An online method for estimating the state of charge of lightweight batteries driven by an electrochemical model

Abstract

This application discloses an electrochemical model-driven online estimation method for the state of charge (SOC) of lightweight batteries, belonging to the field of online SOC estimation technology. This method establishes a high-fidelity electrochemical model and acquires baseline terminal voltage data and sparse terminal voltage frequency response data at each operating point on a three-dimensional grid composed of SOC, current rate, and temperature. Based on a physical information neural network, the sparse terminal voltage frequency response data is reconstructed and enhanced in the frequency domain to obtain enhanced terminal voltage frequency response data. The Loewner order reduction algorithm is used to perform order reduction modeling and stabilization processing on the enhanced terminal voltage frequency response data at each operating point, obtaining a local terminal voltage order reduction model at each operating point. Based on the above model, a three-dimensional scheduling linear parameter variation terminal voltage prediction model is constructed to obtain the predicted terminal voltage. The residual between the predicted and measured terminal voltages is used to correct the predicted SOC value to be corrected, obtaining the final estimated value. This application can reduce the plate-end computational load of pseudo-two-dimensional electrochemical models and improve the real-time performance, stability, and robustness of terminal voltage prediction and online SOC estimation under dynamic operating conditions.

Description

Technical Field

[0001] This application relates to the field of online state-of-charge estimation technology for batteries, and in particular to a lightweight online state-of-charge estimation method for batteries driven by an electrochemical model. Background Technology

[0002] The State of Charge (SOC) of a battery is a crucial state quantity reflecting the remaining usable capacity of the battery and a vital basis for battery management systems to achieve energy dispatch, safety protection, range assessment, and charge / discharge control. In energy storage systems, electric vehicles, and various power supply equipment, inaccurate SOC estimation can easily lead to problems such as inaccurate remaining capacity display, mismatched charge / discharge control strategies, inaccurate range prediction, and unreasonable safety protection threshold settings, thereby affecting the system's operational safety, reliability, and energy utilization efficiency.

[0003] Existing SOC estimation methods mainly include the ampere-hour integration method, the open-circuit voltage method, methods based on equivalent circuit models, and data-driven methods. Among them, the ampere-hour integration method is simple to implement, but it is sensitive to initial values ​​and current measurement accuracy, and is prone to cumulative errors over long periods of operation. The open-circuit voltage method requires a long settling time, making it difficult to apply directly to real-time estimation under dynamic operating conditions. Methods based on equivalent circuit models are easy to implement in engineering and for embedded deployment, but their ability to characterize electrochemical mechanisms such as diffusion, polarization, and mass transfer within the battery is limited, and the accuracy of SOC estimation is difficult to improve further under complex operating conditions and wide operating ranges. Data-driven methods typically rely on a large number of training samples and labeled data, have limited adaptability to different operating conditions, temperatures, and battery types, and lack physical interpretability.

[0004] Compared to the methods mentioned above, the electrochemical mechanism model-based approach can more realistically describe the lithium-ion transport, concentration distribution evolution, and polarization response processes within the battery, exhibiting stronger physical consistency and mechanistic explanation capabilities. Therefore, it has potential advantages in SOC estimation. However, full-order electrochemical models typically have high state dimensions, a large number of parameters, and high computational complexity. If directly used for online estimation, they often fail to meet the real-time operational requirements of resource-constrained platforms such as embedded controllers, development boards, or battery management systems.

[0005] Furthermore, the accuracy of battery terminal voltage and SOC estimation is significantly affected by changes in current rate, ambient temperature, and dynamic operating conditions. Temperature variations alter the diffusion coefficient, reaction kinetics, conductivity, and polarization characteristics, causing the terminal voltage response to vary with SOC, rate, and temperature. Therefore, lightweight models built only around a single SOC, rate, or fixed temperature range struggle to maintain stable terminal voltage prediction and SOC correction capabilities across a wide range of temperatures, rates, and dynamic currents. While some existing reduced-order models can reduce computational complexity, improvements are still needed in dynamic response consistency, scheduling continuity, and the ability to utilize terminal voltage residuals under multiple SOC, rate, and temperature conditions. Summary of the Invention

[0006] The purpose of this application is to provide a lightweight online estimation method for battery state of charge driven by an electrochemical model. This method can transform a high-fidelity electrochemical model into a lightweight online estimation model suitable for plate-side operation while maintaining the consistency of the electrochemical mechanism and the accuracy of the terminal voltage response. This enables stable, accurate and real-time online estimation of the battery state of charge.

[0007] To achieve the above objectives, this application provides the following solution: An electrochemical model-driven online method for estimating the state of charge (SOC) of lightweight batteries includes: A high-fidelity electrochemical model was established based on battery material parameters and geometric parameters.

[0008] On a three-dimensional operating point grid consisting of state of charge, current ratio, and temperature, baseline terminal voltage data and sparse terminal voltage frequency response data at each operating point are obtained based on the high-fidelity electrochemical model.

[0009] The frequency domain reconstruction and enhancement of the sparse end voltage frequency response data are performed based on the physical information neural network to obtain the enhanced end voltage frequency response data. Based on the Loewner data-driven order reduction algorithm, the frequency response data of the enhanced terminal voltage at each operating point are modeled and stabilized to obtain the local terminal voltage order reduction model at each operating point.

[0010] A three-dimensional scheduling linear parameter change terminal voltage prediction model is constructed based on the local reduced-order model at each operating point; the three-dimensional scheduling linear parameter change terminal voltage prediction model retains the independent state of each local terminal voltage reduction model, and establishes a fusion mechanism for the output of the neighboring local terminal voltage reduction model according to the state of charge, current ratio and temperature.

[0011] Based on the current state of charge, dispatch current ratio, and dispatch temperature, the predicted terminal voltage is obtained using a three-dimensional dispatch linear parameter variation terminal voltage prediction model.

