the relaxation process includes the following steps: real-time acquisition of terminal voltage and temperature data during the effective relaxation process of the lithium-ion battery; introduction of charge transfer and diffusion process circuit elements to construct an improved second-order RC equivalent circuit model of the lithium-ion battery to model the relaxation process, assuming that the charge transfer time constant and diffusion time constant change linearly with the open-circuit relaxation time; based on the terminal voltage and temperature data, using an "alternating estimation" strategy and the least squares method to identify key parameters of the second-order RC equivalent circuit model of the lithium-ion battery, and extracting physical model feature parameters; constructing a two-stage iterative ATCN model, and during actual estimation, real-time acquisition of the cell terminal relaxation stage voltage sequence and temperature data, and combining the physical model feature parameters with prior data calculated by ampere-hour integration. Voltage sequence and temperature data are used as inputs and fed into the first-stage ATCN module of the ATCN model, which outputs the battery SOH estimate. The SOH value estimated in the first stage, along with physical characteristic parameters, relaxation stage voltage sequence, and temperature data, are used as inputs and fed into the second-stage ATCN module, which outputs the battery's posterior SOC estimate. The estimation steps of the first-stage ATCN module and the second-stage ATCN module are executed alternately and iteratively until the change in the SOC estimate between the current iteration and the previous iteration, or the change in the SOH estimate between the previous iteration and the previous iteration, is less than a preset convergence threshold. The iteration stops when the SOC estimate and SOH estimate of the current iteration are output as the result.

[0050] The numerical-analog fusion lithium battery state estimation method based on relaxation process has shown significant advantages in technical design, performance and engineering application, and effectively solves the pain points of traditional lithium battery state estimation methods.

[0051] From a technological innovation perspective, this invention first improves the second-order RC equivalent circuit model, overcoming the limitation of the fixed time constant in traditional models. By introducing circuit elements for the charge transfer process and the diffusion process, and assuming that the charge transfer time constant and the diffusion time constant change linearly with the open-circuit relaxation time, it can more accurately fit the dynamic evolution of the two key polarization processes of charge transfer and diffusion inside the battery during the relaxation stage, thus improving the fitting ability of the battery's nonlinear characteristics from a physical mechanism level. Secondly, in the parameter identification stage, an "alternating estimation" strategy combined with the least squares method is adopted to decompose the originally complex nonlinear parameter identification problem into two sub-problems: linear parameter solving and nonlinear parameter updating, which are processed alternately. This reduces computational complexity and can quickly and robustly extract physical model characteristic parameters such as open-circuit voltage, charge transfer, and diffusion process-related features. These retained physical parameters can also be used for fault diagnosis and health tracing, enhancing the interpretability of the technical solution. In addition, the constructed two-stage iterative ATCN model achieves a deep integration of physical features and data-driven approaches. The first stage estimates the SOH using physical model characteristic parameters, relaxation voltage sequence, temperature, and prior SOC as inputs. The second stage estimates the SOC by combining the SOH estimate with the physical characteristic parameters and other inputs. The values ​​of both are dynamically corrected through iterative iteration. This approach utilizes the ATCN model's efficient ability to capture long-term and short-term time series characteristics while avoiding the irrationality of purely data-driven model outputs through physical parameter constraints, thus breaking through the performance bottleneck of traditional single methods.

[0052] In terms of performance superiority, this invention demonstrates outstanding estimation accuracy and robustness. Regarding estimation accuracy, the features input to the ATCN model not only include traditional voltage and temperature data but also incorporate core parameters extracted from the physical model, resulting in a more comprehensive feature dimension and more accurate characterization of the battery's internal state. Simultaneously, the two-stage iterative mechanism, through mutual feedback correction between SOH and SOC, gradually offsets the error of a single estimation, significantly improving the stability and accuracy of the final output SOC and SOH estimates. In terms of robustness, since the physical model is built upon the core electrochemical mechanism of the relaxation process, and the ATCN model dynamically weights key temporal features through an attention mechanism, the method is more adaptable to complex operating conditions such as temperature fluctuations and battery aging (e.g., capacity degradation, increased internal resistance). Even in scenarios where battery aging causes parameter drift or temperature changes affect the polarization process, the physical model can still accurately extract features through dynamic time constants and alternating estimation strategies. The ATCN model can also avoid deviating from the actual state through physical parameter constraints, significantly improving robustness compared to traditional methods.

