A high-precision temperature field modeling method for power batteries based on embedded fiber optic sensors

Abstract

The present invention provides a high-precision temperature field modeling method for power batteries based on embedded optical fiber sensors. It acquires internal and external temperature data of the battery by building a distributed optical fiber temperature sensing system in a non-destructive manner, and only uses a small number of parameters such as battery voltage, current, and temperature to establish a battery internal temperature field prediction model based on a GRU neural network. In the training of the model, a Bayesian optimization algorithm is introduced for hyperparameter search, which can effectively improve the training speed. The combined estimation results of the battery SOC and SOH based on EKF are used as model input to ensure the accuracy of temperature field modeling. The method of the present invention can be widely applied to various models of commercial batteries, and is helpful for early identification of various faults such as negative electrode lithium deposition, internal short circuit, overcharging, etc. that may cause abnormal internal high temperature, thereby improving the safety of the battery and improving the level of intelligent management of power batteries from an overall level.

Description

Technical Field

[0001] The present invention belongs to the technical field of battery testing and management monitoring, and in particular relates to a high-precision temperature field modeling method for a power battery based on an embedded optical fiber sensor. Background Art

[0002] Temperature has a high correlation with the working performance and cycle life of power batteries. Accurate and timely monitoring of it can help to diagnose and early-warn battery thermal runaway failures, thereby ensuring the safety of battery use. At present, some existing technologies related to battery temperature monitoring have been expanded from the battery surface to the inside, which has greatly improved the characterization effect of battery temperature compared to the early days. The internal battery temperature monitoring methods used are mainly divided into three categories: direct measurement method, model method and data-driven method. Among them, the measurement method directly measures the internal temperature through embedded sensors or indirectly measures parameters such as the electrochemical impedance of the power battery to reflect the temperature information inside the power battery. However, there is no clear conclusion on the impact of destructive sensor implantation on important performance parameters such as power battery capacity, life and safety, and the implantation process also lacks clear standards. It is difficult to achieve commercial application in the short term; the measurement of electrochemical impedance parameters requires a large amount of experimental data calibration, and can only reflect the overall temperature state, which is of little help in temperature field modeling. Model-based approaches estimate the internal battery temperature by establishing distributed or lumped parameter models. However, the resulting temperature estimation is overly dependent on modeling accuracy. Furthermore, since most existing battery thermal models are simplified and approximated, this can easily lead to temperature estimation bias and only yields temperature estimates at a few locations. Data-driven approaches utilize machine learning to directly establish internal battery temperature prediction models based on a large amount of temperature data collected during battery operation. This facilitates online applications, but the quality of test and temperature data under different experimental conditions directly impacts the model's generalization and accuracy. While some artificial intelligence-based battery temperature modeling techniques can improve data quality and modeling accuracy to a certain extent, their complex parameter adjustment and slow learning speed remain difficult to address. Summary of the Invention

[0003] In view of this, and in response to the technical problems existing in this field, the present invention provides a method for high-precision temperature field modeling of a power battery based on an embedded optical fiber sensor, which specifically includes the following steps:

[0004] Step 1: Place optical fiber transmission devices on the surface of commercial power battery cells and perform non-destructive optical fiber implantation into the battery cells to form a distributed optical fiber temperature sensor system for measuring the surface and internal temperature of the power battery.

[0005] Step 2: Establish a battery bipolarization equivalent circuit model containing two RC networks and the corresponding state equation for the battery cell, and discretize and linearize the state equation to obtain a form consisting of voltage and current data matrices and a system parameter matrix;

[0006] Step 3: Use the collected battery voltage and current data to identify the system parameter matrix and parse out the ohmic internal resistance R0 and polarization internal resistance R represented by the system parameters. D1 and R D2 And the polarization capacitance C D1 and C D2 ;

[0007] Step 4: The state variables are composed of the power battery SOC and the loop voltages of the two RC networks. The terminal voltage is used as the observation variable. The system state equation and observation equation are established, and the SOC and SOH of the power battery are jointly estimated based on the extended Kalman filter.