[0012] Obtain the predicted state of charge to be corrected, obtain the terminal voltage residual based on the predicted terminal voltage and the measured terminal voltage, and correct the predicted state of charge to be corrected according to the terminal voltage residual to obtain the final estimated state of charge.

[0013] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides an online estimation method for the state of charge (SOC) of lightweight batteries driven by an electrochemical model. First, a high-fidelity electrochemical model is established based on battery material and geometric parameters, avoiding direct solution of high-dimensional electrochemical equations during online estimation while preserving electrochemical mechanism information and temperature effects in the dynamic response of the terminal voltage. Second, on a three-dimensional operating point grid composed of SOC, current rate, and temperature, baseline terminal voltage data and sparse terminal voltage frequency response data at each operating point are obtained based on the high-fidelity electrochemical model. The sparse terminal voltage frequency response data is then reconstructed and enhanced in the frequency domain using a physical information neural network to obtain enhanced terminal voltage frequency response data. This reduces the computational load of the high-fidelity electrochemical model in the frequency domain simulation while ensuring the required frequency response data density and bandwidth coverage for order reduction modeling. Next, a Loewner data-driven order reduction algorithm is used to perform order reduction modeling and stabilization processing on the enhanced terminal voltage frequency response data at each operating point, obtaining the local terminal voltage frequency response data at each operating point. The voltage reduction model can reduce the amount of online recursive calculation. Then, a three-dimensional scheduling linear parameter changing terminal voltage prediction model is constructed based on the local reduction model at each operating point. The three-dimensional scheduling linear parameter changing terminal voltage prediction model retains the independent state of each local terminal voltage reduction model and establishes a fusion mechanism for the output of adjacent local terminal voltage reduction models according to the state of charge, current ratio, and temperature, so that the local terminal voltage model can continuously transition between different operating points. Finally, based on the current state of charge, scheduling current ratio, and scheduling temperature, the predicted terminal voltage is obtained based on the three-dimensional scheduling linear parameter changing terminal voltage prediction model. Based on the predicted value of the state of charge to be corrected, the terminal voltage residual is obtained based on the predicted terminal voltage and the measured terminal voltage. The predicted value of the state of charge to be corrected is corrected based on the terminal voltage residual to obtain the final estimated value of the state of charge. This can improve the real-time performance, stability, and robustness of terminal voltage prediction and online estimation of battery state of charge under dynamic operating conditions while maintaining the physical consistency of the main state of battery state of charge. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1This is a diagram illustrating the application environment of an electrochemical model-driven online estimation method for the state of charge of lightweight batteries in one embodiment of this application. Figure 2 A flowchart illustrating an online estimation method for the state of charge of a lightweight battery driven by an electrochemical model, provided as an embodiment of this application; Figure 3 A technical flowchart of an online method for estimating the state of charge of a lightweight battery driven by an electrochemical model, provided in another embodiment of this application; Figure 4 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0016] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0017] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] The electrochemical model-driven online state-of-charge estimation method for lightweight batteries provided in this application can be applied to, for example... Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be set up independently, integrated into server 104, or placed in the cloud or on other servers. Terminal 102 can send the battery material parameters and geometric parameters to be processed to server 104. After receiving the battery material parameters and geometric parameters, server 104 establishes a high-fidelity electrochemical model based on the battery material parameters and geometric parameters; on a three-dimensional operating point grid composed of state of charge, current rate, and temperature, it acquires baseline terminal voltage data and sparse terminal voltage frequency response data at each operating point based on the high-fidelity electrochemical model; it performs frequency domain reconstruction and enhancement of the sparse terminal voltage frequency response data based on a physical information neural network to obtain enhanced terminal voltage frequency response data; and it performs further processing on the enhanced terminal voltage frequency response data at each operating point based on the Loewner data-driven order reduction algorithm. The process involves order reduction modeling and stabilization to obtain local terminal voltage order reduction models at each operating point. Based on these local order reduction models, a three-dimensional scheduling linear parameter variation terminal voltage prediction model is constructed. This model retains the independent states of each local terminal voltage order reduction model and establishes a fusion mechanism for the outputs of neighboring local terminal voltage order reduction models based on state of charge, current ratio, and temperature. According to the current state of charge, scheduling current ratio, and scheduling temperature, the predicted terminal voltage is obtained based on the three-dimensional scheduling linear parameter variation terminal voltage prediction model. The predicted state of charge to be corrected is acquired, and the terminal voltage residual is obtained based on the predicted and measured terminal voltages. This residual is then used to correct the predicted state of charge to be corrected, resulting in a final estimated state of charge. The server 104 can then feed back the final estimated state of charge to the terminal 102. In addition, in some embodiments, the electrochemical model-driven online estimation method for the state of charge of lightweight batteries can also be implemented by the server 104 or the terminal 102 separately. For example, the terminal 102 can directly estimate the battery material parameters and geometric parameters to be processed online, or the server 104 can obtain the battery material parameters and geometric parameters to be processed from the data storage system and estimate them online.

[0019] The terminal 102 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 104 can be implemented using a standalone server or a server cluster composed of multiple servers, or it can be a cloud server.

[0020] In one exemplary embodiment, such as Figure 2 As shown, an online method for estimating the state of charge of lightweight batteries driven by an electrochemical model is provided. This method is executed by a computer device, specifically a terminal or server, or both. In this embodiment, the method is applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps A1 to A7.

[0021] Step A1: Establish a high-fidelity electrochemical model based on battery material parameters and geometric parameters.

[0022] Step A2: On a three-dimensional operating point grid composed of state of charge, current ratio, and temperature, baseline terminal voltage data and sparse terminal voltage frequency response data at each operating point are obtained based on the high-fidelity electrochemical model.

[0023] Step A3: Based on the physical information neural network, the sparse end voltage frequency response data is reconstructed and enhanced in the frequency domain to obtain the enhanced end voltage frequency response data.