[0053] From an engineering practicality perspective, this invention is highly adaptable to real-world application scenarios and easy to deploy. Firstly, it eliminates the need for prolonged static storage. Traditional open-circuit voltage methods often require several hours of static storage to obtain a stable OCV, while this invention only needs to capture voltage and temperature data from the first 10 minutes of the relaxation process to complete parameter identification and state estimation. This perfectly matches the actual operating conditions of vehicles and energy storage devices, which typically require "short-term static storage," eliminating the need for additional complex testing procedures and lowering the barrier to engineering applications. Secondly, online deployment is convenient. Data acquisition only requires obtaining the terminal voltage, current, and temperature of the lithium battery, which are standard monitoring parameters for battery management systems, requiring no additional sensors. Parameter identification employs the efficient least squares method, and the ATCN model balances accuracy and computational load through optimized network structure, making it suitable for deployment in battery management systems or edge computing devices. The two-stage iteration uses the change in SOC and SOH estimates being less than a preset threshold as the convergence condition. Stable output is typically achieved with a small number of iterations, meeting real-time requirements. Overall computational complexity is controllable, and deployment costs are low.

[0054] The specific execution steps are as follows:

[0055] Step 1: Data Acquisition and Relaxation Segment Identification

[0056] In an embedded battery management system, the battery's terminal voltage U, current I, and surface temperature T are continuously acquired with a sampling period of 100ms. A current threshold I is set. thres (e.g., C / 50) and minimum settling time T min (For example, 10 seconds). When the absolute value of the detected current is less than I... thres And the duration exceeds T minWhen the current drops below the threshold, the system determines that it has entered an effective relaxation process and extracts a voltage and temperature sequence of length N (e.g., N = 6000, corresponding to 10 minutes of data) as a data window.

[0057] Step 2: Modeling the improved second-order RC equivalent circuit

[0058] This embodiment models the relaxation process voltage based on an improved second-order RC equivalent circuit. The modeling process is as follows.

[0059] 1. Improved second-order RC model and analytical expression of relaxation voltage

[0060] Considering the open-circuit relaxation phase (I=0) of a single battery cell, the ohmic voltage drop is 0. The terminal voltage can be written as:

[0061] U(t)=U OCV +v ct (t)+v d (t),

[0062] Where v ct For charge transfer polarization (branch R1C1), v d For diffusion polarization (branch R2C2). Assume the time constants of both branches vary linearly with the open-circuit relaxation time:

[0063] τ ct (t)=α1t+β1,τ d (t)=α2t+β2,

[0064] First-order differential equations of branch states:

[0065]

[0066] Solving for v, we get (Let v) ct (0)=A1,v d (0) = A2):

[0067]

[0068] Therefore, the voltage at the terminal during the relaxation phase is:

[0069]

[0070] 2. Construction of Discrete Recurrence Representation

[0071] For ease of online implementation, let the sampling period be Δt and the sampling time be t. k =kΔt. The discrete recursion is obtained by integrating the time-varying constant term using the midpoint approximation:

[0072]

[0073] And obtain the discrete terminal voltage relationship:

[0074] U[k]=U OCV +v ct [k]+v d [k].

[0075] 3. Linearized Regressor Quantities and Observation Construction

[0076] Least squares (RLS) requires linearity with respect to the parameters. Although U(t) is linear with respect to α i ,β i It is nonlinear, but the "instantaneous decay rate" for each branch can be linearized.

[0077] For any branch v i (i∈{ct,d}), discrete form:

[0078]

[0079] Take the logarithm and rearrange it to define the observable attenuation equivalent:

[0080]

[0081] A strictly linear regression relationship can be given:

[0082]

[0083] in The above formula is for parameter α i ,β i Step 3 of linear regression basic quantities: Extraction of physical model feature parameters

[0084] This implementation employs an "alternating estimation" strategy, decomposing the nonlinear parameter problem into linear subproblems and linearization subproblems that are solved alternately. Furthermore, to rapidly and robustly extract physical characteristic parameters from the battery's relaxation process, this embodiment uses Relative Simplification and Reduction (RLS) to evaluate the key parameter {U}. OCV The identification is performed on the following: α1, β1, α2, β2, A1, A2. The specific steps are as follows:

[0085] 1. Data Preprocessing

[0086] Acquisition of relaxation segment voltage sequence If a historical OCV-SOC table exists, the lookup value can be used as the initial value. Otherwise, the average of several samples from the final relaxation phase can be used as initial values. Calculate the total polarization.