[0008] Step 5: Establish a power battery internal temperature field prediction model based on a gated recurrent unit (GRU) neural network, and perform a hyperparameter search for the GRU neural network using a Bayesian optimization algorithm. The power battery internal temperature field prediction model uses the combined estimation results of the battery voltage, current, SOC, and SOH, as well as the battery's external temperature, as input, and outputs the temperature field distribution inside the battery.

[0009] Step 6: Using the battery surface temperature data collected by the distributed fiber optic temperature sensor system, the power battery voltage and current data, and the joint estimation results of SOC and SOH, the established power battery internal temperature field prediction model is trained, and the battery internal temperature collected by the distributed fiber optic temperature sensor is used to verify the training accuracy of the model;

[0010] Step seven: Apply the trained power battery internal temperature field prediction model online so that it can directly output the internal temperature field distribution results based on the power battery voltage, current, SOC and SOH and external temperature data.

[0011] Furthermore, the process of non-destructive optical fiber implantation of the battery cell in step 1 includes:

[0012] (1) Drilling the negative electrode: Use a drill bit with a diameter of 0.9 mm to drill a hole in the center of the negative electrode of the cylindrical power battery cell. The drill bit should drill into the hole to a depth of 0.8 to 1 mm to ensure that the negative electrode shell and the electrode are drilled open. After the negative electrode is drilled, it should be sealed with Kapton glue immediately to prevent leakage of electrolyte.

[0013] (2) Positive electrode cutting: Use pipe cutters to cut open the positive terminal cover of the battery cell. After cutting, check the integrity of the positive electrode sheet and use Kapton tape to insulate the positive electrode sheet from the shell.

[0014] (3) Distributed fiber optic implantation: The distributed fiber optic sensor is implanted from the negative electrode hole, and the optical fiber is led out from the positive electrode and attached to the shells on both sides of the battery cell and fixed with Kapton glue, and then the positive and negative electrodes of the battery cell are resealed with epoxy resin glue;

[0015] (4) Battery cell performance test and temperature perception performance test: After the epoxy resin glue solidifies, measure the open circuit voltage of the battery cell. After confirming that it is correct, place the battery cell on the fixture and carry out charge and discharge tests and internal and external temperature perception performance tests of the battery cell in the temperature chamber;

[0016] The above steps (1) to (3) are all carried out under an environment protected by an inert gas.

[0017] Furthermore, in step 2, the following state equation is established for the cylindrical power battery cell:

[0018] U t (t) = U OC (t)-i L (t)R0(t)-U D1 (t)-U D2 (t)

[0019]

[0020]

[0021] Where U OC is the open circuit voltage, U t is the terminal voltage, U D1 and U D2 are the loop voltages of the two RC networks, i L is the load current, R0 is the battery internal resistance in ohms, R D1 and R D2 is the polarization internal resistance of the two RC networks, C D1 and C D2 are the polarized capacitances of the two RC networks, t represents time, and the superscript · represents the derivative of the corresponding parameter;

[0022] After discretization and linearization, we get:

[0023] U t,k =Φ 2,k θ 2,k

[0024] Φ 2,k =[1,U t,k-1 ,U t,k-2 ,i L,k ,i L,k-1 ,i L,k-2 ]

[0025] θ 2,k =[(1-θ1-θ2)U OC,k ,θ1,θ2,θ3,θ4,θ5] T

[0026] Among them, Φ 2,k is the system voltage and current data matrix, θ 2,k is the system parameter matrix, and k represents the sampling time.