[0024] Step A4: Based on the Loewner data-driven order reduction algorithm, the enhanced terminal voltage frequency response data at each operating point are modeled and stabilized to obtain the local terminal voltage order reduction model at each operating point.

[0025] Step A5: Construct a three-dimensional scheduling linear parameter change terminal voltage prediction model based on the local reduced-order model at each operating point; the three-dimensional scheduling linear parameter change terminal voltage prediction model retains the independent state of each local terminal voltage reduction model, and establishes a fusion mechanism for the output of the neighboring local terminal voltage reduction model according to the state of charge, current ratio and temperature.

[0026] Step A6: Based on the current state of charge, dispatch current ratio, and dispatch temperature, the predicted terminal voltage is obtained using a three-dimensional dispatch linear parameter variation terminal voltage prediction model.

[0027] Step A7: Obtain the predicted state of charge to be corrected, obtain the terminal voltage residual based on the predicted terminal voltage and the measured terminal voltage, and correct the predicted state of charge to be corrected according to the terminal voltage residual to obtain the final estimated state of charge.

[0028] By implementing steps A1 to A7 above, a lightweight online estimation model is constructed based on the high-fidelity electrochemical model. A physical information neural network is used to reconstruct and enhance the sparse frequency response data in the frequency domain, reducing the number of direct simulation frequencies in the high-fidelity electrochemical model and lowering the computational load of offline modeling. Based on the Loewner data-driven order reduction algorithm, the enhanced frequency response data is processed to maintain stability, resulting in a local terminal voltage order reduction model. This reduces the model order while preserving the main dynamic features and terminal voltage response capability. A three-dimensional scheduling linear parameter variation terminal voltage prediction model is used to achieve continuous transition between multiple operating points and real-time operation at the board end, meeting the real-time deployment requirements of the embedded controller. Terminal voltage residual feedback is used to correct the predicted state of charge, improving the accuracy and stability of online state of charge estimation under complex operating conditions. By employing a technical approach of "high-fidelity electrochemical model-driven—lightweight model reduction—online estimation at the plate end," this approach addresses, to some extent, the limitations of existing technologies where high-fidelity electrochemical models are difficult to directly apply to real-time plate-end operation, traditional simplified models fail to adequately characterize internal mechanisms, and online estimation accuracy and stability under complex operating conditions are poor. This approach achieves a balance between mechanism consistency, model lightweighting, and real-time engineering deployment capabilities. Furthermore, this application can be extended to other electrochemical energy storage systems. By replacing the corresponding high-fidelity electrochemical model parameters and geometric parameters, online estimation of the state of charge (SOC) in different systems such as sodium-ion batteries, solid-state batteries, and flow batteries can be achieved.

[0029] In another exemplary embodiment of this application, the high-fidelity electrochemical model in step A1 above includes a pseudo-two-dimensional electrochemical model and a lumped thermal model; the pseudo-two-dimensional electrochemical model is used to characterize the mass transfer, diffusion, polarization and electrochemical reaction processes inside the battery, and the lumped thermal model is used to characterize the influence of the overall temperature change of the battery cell on the dynamic characteristics of the terminal voltage.

[0030] In this embodiment, the high-fidelity electrochemical model uses the P2D model built into Pybamm and selects the Chen2020 parameter set. The high-fidelity electrochemical model is a model built on a pseudo-two-dimensional electrochemical model to characterize the internal mass transfer, diffusion, polarization, and electrochemical reaction processes of lithium-ion batteries. Specifically, the pseudo-two-dimensional electrochemical model includes: a volume-averaged approximate equation for charge conservation in the porous electrode solid phase, a solid phase mass conservation equation, a volume-averaged approximate equation for charge conservation in the porous electrode electrolyte phase, a volume-averaged approximate equation for mass conservation in the porous electrode electrolyte phase, a volume-averaged approximate equation for the microscopic Butler-Volmer kinetic relationship, and open-circuit potential functions for the negative and positive electrodes.

[0031] The volume-averaged approximate equation for the solid-phase charge conservation of porous electrodes is as follows: ; in, This is the divergence operator, used to measure the net outflow of a vector field at a point in space. In this formula, it calculates... The divergence of this vector. The volume average conductivity of a solid medium in a porous medium near a given point is given in units of . , The volume-average potential near a given point in a solid electrode, in units of . , The gradient of the solid-state potential represents the spatial rate of change of the potential, with units of . , Interfacial area, representing the area of ​​the boundary between the solid and electrolyte per unit volume, is expressed in units of 1000 m². , It is the Faraday constant, with units of 1000 ppm. , The volume average flux density is expressed in units of 1000 kJ / m². .

[0032] The solid mass conservation equation is: ; in, The lithium concentration in the solid electrode, in units of , For time, the unit is , The radial coordinates inside the spherical particle are given in units of 1. , The diffusion coefficient in a solid electrode, in units of . , The concentration is the rate of change over time.

[0033] The volume-averaged approximate equation for the charge conservation of the porous electrode electrolyte phase is as follows: ; in, This is the divergence operator, used to measure the net outflow of a vector field at a point in space. In this formula, it calculates... The divergence of this vector, The effective ionic conductivity, averaged by volume near a given point, is expressed in units of... , The effective diffusion conductivity is related to the diffusion potential caused by the difference in ion concentration. ; The volume average potential of the liquid phase, in units of , This represents the lithium concentration in the electrolyte, in units of... , Interfacial area, representing the area of ​​the boundary between the solid and electrolyte per unit volume, is expressed in units of 1000 m². , It is the Faraday constant, with units of 1000 ppm. , The volume average flux density is expressed in units of 1000 kJ / m². .