[0087] 2. Initial value estimation for two windows

[0088] In actual observations, only the total polarization y[k] = U[k] - U can be directly measured.OCV rather than just v ct ,v d Therefore, the following engineering strategy is required to obtain the initial estimates for each branch. Using engineering experience, an early short window (fast model dominant) approximation y≈v is adopted. ct In the late stage, long window (slow mode dominates) y≈v d Construct a local r for the data within the two windows. i And obtained using RLS As initial values.

[0089] 3. Solving for linear parameters

[0090] In the given Under the condition that n represents the result of the nth iteration, construct the basis function:

[0091]

[0092] Design matrices can be constructed

[0093]

[0094] U=Xθ lin

[0095] Solving the linear least squares problem yields the linear parameter vector.

[0096]

[0097] To enhance online capabilities, recursive least squares can be preferred over batch solutions.

[0098] 4. Nonlinear parameter update

[0099] With a fixed linear parameter (U) OCV Under the conditions of A1, A2), this embodiment addresses the nonlinear parameter (α) i ,β i The direct least squares method is used for identification.

[0100] First, calculate the estimated voltage sequence for each branch based on the current parameters.

[0101]

[0102] And construct the observations

[0103]

[0104] when (e.g., ε = 10) -6 When the value is in a certain position, skip the update at that moment or use a longer difference interval to avoid numerical instability.

[0105] For each branch i, define a regression vector.

[0106]

[0107] And stack all samples to obtain the observation matrix.

[0108]

[0109] z i =Φ i θ i

[0110] Solve for the standard least squares estimate:

[0111]

[0112] To enhance online capabilities, recursive least squares can be preferred over batch solutions.

[0113] 5. Convergence Determination and Termination

[0114] Execute steps 3 and 4 alternately until either convergence condition is met:

[0115] (1) The norm increments of linear and nonlinear parameters are less than a preset threshold.

[0116] (2) Reconstruction error |U-Xθ lin |2 is less than the preset threshold

[0117] (3) Reach the maximum number of iterations.

[0118] Output the final parameter set {U OCV The parameters α1,β1,α2,β2,A1,A2} are used both for physical interpretation (OCV curves and relaxation features) and as physical feature inputs to the ATCN network, improving the accuracy and robustness of subsequent SOH and SOC estimations.

[0119] Step 4: Construct a two-stage iterative ATCN model to estimate battery SOC and SOH.

[0120] 1. Dataset Construction: Cyclic aging experiments were conducted on lithium batteries, and the collected battery charge-discharge and rest-relaxation experimental data were processed. First, the continuous voltage and current sequences were divided into fixed-length time segments using the sliding window method to construct the model input. The voltage sequence and temperature at the relaxation stage after each charge-discharge cycle were extracted, and then the physical model feature parameters were extracted in step three. The voltage, temperature, physical model feature parameters, and SOC (State of Charge) were then combined.

[0121] Using SOH as input and experimentally measured battery standard SOC and SOH as output labels, training and validation datasets are constructed.

[0122] 2. Constructing the ATCN model: This implementation uses a temporal convolutional network with an integrated attention mechanism.

[0123] (ATCN) is used to estimate SOC and SOH; the model structure is shown in the attached figure. Figure 5 As shown. The network input is a sequence length × 10-dimensional feature, including real-time voltage, temperature, and prior data calculated by ampere-hour integration. Feature parameters of the equivalent circuit model extracted in step two

[0124] {U OCV ,α1,β1,α2,β2,A1,A2}.