[0027] Furthermore, in step 3, the system parameter matrix θ is identified based on the recursive least squares method with forgetting factor. 2,k :

[0028]

[0029] Where K k is the gain matrix, P k is the covariance error matrix, I is the identity matrix, and μ is the forgetting factor;

[0030] In the identification process, the total voltage and total current in the power battery test data are input in matrix format; after identification, the ohmic internal resistance R0 and polarization internal resistance R D1 and R D2 And the polarization capacitance C D1 and C D2 They are:

[0031]

[0032] Furthermore, the system state equation x established in step 4 k and the observation equation U t,k The specific form is as follows:

[0033] x k =A k x k-1 +B k I k +ω k

[0034] U t,k =C k x k -I k R0+D k-1 +v k

[0035]

[0036] The process of performing joint estimation of the SOC and SOH of the power battery based on the extended Kalman filter includes:

[0037] (1) Initialize the system state: Set the initial state, the state estimation error covariance matrix to P0, the system noise covariance matrix to Q, and the measurement noise covariance matrix to R, where Q and R are constants;

[0038] (2) System state prediction: predict the system state x at the next moment k And update the error covariance matrix P k :

[0039]

[0040] (3) State update: Use the measurement data to correct the state estimate and covariance estimate, and calculate the Kalman gain matrix K in sequence k , correct the system state x k And update the covariance matrix P k :

[0041]

[0042] (4) System state recursion: Let And start the next iteration;

[0043] The EKF is used to quickly recursively obtain the SOC result during the battery cycle. For adjacent sampling points, the changes in battery capacity and SOC are counted to obtain the actual maximum available capacity of the battery and estimate the battery SOH:

[0044]

[0045] After one charge and discharge cycle, the average value of the maximum available capacity of the battery in the entire cycle is calculated, and the capacity parameters of the system state equation in the EKF method are updated in real time.

[0046] Furthermore, the process of searching for hyperparameters of the GRU neural network through Bayesian optimization in step 5 includes:

[0047] (1) Initialization exploration: Randomly sample the initial observation points in the set hyperparameter space to obtain the observation points x1,…,x t , and then obtain the GRU model hyperparameter objective function observations F(x1),…,F(x t );

[0048] (2) Surrogate function fitting: Based on the existing observation points and observation values, solve the distribution parameter Θ of the surrogate function and calculate the 95% confidence interval of all unexplored spaces;

[0049] (3) Exploration point selection: According to the upper bound of the confidence interval of the objective function, the exploration utility function of the hyperparameter objective function is obtained, and then the maximum point of the exploration utility function is calculated, which is the next exploration point x* ;

[0050] (4) Iterative exploration: Obtain the hyperparameter objective function observation value F(x * ), repeat the above steps (2) and (3) until the target accuracy requirement or the search number limit is reached, and finally output the optimal hyperparameter combination obtained by the search.

[0051] The high-precision temperature field modeling method for power batteries based on embedded optical fiber sensors provided by the present invention acquires the internal and external temperature data of the battery by building a distributed optical fiber temperature sensing system in a non-destructive manner, and only uses a small number of parameters such as battery voltage, current, and temperature to establish a battery internal temperature field prediction model based on a GRU neural network; the Bayesian optimization algorithm is introduced in the training of the model for hyperparameter search, which can effectively improve the training speed, and the joint estimation results of the battery SOC and SOH based on the EKF are used as the model input, so that the accuracy of the temperature field modeling is also guaranteed. The method of the present invention can be widely applied to various models of commercial batteries, and is helpful for early identification of various faults such as negative electrode lithium plating, internal short circuit, overcharging, etc. that may cause abnormal internal high temperature, thereby improving the safety of the battery and improving the level of intelligent management of power batteries from an overall level. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A flowchart of the method provided by the present invention;

[0053] Figure 2 A flow chart for non-destructive distributed fiber optic implantation into battery cells;

[0054] Figure 3 Flowchart for modeling the GRU neural network based on the Bayesian optimization algorithm. DETAILED DESCRIPTION

[0055] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0056] The present invention provides a high-precision temperature field modeling method for power batteries based on embedded optical fiber sensors, such as Figure 1 As shown, the specific steps include:

[0057] Step 1: Place optical fiber transmission devices on the surface of commercial power battery cells and perform non-destructive optical fiber implantation into the battery cells to form a distributed optical fiber temperature sensor system for measuring the surface and internal temperature of the power battery.