[0034] The volume-averaged approximate equation for the mass conservation of the electrolyte phase in porous electrodes is as follows: ; in, It represents the volume fraction of the electrolyte near a certain point, with units of . , This represents the lithium concentration in the electrolyte, in units of... , The transference number of the cation relative to the solvent, in units of 1. , This represents the number of moles of lithium in the liquid phase per unit volume of the electrode. The effective diffusion coefficient represents the actual diffusion rate of ions in a complex porous electrode. This is the divergence operator, used to measure the net outflow of a vector field at a point in space. In this formula, it calculates... The divergence of this vector, Interfacial area, representing the area of ​​the boundary between the solid and electrolyte per unit volume, is expressed in units of 1000 m². , For time, the unit is , The volume average flux density is expressed in units of 1000 kJ / m². .

[0035] The volume-averaged approximation equation for the microscopic Butler-Volmer dynamics is: ; The volume average flux density is expressed in units of 1000 kJ / m². ; Volumetric flux density, in units of ; The concentration of solid lithium on the surface of the particles is used to calculate the overpotential of the electrochemical reaction. The average lithium-ion concentration in the electrolyte, in units of ; The volume-average potential near a given point in a solid electrode, in units of . ; The volume average potential of the liquid phase, in units of ; This represents the reaction overpotential at the battery scale. The unit is , This is an open-circuit potential function, representing the equilibrium potential of the electrode material at a specific lithium concentration. For SEI film resistance; Exchange current density, unit: R is the ideal gas constant, R = ; T Temperature, unit: K ; The charge transfer coefficient; It is the Faraday constant, with units of 1000 ppm. ; The volume average flux density is expressed in units of 1000 kJ / m². .

[0036] The open-circuit potential functions of the positive and negative terminals are given by the parameter set used, and can be uniformly expressed as: ; in, The open-circuit potential function of the negative electrode. The positive open-circuit potential function. and The values ​​represent the lithiation fractions of the negative and positive electrode materials, respectively.

[0037] The battery terminal voltage is expressed as: ; in, Terminal voltage, and These are the solid-phase potentials of the positive and negative electrodes, respectively.

[0038] The formula for the lumped heat model is: ; in T For cell temperature, For ambient temperature, For equivalent heat capacity, The equivalent heat dissipation coefficient, This is for heat generation in battery cells.

[0039] In another exemplary embodiment of this application, the three-dimensional operating point grid in step A2 above includes multiple state-of-charge (SOC) operating points, multiple current-rate operating points, and multiple temperature operating points; the SOC operating points cover low SOC region, medium SOC region, and high SOC region; the current-rate operating points cover charging condition, resting condition, and discharging condition, with positive current rate representing discharging and negative current rate representing charging; the temperature operating points cover low temperature condition, medium temperature condition, and high temperature condition.

[0040] In this embodiment, the three-dimensional working point mesh can be represented as: ; in, For the set of SOC working points, For the C-rate working point set, This is the set of operating temperature points, where a positive C-rate represents discharge and a negative C-rate represents charging.

[0041] For any working point A high-fidelity model is initialized with the target SOC and a DC bias current is applied. The pre-bias time is set to establish a baseline for local concentration polarization and electrochemical dynamics.

[0042] In another exemplary embodiment of this application, in order to reduce the direct simulation computation of the high-fidelity electrochemical model, step A2 above is replaced by steps B1 to B4: Step B1: Construct a three-dimensional operating point mesh based on the state of charge, current ratio, and temperature.

[0043] Step B2: Set the DC bias current according to the current multiplier at each operating point, perform DC pre-biasing to establish quasi-steady-state polarization, and run pure DC baseline simulation and DC superimposed AC disturbance simulation from the same pre-bias final state.

[0044] Step B3: Obtain baseline terminal voltage data through the pure DC baseline simulation.

[0045] Step B4: Align the outputs of the pure DC baseline simulation and the DC superimposed AC disturbance simulation with the same equidistant timestamp array, obtain the small-signal voltage response by subtracting them point by point, and obtain the sparse-end voltage frequency response data by frequency domain analysis.

[0046] In this embodiment, simulation operating points consisting of different states of charge, current ratios, and temperatures are selected. The required DC bias excitation current is calculated from the current ratio and nominal capacity. The DC bias excitation current at the corresponding operating point is input to make the internal state of the battery evolve to the vicinity of the operating point. The voltage at this time is recorded as the baseline terminal voltage, and a baseline terminal voltage space for multiple operating points is constructed. Then, pure DC baseline simulation and DC superimposed small signal perturbation simulation are performed on this basis. The results of the two simulations are subtracted and then a fast Fourier transform is performed to obtain the frequency response of the sparse terminal voltage of the battery at different operating points.

[0047] After the pre-biasing is complete, a pure DC baseline simulation and a DC-plus-AC disturbance simulation are performed from the same final state. Both simulations use the exact same equidistant timestamp array. ; in, Δt The sampling interval is... This is the length of the frequency response extraction window. By forcing the two simulation trajectories to be output at the same timestamp, secondary numerical errors caused by post-processing interpolation are avoided.

[0048] The small-signal response of the terminal voltage is obtained by subtracting it point by point: .

[0049] The input small signal is: .

[0050] right ΔV(t) and ΔI(t) Frequency domain analysis was performed to obtain the terminal voltage frequency response at the target frequency point: .

[0051] In another exemplary embodiment of this application, the lowest frequency of the AC disturbance in step B2 is dynamically truncated according to the current ratio to limit the state of charge drift within the simulation window; the amplitude of the AC disturbance is determined according to a preset ratio of the nominal 1C current, and the sparse terminal voltage frequency response data is sampled for quality inspection by phase-locked loop or least squares method.

[0052] In this embodiment, the frequency response data extraction process employs low-frequency dynamic truncation to control low-frequency data drift and uses single-frequency phase-locked loop to verify the correctness of the frequency response results.

[0053] The low-frequency dynamic truncation is to avoid excessive SOC drift within a single operating point caused by low-frequency long-period disturbances. The minimum excitation frequency is dynamically truncated according to the C-rate. The minimum frequency is increased at high-rate charge / discharge operating points, and lower frequencies are allowed at 0C or low-rate operating points.