[0125] The ATCN model consists of two cascaded ATCN modules. The TCN module contains three causal convolutional layers with kernel sizes of 3 and dilation rates of 1, 2, and 4, respectively. Residual connections are used to mitigate the vanishing gradient problem and accelerate convergence. Each residual block contains two convolutional layers using the ReLU activation function, with WeightNorm and Dropout added to each layer for network regularization. A standard multi-head self-attention module (4 heads) is appended after each residual block to enhance the model's ability to capture temporal features and dynamically weight the importance of features at different time points, thereby improving its ability to represent battery state evolution. The final output of the ATCN network is mapped through fully connected layers to obtain estimates of the battery's State of Interest (SOH) and State of Charge (SOC).

[0126] 3. Network training strategy: Compare the preprocessed sample data with the corresponding actual SOH of the battery or

[0127] SOCs are used as labels in network training. The ATCN model employs a residual structure composed of multiple layers of causal dilated convolutions and embeds an attention mechanism within the network to improve long sequence feature extraction. Mean squared error (MSE) is used as the loss function during training.

[0128] The SOH and SOC branches are optimized separately. To improve generalization ability, strategies such as batch normalization, Dropout regularization, and early stopping can be used, and the learning rate and training epochs can be automatically adjusted based on the validation set error.

[0129] 4. Model Deployment and Convergence Judgment: After training, the model is deployed in the battery management system or edge computing device. During actual estimation, real-time data such as cell terminal voltage and current are collected, and the voltage relaxation curve is recorded after the vehicle's charging and discharging processes are completed. The equivalent circuit model feature parameters are extracted according to step three, and compared with the voltage sequence, temperature, and prior data calculated from the ampere-hour integral.

[0130] Construct the model input, use the first-stage ATCN to estimate the SOH sequentially, and then input the model characteristic parameters, voltage sequence, temperature, and the SOH estimated in the first stage into the second-stage ATCN.

[0131] The system estimates the State of Charge (SOC) and feeds back the two-stage results for iterative updates. The difference between the SOC and State of Hypoxia (SOH) from the two iterations is compared to a preset threshold (e.g., 0.1%). If the difference is less than the threshold, convergence is considered achieved, and the final SOC and SOH estimates are output; otherwise, iteration continues until the maximum number of iterations or the convergence condition is reached. This process requires no additional experimental conditions and can quickly estimate the battery's SOC and SOH.

[0132] The real-time acquisition of terminal voltage and temperature data during the effective relaxation process of the lithium battery includes the following steps: setting a sampling period, current threshold, and minimum resting time; continuously acquiring terminal voltage, current, and surface temperature data of the battery; when the current data is detected to be less than the current threshold and the duration exceeds the minimum resting time, it is determined that an effective relaxation process has been entered, and starting from the moment when the current drops below the threshold, voltage and temperature data of length N sampling periods are extracted as a data window.

[0133] Improved second-order RC equivalent circuit model of lithium battery, such as Figure 2 As shown.

[0134] An improved second-order RC equivalent circuit model for lithium batteries is constructed by introducing charge transfer process circuit elements and diffusion process circuit elements to model the relaxation process. The steps include: constructing an improved second-order RC equivalent circuit model for lithium batteries, which includes charge transfer process circuit elements and diffusion process circuit elements. The charge transfer process circuit elements are electrically connected to the battery terminals, and each includes a resistor R1 and a capacitor C1 connected in parallel. The diffusion process circuit elements are also electrically connected to the battery terminals, and each includes a resistor R2 and a capacitor C2 connected in parallel. Furthermore, the charge transfer time constant and diffusion time constant are set to change linearly with the open-circuit relaxation time. The charge transfer time constant is calculated based on the charge transfer linearity coefficient and the charge transfer constant term, and the diffusion time constant is calculated based on the diffusion linearity coefficient and the diffusion constant term.

[0135] The physical model features include: open-circuit voltage, parameters related to the charge transfer process, and parameters related to the diffusion process. The parameters related to the charge transfer process include: charge transfer linearity coefficient, charge transfer constant, and initial value of charge transfer polarization. The parameters related to the diffusion process include: diffusion linearity coefficient, diffusion constant term, and initial value of diffusion polarization.