[0058] Step 2: Establish a battery bipolarization equivalent circuit model containing two RC networks and the corresponding state equation for the battery cell, and discretize and linearize the state equation to obtain a form consisting of voltage and current data matrices and a system parameter matrix;

[0059] Step 3: Use the collected battery voltage and current data to identify the system parameter matrix and parse out the ohmic internal resistance R0 and polarization internal resistance R represented by the system parameters. D1 and R D2 And the polarization capacitance C D1 and C D2 ;

[0060] Step 4: The state variables are composed of the power battery SOC and the loop voltages of the two RC networks. The terminal voltage is used as the observation variable. The system state equation and observation equation are established, and the SOC and SOH of the power battery are jointly estimated based on the extended Kalman filter.

[0061] Step 5: Establish a power battery internal temperature field prediction model based on a gated recurrent unit (GRU) neural network, and perform a hyperparameter search for the GRU neural network using a Bayesian optimization algorithm. The power battery internal temperature field prediction model takes the estimated results of the battery voltage, current, SOC, and SOH, as well as the external temperature, as input, and outputs the temperature field distribution inside the battery.

[0062] Step 6: Using the battery surface temperature data collected by the distributed fiber optic temperature sensor system, the power battery voltage and current data, and the estimated results of SOC and SOH, the established power battery internal temperature field prediction model is trained, and the battery internal temperature collected by the distributed fiber optic temperature sensor is used to verify the training accuracy of the model;

[0063] Step seven: Apply the trained power battery internal temperature field prediction model online so that it can directly output the internal temperature field distribution results based on the power battery voltage, current, SOC and SOH and external temperature data.

[0064] In a preferred embodiment of the present invention, the process of non-destructive optical fiber implantation of the battery cell in step 1 is as follows: Figure 2 Shown, including:

[0065] (1) Drilling the negative electrode: Use a drill bit with a diameter of 0.9 mm to drill a hole in the center of the negative electrode of the cylindrical power battery cell. The drill bit should drill into the hole to a depth of 0.8 to 1 mm to ensure that the negative electrode shell and the electrode are drilled open. After the negative electrode is drilled, it should be sealed with Kapton glue immediately to prevent leakage of electrolyte.

[0066] (2) Positive electrode cutting: Use pipe cutters to cut open the positive terminal cover of the battery cell. After cutting, check the integrity of the positive electrode sheet and use Kapton tape to insulate the positive electrode sheet from the shell.

[0067] (3) Distributed fiber optic implantation: The distributed fiber optic sensor is implanted from the negative electrode hole, and the optical fiber is led out from the positive electrode and attached to the shells on both sides of the battery cell and fixed with Kapton glue, and then the positive and negative electrodes of the battery cell are resealed with epoxy resin glue;

[0068] (4) Battery cell performance test and temperature perception performance test: After the epoxy resin glue solidifies, measure the open circuit voltage of the battery cell. After confirming that it is correct, place the battery cell on the fixture and carry out charge and discharge tests and internal and external temperature perception performance tests of the battery cell in the temperature chamber;

[0069] The above steps (1) to (3) are all carried out under an environment protected by an inert gas.

[0070] The above process does not require disassembly of the battery cells, so it will not affect the normal performance and safety of the battery.

[0071] In a preferred embodiment of the present invention, in step 2, the following state equation is established for the cylindrical power battery cell:

[0072] U t (t) = U OC (t)-i L (t)R0(t)-U D1 (t)-U D2 (t)

[0073]

[0074] Where U OC is the open circuit voltage, U t is the terminal voltage, U D1 and U D2 are the loop voltages of the two RC networks, i L is the load current, R0 is the battery internal resistance in ohms, R D1 and R D2 is the polarization internal resistance of the two RC networks, C D1 and C D2 are the polarized capacitances of the two RC networks, t represents time, and the superscript · represents the derivative of the corresponding parameter;

[0075] After discretization and linearization, we get:

[0076] U t,k =Φ 2,k θ 2,k

[0077] Φ 2,k =[1,U t,k-1 ,U t,k-2 ,i L,k ,i L,k-1 ,i L,k-2 ]

[0078] θ 2,k =[(1-θ1-θ2)U OC,k ,θ1,θ2,θ3,θ4,θ5] T

[0079] Among them, Φ 2,k is the system voltage and current data matrix, θ 2,k is the system parameter matrix, and k represents the sampling time.