[0054] The single-frequency phase-locked loop (PLL) is used to ensure the effectiveness of the small-signal frequency response. The AC disturbance amplitude should not be too small to avoid being overwhelmed by numerical noise, nor should it be too large to avoid triggering Butler-Volmer nonlinear high-order harmonics. The multi-sinusoidal frequency response extraction results can be sampled and checked using the PLL method or the least squares method.

[0055] In another exemplary embodiment of this application, the physical information neural network in step A3 above takes the normalized logarithmic frequency as input and the real and imaginary parts of the complex frequency response of the terminal voltage as output; the physical information neural network is trained using data consistency loss, frequency direction smoothing loss and causality loss based on Kramers-Kronig relationship.

[0056] In this embodiment, the total loss function of the physical information neural network is a weighted combination of three types of losses: data consistency loss, frequency direction second-order smoothing loss, and causality loss based on the Kramers-Kronig relationship.

[0057] In this embodiment, the total loss function of the physical information neural network is: ; in, This is the total loss function of the physical information neural network; These are the weighting coefficients for data consistency, smoothing regularization, and causality constraints, respectively. Let this be the data consistency loss function; The frequency-direction second-order smoothing loss function; This is the causal loss function.

[0058] Let the frequency response data of any of the above models be: ; The input is the applied current. The output is the frequency response of the above models; The frequency response of the model; ω is the angular frequency.

[0059] To control costs, only a finite number of sparse sampling frequency points are obtained, denoted as the sparse sample set: ; in, It is a sparse sample set; The number of sparse samples; For the first angular frequency at each sampling point; In the first Frequency points The actual complex frequency response value of the system; Given The source definition, in At this purely imaginary point, the frequency response function The chosen values ​​are based solely on the Laplace variable, as only the steady-state sinusoidal response is considered. The value on the imaginary axis; It is the imaginary unit; the goal is to reconstruct the continuous complex frequency response function over the entire target frequency band. This provides enhanced frequency domain samples for subsequent Loewner interpolation order reduction.

[0060] Since the frequency range typically spans multiple orders of magnitude, directly using Using these as inputs leads to unstable training; therefore, the frequencies are first logarithmed and normalized. Defined as: ; in, The normalized logarithmic frequency, ; ω is the angular frequency.

[0061] Therefore, the input to PINN is no longer the original frequency. It is the normalized logarithmic frequency. .

[0062] For each fixed target, construct a one-dimensional frequency PINN: ; in, For network parameters; The frequency response function predicted by PINN; For weights The neural network.

[0063] Since the frequency response is complex, the network output is denoted as: ; in, The frequency response function predicted by PINN; and These represent the real and imaginary parts of the predicted frequency response, respectively.

[0064] Thus, PINN's objective narrows from "two-dimensional frequency response field reconstruction in space and frequency" to "one-dimensional frequency function reconstruction with a fixed objective".

[0065] At sparse measurement frequencies, the network output must be consistent with the original frequency response samples. The data consistency loss function is defined as follows: ; in, Let this be the data consistency loss function; For normalization; The number of sparse samples; This is the frequency response function predicted by PINN. To avoid large-amplitude frequencies dominating training, a relative scale form is usually used.

[0066] The normalization scale is expressed by the following formula: ; in, For normalization, In the first Frequency points The actual complex frequency response value of the system; This serves as a lower bound for the scale, used to prevent numerical instability caused by extremely small amplitude frequencies.

[0067] Therefore, the role of the data items is to ensure that the network maintains consistency with the true frequency response at known sparse frequency points.

[0068] Since the frequency response function typically exhibits strong continuity and smoothness as it changes with the logarithmic frequency, a second-order smoothing term in the frequency direction is introduced to suppress non-physical oscillations in PINN between sparse sampling intervals. The second-order smoothing loss function in the frequency direction is defined as: ; in, The frequency-direction second-order smoothing loss function; This represents the number of dense samples; These are the smooth constraint points selected on the dense frequency grid; This is the frequency response function predicted by PINN.

[0069] This essentially constrains the curvature of the frequency response curve on the logarithmic frequency axis, thereby making the reconstruction results smoother and more stable, and improving the quality of subsequent Loewner interpolation samples.

[0070] For linear causal systems, the real and imaginary parts of the frequency response satisfy the Kramers-Kronig relation. Let's assume a discrete frequency grid... Constructing the discrete Hilbert approximation matrix Then the discrete approximation relation can be written as: ; ; in, For the complex frequency response function predicted by PINN; and For points on a discrete frequency grid, the subscript is used to distinguish the current calculation point from all points involved in the calculation; and These refer to taking the imaginary and real parts of a complex number, respectively. The elements of the Kramers-Kronig (KK) matrix KK are numerical approximations of the continuous Hilbert transform at discrete frequency points. The values ​​of the matrix depend only on the choice of frequency grid and are fixed and can be pre-calculated. This represents the total number of frequency grid points.

[0071] Based on this, a causal loss function is constructed: ; in, The loss function is causal. This represents the total number of frequency grid points. For the complex frequency response function predicted by PINN; and For points on a discrete frequency grid, the subscript is used to distinguish the current calculation point from all points involved in the calculation; and These refer to taking the imaginary and real parts of a complex number, respectively. The elements of the Kramers-Kronig (KK) matrix KK are numerical approximations of the continuous Hilbert transform at discrete frequency points. The values ​​of the matrix depend only on the choice of frequency grid and are fixed and can be pre-calculated.

[0072] The PINN reconstructed frequency response is as self-consistent as possible in terms of resolution and causality, rather than simply fitting an interpolation curve with a finite number of sample points. Used to ensure accurate reconstruction at known sampling points; Used to constrain the continuous and smooth variation of the frequency response on the frequency axis; Complex analytical structures used to constrain frequency response to satisfy causal systems.

[0073] The network parameters are determined through the following optimization problem: ; in, The optimal network parameters; To find the parameters that minimize the objective function The operation; This is the total loss function of the physical information neural network.