[0136] In one embodiment, a fusion-based lithium battery state estimation system based on the relaxation process includes: a data acquisition module for real-time acquisition of terminal voltage and temperature data during the effective relaxation process of the lithium battery; a circuit model construction module for introducing charge transfer process circuit elements and diffusion process circuit elements to construct an improved second-order RC equivalent circuit model of the lithium battery to model the relaxation process, assuming that the charge transfer time constant and diffusion time constant change linearly with the open-circuit relaxation time; a feature parameter extraction module for identifying key parameters of the second-order RC equivalent circuit model of the lithium battery based on the terminal voltage data and the temperature data, using an "alternating estimation" strategy and the least squares method, and extracting physical model feature parameters; and a battery SOH estimation module for real-time acquisition of the cell terminal relaxation stage voltage sequence and temperature data during actual estimation, and combining the physical model feature parameters with prior data calculated by ampere-hour integration. Voltage sequence and temperature data are input to the first-stage ATCN module of the ATCN model, which outputs the battery SOH estimate. The SOC estimation module takes the SOH value estimated in the first stage, physical characteristic parameters, and voltage sequence and temperature data from the relaxation stage as input to the second-stage ATCN module, which outputs the battery's posterior SOC estimate. The result output module iteratively executes the estimation steps of the first-stage ATCN module and the second-stage ATCN module alternately until the change in the current iteration's SOC estimate or the change in the previous iteration's SOH estimate is less than a preset convergence threshold. When this happens, the iteration stops, and the current iteration's SOC estimate and SOH estimate are output as the result.

[0137] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0138] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0139] Furthermore, one embodiment of the present invention provides a computer-readable storage medium storing computer-executable instructions that are executed by a processor or controller, for example, by a processor in the above-described terminal embodiment, such that the processor performs the analog-digital fusion lithium battery state estimation method based on relaxation process described in the above-described embodiment.

[0140] Those skilled in the art will understand that all or some of the steps and systems in the methods disclosed above can be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, as is known to those skilled in the art, communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0141] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of the present invention.

[0142] The specific embodiments of the present invention described above do not constitute a limitation on the scope of protection of the present invention. Any other corresponding changes and modifications made in accordance with the technical concept of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A numerical-analog fusion method for lithium battery state estimation based on relaxation processes, characterized in that, Including the following steps: Real-time acquisition of terminal voltage and temperature data during the effective relaxation process of lithium batteries; By introducing circuit elements for the charge transfer process and the diffusion process, an improved second-order RC equivalent circuit model of a lithium battery is constructed to model the relaxation process, and it is assumed that the charge transfer time constant and the diffusion time constant change linearly with the open-circuit relaxation time. Based on the terminal voltage data and the temperature data, the key parameters of the second-order RC equivalent circuit model of the lithium battery are identified by the "alternating estimation" strategy and the least squares method, and the physical model feature parameters are extracted. A two-stage iterative ATCN model is constructed. During actual estimation, the voltage sequence and temperature data of the cell terminal relaxation stage are collected in real time. The physical model characteristic parameters and prior data calculated by ampere-hour integration are then used. Voltage sequence and temperature data are used as inputs and fed into the first stage ATCN module of the ATCN model to output the battery SOH estimate. The SOH value estimated in the first stage, along with physical characteristic parameters, relaxation stage voltage sequence, and temperature data, are input to the second-stage ATCN module, which outputs the posterior SOC estimate of the battery. The estimation steps of the first-stage ATCN module and the second-stage ATCN module are executed alternately and iteratively until the change in the current iteration's SOC estimate or the change in the previous iteration's SOH estimate is less than a preset convergence threshold. The iteration stops when the current iteration's SOC estimate and SOH estimate are output as the result.

2. The method for state estimation of lithium batteries based on relaxation process fusion of numerical and analog models according to claim 1, characterized in that, Real-time acquisition of terminal voltage and temperature data during the effective relaxation process of lithium batteries, including the following steps: Set the sampling period, current threshold, and minimum resting time to continuously collect battery terminal voltage data, current data, and surface temperature data; When the current data is detected to be less than the current threshold and the duration exceeds the minimum rest time, it is determined that an effective relaxation process is entered, and starting from the moment when the current drops below the threshold, voltage data and temperature data with a length of N sampling cycles are extracted as a data window.