[0080] In a preferred embodiment of the present invention, in step 3, the system parameter matrix θ is identified based on the recursive least squares method with forgetting factor. 2,k :

[0081]

[0082] Where K k is the gain matrix, P k is the covariance error matrix, I is the identity matrix, and μ is the forgetting factor;

[0083] In the identification process, the total voltage and total current in the power battery test data are input in matrix format; after identification, the ohmic internal resistance R0 and polarization internal resistance R D1 and R D2 And the polarization capacitance C D1 and C D2 They are:

[0084]

[0085] In a preferred embodiment of the present invention, the system state equation x established in step 4 is k and the observation equation U t,k The specific form is as follows:

[0086] x k =A k x k-1 +B k I k +ω k

[0087] U t,k =C k x k -I k R0+D k-1 +v k

[0088]

[0089] The process of performing joint estimation of the SOC and SOH of the power battery based on the extended Kalman filter includes:

[0090] (1) Initialize the system state: Set the initial state, the state estimation error covariance matrix to P0, the system noise covariance matrix to Q, and the measurement noise covariance matrix to R, where Q and R are constants;

[0091] (2) System state prediction: predict the system state x at the next moment k And update the error covariance matrix P k :

[0092]

[0093] (3) State update: Use the measurement data to correct the state estimate and covariance estimate, and calculate the Kalman gain matrix K in sequence k , correct the system state x k And update the covariance matrix P k :

[0094]

[0095] (4) System state recursion: Let And start the next iteration;

[0096] The EKF is used to quickly recursively obtain the SOC result during the battery cycle. For adjacent sampling points, the changes in battery capacity and SOC are counted to obtain the actual maximum available capacity of the battery and estimate the battery SOH:

[0097]

[0098] After one charge and discharge cycle, the average value of the maximum available capacity of the battery in the entire cycle is calculated, and the capacity parameters of the system state equation in the EKF method are updated in real time.

[0099] Since the number of GRU network layers, the number of neurons in each layer and other hyperparameters will greatly affect the model training speed and modeling results, the present invention uses a hyperparameter search based on Bayesian optimization to conduct a rapid search of high-dimensional and complex hyperparameter spaces to optimize the modeling results. Bayesian optimization mainly includes the fitting of surrogate functions and the calculation of exploration utility functions. The present invention uses a multivariate normal distribution surrogate function to approximate the objective function. For the sampling points x1,…,x t (abbreviated as x 1:t ) and the observed values ​​of the objective function F(x1),…,F(x t)(abbreviated as F(x 1:t ), the multivariate normal distribution it obeys is:

[0100]

[0101] Among them, σ t 2 is the noise covariance matrix, μ t is the mean of the multivariate normal distribution, K t The covariance of the distribution is the probability density function that obeys the above multivariate normal distribution law as follows:

[0102]

[0103] Where Θ is the parameter set of the multivariate normal distribution. Since the observation points and observation values ​​are known, the probability density function only uses the parameter set as a variable. Through maximum likelihood estimation, the parameter value that maximizes the probability density can be selected. Convert the probability density function to the log-likelihood function:

[0104]

[0105] This function is a function whose gradient can be calculated by a known formula. The optimal value of the parameter set of the multivariate normal distribution can be solved by the marginal likelihood partial derivative and gradient descent method. The joint Gaussian prior distribution of the target function can be fitted based on the existing observations. For other unobserved points (x * ,y * ), the posterior distribution conditions it satisfies are as follows:

[0106]

[0107] Among them, μ(y * ) is the mean based on the observed values, σ 2 (y * ) is the variance of the observed data. For all undetected points, the 95% confidence interval of the observed value can be calculated as follows:

[0108] 95% CI C =μ(y * )±1.96×σ 2 (y * )