[0074] A two-stage training strategy is employed: first, the Adam optimizer is used to perform a coarse global search; then, L-BFGS-B is used for fine optimization within the local neighborhood. This approach balances global convergence capability in the early stages of training with local refinement capability in the later stages.

[0075] After training, the continuous frequency response surrogate function of the target transfer function is obtained: ; in, This is the optimal continuous frequency response surrogate function obtained after training. Optimal parameters were loaded Neural networks; Normalized logarithmic frequency; The target frequency band range.

[0076] The surrogate function can output complex frequency response values ​​at any frequency point, thus forming an enhanced frequency domain sample input for the subsequent Loewner algorithm.

[0077] Therefore, the essence of this step can be summarized as follows: using one-dimensional frequency PINN to reconstruct the sparse frequency response samples of the fixed target transfer function into a continuous complex frequency response function that satisfies the constraints of data consistency, smoothness and causality.

[0078] In another exemplary embodiment of this application, in order to clarify the algorithm flow of Loewner's data-driven order reduction, step A4 above is replaced by steps C1 to C2: Step C1: Construct the Loewner matrix and displacement Loewner matrix based on the terminal voltage frequency response data at each operating point, determine the target model order at each operating point, and extract the continuous-time state-space model at each operating point.

[0079] Step C2: Perform pole stability checks on the continuous-time state-space model at each operating point. When unstable poles exist, use the stable pole reflection method to stabilize them. Discretize the stabilized model to obtain the local terminal voltage reduced-order model at each operating point.

[0080] In another exemplary embodiment of this application, the three-dimensional scheduling linear parameter change terminal voltage prediction model in step A5 above synchronously updates the state of the plurality of local terminal voltage reduction models in each control cycle; the scheduling current multiplier is obtained by dividing the slow-changing bias current by the nominal IC current; the slow-changing bias current is obtained by the real-time input current through a first-order low-pass filter; and the scheduling temperature is obtained by the real-time acquired battery temperature through a low-pass filter.

[0081] In this embodiment, multiple local terminal voltage reduction models are run simultaneously during online operation. Based on the current state of charge, current ratio, and temperature, eight local terminal voltage reduction models at the vertices of the cube containing the current state point are linearly fused in the three-dimensional baseline terminal voltage space using distance weights. The baseline terminal voltages of the eight models are fused to obtain the predicted baseline terminal voltage. The dynamic outputs of the eight models are then fused using the same linear weights to obtain the dynamic compensation amount of the terminal voltage. The predicted baseline terminal voltage is then added to the dynamic compensation amount of the terminal voltage to obtain the predicted terminal voltage.

[0082] A fusion mechanism is established based on the state of charge, current rate, and temperature to create a reduced-order model output of the neighboring local terminal voltage. Specifically, the fusion mechanism is as follows: the state of all local terminal voltage ROMs is synchronously advanced in each sampling period, and the current output is obtained by fusing the local model outputs in the rectangular neighborhood where the previous SOC, slow bias scheduling C-rate, and filtered temperature are located according to multilinear weights.

[0083] The slow bias scheduling C-rate and filtered temperature are obtained by first-order filtering of the real-time input current and the real-time temperature, respectively. In order to prevent the scheduling of C-rate and temperature from causing large drifts due to abrupt changes, they are filtered first before LPV output fusion scheduling.

[0084] The slow bias scheduling C-rate and the filter temperature are calculated as follows: Real-time input current is Slow bias current is Its first-order filter update form is: ; ; in, This is the time constant for slow bias current filtering.

[0085] The actual C-rate and the scheduling C-rate are as follows: ; ; in, The nominal 1C current can be obtained by multiplying the given C-rate by the nominal capacity Q.

[0086] The actual C-rate reflects the current instantaneous current magnitude, while the scheduling C-rate reflects the slowly changing operating point, used for querying the baseline table and calculating the LPV fusion weights. The dynamic input to the local model is: .

[0087] The filter temperature is: ; .

[0088] Let the current scheduling variable be The trilinear weights of neighboring operating points are calculated using the SOC grid, C-rate grid, and temperature grid. The dynamic compensation amount for the terminal voltage is: .

[0089] The baseline voltage is obtained by interpolation from a three-dimensional baseline voltmeter: .

[0090] Therefore, the terminal voltage predicted by the LPV model is: .

[0091] The residual between the measured terminal voltage and the predicted terminal voltage is: .

[0092] To suppress voltage measurement noise, a first-order low-pass filter is applied to the terminal voltage residual: .

[0093] .

[0094] in, This is the residual filtering time constant.

[0095] In another exemplary embodiment of this application, in order to clarify the closed-loop correction process for online state of charge estimation, step A7 above is replaced by steps D1 to D4: Step D1: Perform integral calculations based on the input current and battery capacity parameters to obtain the predicted state of charge value to be corrected.

[0096] Step D2: Obtain the terminal voltage residual based on the predicted terminal voltage and the measured terminal voltage.

[0097] Step D3: Construct a residual correction model based on the terminal voltage residual.

[0098] Step D4: Correct the predicted state of charge to be corrected using the residual correction model to obtain the final estimated state of charge.

[0099] In this embodiment, to clarify the closed-loop correction process for online state of charge estimation, the baseline voltage slope is calculated using the center difference of a three-dimensional baseline voltmeter near the current SOC main trend, scheduling C-rate, and scheduling temperature. ; when When the slope is greater than the preset minimum slope threshold and the scheduling point does not exceed the limit, the SOC correction amount is: ; in, For residual gain, Limiting for single-step SOC correction.

[0100] The final SOC estimate is: ; in, and Estimate the boundary for SOC.

[0101] If the baseline slope is too small, the scheduling point is out of bounds, or the residual is abnormal, the current residual correction can be paused, and only the following output will be provided: .