3. The method for state estimation of lithium batteries based on relaxation processes according to claim 1, characterized in that, By introducing circuit elements for the charge transfer process and the diffusion process, an improved second-order RC equivalent circuit model of a lithium battery is constructed to model the relaxation process, including the following steps: An improved second-order RC equivalent circuit model for a lithium battery is constructed. The second-order RC equivalent circuit model for a lithium battery includes charge transfer process circuit elements and diffusion process circuit elements. The charge transfer process circuit elements are electrically connected to the battery terminals. The charge transfer process circuit elements include a resistor R1 and a capacitor C1 connected in parallel. The diffusion process circuit elements are electrically connected to the battery terminals. The diffusion process circuit elements include a resistor R2 and a capacitor C2 connected in parallel. The charge transfer time constant and diffusion time constant are set to vary linearly with the open-circuit relaxation time. The charge transfer time constant is calculated based on the charge transfer linear coefficient and the charge transfer constant term, and the diffusion time constant is calculated based on the diffusion linear coefficient and the diffusion constant term.

4. The method for state estimation of lithium batteries based on relaxation process fusion of numerical and analog models according to claim 3, characterized in that, The physical model's characteristic parameters include: The parameters include open-circuit voltage, charge transfer process parameters, and diffusion process parameters. The charge transfer process parameters include: charge transfer linearity coefficient, charge transfer constant, and initial value of charge transfer polarization. The diffusion process parameters include: diffusion linearity coefficient, diffusion constant term, and initial value of diffusion polarization.

5. The method for state estimation of lithium batteries based on relaxation process fusion of numerical and analog models according to claim 1, characterized in that, The two-stage iterative ATCN model includes: a first-stage ATCN module, a second-stage ATCN module, and a fully connected layer connected in series. Each ATCN module contains three causal convolutional structures: the kernel size is 3, and the hole ratios are 1, 2, and 4 respectively. The first-stage ATCN module and the second-stage ATCN module use residual connections to alleviate the gradient vanishing problem and accelerate convergence; A residual block contains two convolutional layers, using the ReLU activation function, and WeightNorm and Dropout are added to each layer to regularize the network. A standard multi-head self-attention module is connected after each residual block to enhance the model's ability to capture temporal features and dynamically weight the importance of features at different time points. The final outputs of the first-stage ATCN module and the second-stage ATCN module are mapped through a fully connected layer to obtain the battery SOH estimate and SOC estimate, respectively.

6. The method for state estimation of lithium batteries based on relaxation process fusion of numerical and analog models according to claim 1, characterized in that, The preset convergence threshold is 0.1%.

7. A state estimation system for lithium batteries based on a fusion of numerical and analog models using relaxation processes, characterized in that, include: The data acquisition module is used to collect terminal voltage and temperature data of the lithium battery in real time during the effective relaxation process. The circuit model building module is used to introduce circuit elements for the charge transfer process and the diffusion process, and to build an improved second-order RC equivalent circuit model of the lithium battery to model the relaxation process. It is assumed that the charge transfer time constant and the diffusion time constant change linearly with the open-circuit relaxation time. The feature parameter extraction module, based on the terminal voltage data and the temperature data, uses an "alternating estimation" strategy and the least squares method to identify key parameters of the second-order RC equivalent circuit model of the lithium battery and extracts the physical model feature parameters. The battery SOH estimation module collects real-time voltage sequence and temperature data during the cell's terminal relaxation phase during actual estimation, and incorporates the physical model characteristic parameters and prior data calculated from the ampere-hour integral. Voltage sequence and temperature data are used as inputs and fed into the first stage ATCN module of the ATCN model to output the battery SOH estimate. The SOC estimation module takes the SOH value estimated in the first stage, along with physical characteristic parameters, relaxation stage voltage sequence, and temperature data, as inputs and sends them to the second-stage ATCN module to output the posterior SOC estimate of the battery. The results output module alternately and iteratively executes the estimation steps of the first-stage ATCN module and the second-stage ATCN module until the change in the current iteration's SOC estimate or the change in the previous iteration's SOH estimate is less than a preset convergence threshold. The iteration then stops and outputs the current iteration's SOC estimate and SOH estimate as the results.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions for causing a computer to perform the state estimation method for a fusion of numerical and analog data based on a relaxation process as described in any one of claims 1 to 6.

Images