[0109] Specifically in a preferred embodiment of the present invention, the process of performing hyperparameter search for the GRU neural network through Bayesian optimization includes:

[0110] (1) Initialization exploration: Randomly sample the initial observation points in the set hyperparameter space to obtain the observation points x1,…,x t, and then obtain the GRU model hyperparameter objective function observations F(x1),…,F(x t );

[0111] (2) Surrogate function fitting: Based on the existing observation points and observation values, solve the distribution parameter Θ of the surrogate function and calculate the 95% confidence interval of all unexplored spaces;

[0112] (3) Exploration point selection: According to the upper bound of the confidence interval of the objective function, the exploration utility function of the hyperparameter objective function is obtained, and then the maximum point of the exploration utility function is calculated, which is the next exploration point x * ;

[0113] (4) Iterative exploration: Obtain the hyperparameter objective function observation value F(x * ), repeat the above steps (2) and (3) until the target accuracy requirement or the search number limit is reached, and finally output the optimal hyperparameter combination obtained by the search.

[0114] It should be understood that the size of the serial numbers of the steps in the embodiment of the present invention does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0115] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A high-precision temperature field modeling method for a power battery based on an embedded optical fiber sensor, characterized by: The specific steps include: Step 1: Place optical fiber sensors on the surface of commercial power battery cells and perform non-destructive optical fiber implantation into the battery cells to form a distributed optical fiber temperature sensor system for measuring the surface and internal temperature of the power battery. Step 2: Establish a battery bipolarization equivalent circuit model containing two RC networks and the corresponding state equation for the battery cell, and discretize and linearize the state equation to obtain a form consisting of voltage and current data matrices and a system parameter matrix; Step 3: Use the collected battery voltage and current data to identify the system parameter matrix and parse out the ohmic internal resistance R0 and polarization internal resistance R represented by the system parameters. D1 and R D2 And the polarization capacitance C D1 and C D2 ; Step 4: The state variables are composed of the power battery SOC and the loop voltages of the two RC networks. The terminal voltage is used as the observation variable. The system state equation and observation equation are established, and the SOC and SOH of the power battery are jointly estimated based on the extended Kalman filter. Step 5: Establish a power battery internal temperature field prediction model based on a gated recurrent unit (GRU) neural network, and perform a hyperparameter search for the GRU neural network using a Bayesian optimization algorithm. The power battery internal temperature field prediction model uses the combined estimation results of the battery voltage, current, SOC, and SOH, as well as the battery's external temperature, as input, and outputs the temperature field distribution inside the battery. Step 6: Using the battery surface temperature data collected by the distributed fiber optic temperature sensor system, the power battery voltage and current data, and the joint estimation results of SOC and SOH, the established power battery internal temperature field prediction model is trained, and the battery internal temperature collected by the distributed fiber optic temperature sensor is used to verify the training accuracy of the model; Step seven: Apply the trained power battery internal temperature field prediction model online so that it can directly output the internal temperature field distribution results based on the power battery voltage, current, SOC and SOH and external temperature data.

2. The method according to claim 1, wherein: The process of non-destructive optical fiber implantation in the battery cell in step 1 includes: (1) Drilling the negative electrode: Use a drill bit with a diameter of 0.9 mm to drill a hole in the center of the negative electrode of the cylindrical power battery cell. The drill bit should drill into the hole to a depth of 0.8 to 1 mm to ensure that the negative electrode shell and the electrode are drilled open. After the negative electrode is drilled, it should be sealed with Kapton glue immediately to prevent leakage of electrolyte. (2) Positive electrode cutting: Use pipe cutters to cut open the positive terminal cover of the battery cell. After cutting, check the integrity of the positive electrode sheet and use Kapton tape to insulate the positive electrode sheet from the shell. (3) Distributed fiber optic implantation: The distributed fiber optic sensor is implanted from the negative electrode hole, and the optical fiber is led out from the positive electrode and attached to the shells on both sides of the battery cell and fixed with Kapton glue, and then the positive and negative electrodes of the battery cell are resealed with epoxy resin glue; (4) Battery cell performance test and temperature perception performance test: After the epoxy resin glue solidifies, measure the open circuit voltage of the battery cell. After confirming that it is correct, place the battery cell on the fixture and carry out charge and discharge tests and internal and external temperature perception performance tests of the battery cell in the temperature chamber; The above steps (1) to (3) are all carried out under an environment protected by an inert gas.