[0102] In another exemplary embodiment of this application, the residual correction model in step C3 is used to filter the terminal voltage residual to obtain the filtered terminal voltage residual, and convert the filtered terminal voltage residual into a state of charge correction amount based on the sensitivity of the current state of charge, the dispatch current ratio, and the baseline terminal voltage data near the dispatch temperature to the state of charge. The state of charge correction amount is subjected to single-step limiting processing to update the predicted state of charge value to be corrected, and the final state of charge estimate is obtained. The sensitivity of the current state of charge, the dispatch current ratio, and the baseline terminal voltage data near the dispatch temperature to the state of charge is calculated based on a three-dimensional baseline terminal voltage table.

[0103] In another exemplary embodiment of this application, Figure 3 This paper illustrates the overall flow of an electrochemical model-driven online method for estimating the state of charge (SOC) of lightweight batteries, as provided in this application. The method comprises two main parts: an offline stage and an online stage.

[0104] The offline phase is used to generate a stable reduced-order terminal voltage model that can be run online, specifically including: Based on a high-fidelity electrochemical model, sparse terminal voltage frequency responses and corresponding baseline terminal voltages are extracted from a pre-defined grid of operating points (SOC, rate, and temperature) to form a multi-operating-point dataset. A physical information neural network is used to reconstruct and enhance the sparse terminal voltage frequency responses in the frequency domain, resulting in enhanced frequency domain samples. The Loewner order reduction method is applied to the enhanced frequency response data for each operating point to construct a local low-order state-space model, yielding a continuous state-space matrix. After discretization and stability checks, the discrete low-order model matrix corresponding to each operating point is obtained. Finally, a three-dimensional LPV output fusion model is established based on SOC, slow bias rate, and battery temperature, enabling the plate to fuse the local reduced-order model output under different operating conditions through weighted scheduling, thus obtaining the terminal voltage baseline and dynamic terms.

[0105] The online phase is used for real-time estimation and collaboration on the board side, specifically including: The host computer or battery management system collects and processes downlink current, measured terminal voltage, battery temperature, and sampling time intervals in real time. First, the SOC (State of Charge) master state is predicted based on the input current. Simultaneously, the three-dimensional LPV (Limited Voltage Variation) terminal voltage model advances only the local ROM state within the current neighborhood based on the current predicted SOC, slow bias ratio, and temperature scheduling point, calculating the terminal voltage dynamic response. Then, the baseline terminal voltage is obtained by interpolating the baseline terminal voltage data from eight neighboring grid points, and the terminal voltage dynamics are obtained by fusing the outputs of the eight neighboring local ROMs. The two are added together to obtain the predicted terminal voltage, and the residual between the measured and predicted terminal voltages is calculated. To improve the robustness of online estimation, a measurement gating mechanism based on normalized residual squares is introduced to identify and suppress single-point abnormal voltage measurements; simultaneously, weakly observable updates under static or low-dynamic conditions are protected. Finally, a low-dimensional unscented Kalman filter can be used to recursively correct the SOC and necessary slow corrections. The recommended form updates the empirical dynamic polarization gain only when the current step meets the threshold condition, thus avoiding static interval parameter edge-grabbing. After the above prediction, gating and filtering correction, the final state of charge estimate is output.

[0106] In another exemplary embodiment of this application, the parameters used are the Chen2020 parameter set. The Chen2020 parameter set is used for LG M50 lithium-ion cells and includes parameters such as geometric parameters, electrochemical parameters, cutoff voltage, and initial concentration. Table 1 lists the commonly used physical parameters in this embodiment. In specific implementation, these parameters can be recalibrated according to the target cell. Under multiple temperature conditions, multiple operating temperature points can be set around the reference temperature, and the temperature-related parameters, thermal model parameters, and boundary conditions can be calibrated or corrected accordingly.

[0107] Table 1 Commonly Used Physical Parameters in Chen2020 Parameter Set

[0108] This application also provides an application scenario in which the above-mentioned electrochemical model-driven online estimation method for lightweight battery state of charge (SOC) is applied. Specifically, the electrochemical model-driven online estimation method for lightweight battery SOC provided in this embodiment is applied to battery management scenarios in electric vehicles or energy storage systems. The battery management scenario includes a data acquisition stage, a state estimation stage, and an energy management stage. Battery operating data enters the state estimation stage from the data acquisition stage, undergoes electrochemical model-driven lightweight online estimation processing to obtain an accurate SOC, and then enters the downstream energy management stage. The online SOC estimation method provided in this embodiment belongs to the model estimation sub-stage within the state estimation stage. Specifically, in the process of estimating the battery's SOC, a collaborative approach based on an electrochemical mechanism model and data-driven order reduction can be used to estimate the battery's SOC, thus providing high-precision SOC information to the battery management system.

[0109] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 4As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores the final state-of-charge (SOC) estimate. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements an electrochemical model-driven online SOC estimation method for lightweight batteries.

[0110] Those skilled in the art will understand that Figure 4 The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0111] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0112] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0113] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0114] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0115] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0116] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0117] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A lightweight battery state-of-charge estimation method driven by an electrochemical model, characterized in that, include: A high-fidelity electrochemical model was established based on battery material parameters and geometric parameters. On a three-dimensional operating point grid consisting of state of charge, current ratio, and temperature, baseline terminal voltage data and sparse terminal voltage frequency response data at each operating point are obtained based on the high-fidelity electrochemical model. The frequency domain reconstruction and enhancement of the sparse end voltage frequency response data are performed based on the physical information neural network to obtain the enhanced end voltage frequency response data. Based on the Loewner data-driven order reduction algorithm, the frequency response data of the enhanced terminal voltage at each operating point are modeled and stabilized to obtain the local terminal voltage order reduction model at each operating point. A three-dimensional scheduling linear parameter change terminal voltage prediction model is constructed based on the local reduced-order model at each operating point; the three-dimensional scheduling linear parameter change terminal voltage prediction model retains the independent state of each local terminal voltage reduction model, and establishes a fusion mechanism for the output of the neighboring local terminal voltage reduction models according to the state of charge, current ratio and temperature; Based on the current state of charge, dispatch current ratio, and dispatch temperature, the predicted terminal voltage is obtained using a three-dimensional dispatch linear parameter variation terminal voltage prediction model. Obtain the predicted state of charge to be corrected, obtain the terminal voltage residual based on the predicted terminal voltage and the measured terminal voltage, and correct the predicted state of charge to be corrected according to the terminal voltage residual to obtain the final estimated state of charge.