3. The method according to claim 1, wherein: In step 2, the following state equation is established for the cylindrical power battery cell: U t (t)=U OC (t)-i L (t)R0(t)-U D1 (t)-U D2 (t) Where U OC is the open circuit voltage, U t is the terminal voltage, U D1 and U D2 are the loop voltages of the two RC networks, i L is the load current, R0 is the battery internal resistance in ohms, R D1 and R D2 is the polarization internal resistance of the two RC networks, C D1 and C D2 are the polarized capacitances of the two RC networks, t represents time, and the superscript · represents the derivative of the corresponding parameter; After discretization and linearization, we get: U t,k =Φ 2,k i 2,k Φ 2,k =[1,U t,k-1 ,IN t,k-2 ,and L,k ,and L,k-1 ,and L,k-2 ] i 2,k =[(1-θ1-θ2)U OC,k ,θ1,θ2,θ3,θ4,θ5] T Among them, Φ 2,k is the system voltage and current data matrix, θ 2,k is the system parameter matrix, and k represents the sampling time.

4. The method according to claim 3, wherein: In step 3, the system parameter matrix θ is identified based on the recursive least squares method with forgetting factor. 2,k : Where K k is the gain matrix, P k is the covariance error matrix, I is the identity matrix, and μ is the forgetting factor; In the identification, the total voltage and total current of the power battery test data are specifically input in a matrix format; After identification, the ohmic internal resistance R0 and polarization internal resistance R D1 and R D2 And the polarization capacitance C D1 and C D2 They are:

5. The method according to claim 4, wherein: The system state equation x established in step 4 k and the observation equation U t,k The specific form is as follows: x k =A k x k-1 +B k I k +ω k U t,k =C k x k -I k R0+D k-1 +v k The process of performing joint estimation of the SOC and SOH of the power battery based on the extended Kalman filter includes: (1) Initialize the system state: Set the initial state, the state estimation error covariance matrix to P0, the system noise covariance matrix to Q, and the measurement noise covariance matrix to R, where Q and R are constants; (2) System state prediction: predict the system state x at the next moment k And update the error covariance matrix P k : (3) State update: Use the measurement data to correct the state estimate and covariance estimate, and calculate the Kalman gain matrix K in sequence k , correct the system state x k And update the covariance matrix P k : (4) System state recursion: Let And start the next iteration; The EKF is used to quickly recursively obtain the SOC result during the battery cycle. For adjacent sampling points, the changes in battery capacity and SOC are counted to obtain the actual maximum available capacity of the battery and estimate the battery SOH: After one charge and discharge cycle, the average value of the maximum available capacity of the battery in the entire cycle is calculated, and the capacity parameters of the system state equation in the EKF method are updated in real time.

6. The method according to claim 1, wherein: The process of searching for hyperparameters of the GRU neural network through Bayesian optimization in step 5 includes: (1) Initialization exploration: Randomly sample the initial observation points in the set hyperparameter space to obtain the observation points x1,…,x t , and then obtain the GRU model hyperparameter objective function observations F(x1),…,F(x t ); (2) Surrogate function fitting: Based on the existing observation points and observation values, solve the distribution parameter Θ of the surrogate function and calculate the 95% confidence interval of all unexplored spaces; (3) Exploration point selection: According to the upper bound of the confidence interval of the objective function, the exploration utility function of the hyperparameter objective function is obtained, and then the maximum point of the exploration utility function is calculated, which is the next exploration point x * ; (4) Iterative exploration: Obtain the hyperparameter objective function observation value F(x * ), repeat the above steps (2) and (3) until the target accuracy requirement or the search number limit is reached, and finally output the optimal hyperparameter combination obtained by the search.

Images