2. The electrochemical model-driven online estimation method for the state of charge of lightweight batteries according to claim 1, characterized in that, The high-fidelity electrochemical model includes a pseudo-two-dimensional electrochemical model and a lumped thermal model. The pseudo-two-dimensional electrochemical model is used to characterize the mass transfer, diffusion, polarization and electrochemical reaction processes inside the battery, while the lumped thermal model is used to characterize the impact of the overall temperature change of the battery cell on the dynamic characteristics of the terminal voltage.

3. The electrochemical model-driven online estimation method for the state of charge of lightweight batteries according to claim 1, characterized in that, The three-dimensional operating point grid includes multiple state-of-charge operating points, multiple current-rate operating points, and multiple temperature operating points; the state-of-charge operating points cover low-charge state region, medium-charge state region, and high-charge state region. The current ratio operating point covers charging, resting, and discharging conditions, with a positive current ratio representing discharging and a negative current ratio representing charging; the temperature operating point covers low-temperature, medium-temperature, and high-temperature conditions.

4. The electrochemical model-driven online estimation method for the state of charge of lightweight batteries according to claim 1, characterized in that, On a three-dimensional operating point grid composed of state of charge, current rate, and temperature, baseline terminal voltage data and sparse terminal voltage frequency response data at each operating point are obtained based on the high-fidelity electrochemical model, specifically including: A three-dimensional operating point mesh is constructed based on the state of charge, current ratio, and temperature. The DC bias current is set according to the current ratio at each operating point, and DC pre-biasing is performed to establish quasi-steady-state polarization. Pure DC baseline simulation and DC superimposed AC disturbance simulation are run from the same pre-bias final state. Baseline terminal voltage data were obtained through the pure DC baseline simulation. The outputs of the pure DC baseline simulation and the DC superimposed AC disturbance simulation are aligned with the same equidistant timestamp array. The small-signal voltage response is obtained by subtracting them point by point, and the sparse-end voltage frequency response data is obtained by frequency domain analysis.

5. The method according to claim 4, characterized in that, The lowest frequency of the AC disturbance is dynamically truncated according to the current ratio to limit the state of charge drift within the simulation window; the amplitude of the AC disturbance is determined according to a preset ratio of the nominal 1C current, and the quality of the sparse terminal voltage frequency response data is sampled by phase-locked loop or least squares method.

6. The electrochemical model-driven online estimation method for the state of charge of lightweight batteries according to claim 1, characterized in that, The physical information neural network takes the normalized logarithmic frequency as input and the real and imaginary parts of the complex frequency response of the terminal voltage as output; the physical information neural network is trained using data consistency loss, frequency direction smoothing loss and causality loss based on Kramers-Kronig relationship.

7. The electrochemical model-driven online estimation method for the state of charge of lightweight batteries according to claim 1, characterized in that, Based on the Loewner data-driven order reduction algorithm, the enhanced terminal voltage frequency response data at each operating point are subjected to order reduction modeling and stabilization processing to obtain the local terminal voltage order reduction model at each operating point, specifically including: Based on the terminal voltage frequency response data at each operating point, construct the Loewner matrix and displacement Loewner matrix respectively, determine the target model order at each operating point and extract the continuous time state space model at each operating point. Pole stability checks are performed on the continuous-time state-space model at each operating point. When unstable poles exist, the stable pole reflection method is used for stabilization. The stabilized model is then discretized to obtain the local terminal voltage reduced-order model at each operating point.

8. The electrochemical model-driven online estimation method for the state of charge of lightweight batteries according to claim 1, characterized in that, The three-dimensional scheduling linear parameter change terminal voltage prediction model synchronously updates the state of the multiple local terminal voltage reduction models in each control cycle. The scheduling current multiplier is obtained by dividing the slow-variable bias current by the nominal 1C current; The slow-varying bias current is obtained by filtering the real-time input current through a first-order low-pass filter. The scheduling temperature is obtained by low-pass filtering the real-time collected battery temperature.

9. The electrochemical model-driven online estimation method for the state of charge of lightweight batteries according to claim 1, characterized in that, Obtain the predicted state of charge (SCC) value to be corrected, obtain the terminal voltage residual based on the predicted terminal voltage and the measured terminal voltage, and correct the predicted SCC value to be corrected based on the terminal voltage residual to obtain the final estimated SCC value. Specifically, this includes: The predicted state of charge to be corrected is obtained by integral calculation based on the input current and battery capacity parameters. The terminal voltage residual is obtained based on the predicted terminal voltage and the measured terminal voltage. A residual correction model is constructed based on the terminal voltage residual; The predicted state of charge is corrected by the residual correction model to obtain the final estimated state of charge.

10. The electrochemical model-driven online estimation method for the state of charge of lightweight batteries according to claim 9, characterized in that, The residual correction model is used to filter the terminal voltage residual to obtain the filtered terminal voltage residual. Based on the sensitivity of the current state of charge, dispatch current ratio, and baseline terminal voltage data near the dispatch temperature to the state of charge, the filtered terminal voltage residual is converted into a state of charge correction amount. The state of charge correction amount is subjected to single-step limiting processing to update the predicted state of charge value to be corrected, and the final state of charge estimate is obtained. The sensitivity of the current state of charge, dispatch current ratio, and baseline terminal voltage data near the dispatch temperature to the state of charge is calculated based on a three-dimensional baseline terminal voltage meter.

Images