Energy storage battery SOH estimation method based on electrochemical model and machine learning

Abstract

The invention discloses an energy storage battery SOH estimation method based on an electrochemical model and machine learning. The method comprises the following steps: S1, constructing an electrochemical model of an energy storage battery; s2, performing parameter identification on the electrochemical model under the initial aging cycle; s3, extracting model features related to battery aging based on the electrochemical model after parameter identification; s4, extracting data characteristics related to battery aging based on the experimental data; s5, performing correlation analysis, feature dimension reduction and feature fusion on the model features and the data features to obtain fused features; and S6, inputting the fusion features into a trained support vector machine model, and outputting a corresponding battery health state prediction value. According to the method, the characteristics related to battery aging are extracted from the electrochemical model and the experimental data, dimension reduction processing is carried out on the characteristics through the principal component analysis method to obtain the fusion characteristics, and the support vector machine is utilized to establish the battery aging model to realize SOH estimation, so that the accuracy and generalization of SOH estimation of the energy storage battery are effectively improved.

Description

Technical Field

[0001] The present invention relates to the fields of Internet big data and battery management, and in particular to a method for estimating the SOH of an energy storage battery based on an electrochemical model and machine learning. Background Art

[0002] With the intensification of global pollution and the depletion of fossil energy, energy conservation and environmental protection have become the common development goals of every country. Among them, the top priority of achieving the "dual carbon" goal is to use clean energy as the main energy source of the power system. New energy vehicles play an important role in the adjustment of the national energy structure. As a clean energy storage technology, lithium-ion batteries have also become one of the most promising candidate batteries for electric vehicles.

[0003] Although lithium-ion batteries have the advantages of energy saving, environmental protection, no memory effect, long cycle life and high energy density, their performance in terms of capacity and power will gradually deteriorate with the increase of usage time, seriously affecting the mileage and service life of electric vehicles, and may even cause electrolyte leakage and micro short circuits, causing battery failure and thermal runaway, thus causing catastrophic accidents. Therefore, the safety of lithium batteries has become the main bottleneck restricting the development of new energy vehicles. In order to ensure the safe operation of electric vehicles, batteries must be effectively managed. Among them, the battery management system (BMS) can realize battery status monitoring and safety warning to ensure the long-term safe and reliable operation of the battery. The battery health state (SOH) is one of the most important performance indicators of BMS.

[0004] At present, the main methods for estimating SOH are model-based methods and data-driven methods. The model-based method mainly establishes equivalent circuit models and electrochemical models to realize SOH estimation. The equivalent circuit model is a model that does not consider the internal chemical composition and corresponding reactions of the battery. Since the internal resistance and capacity can effectively characterize the aging of the battery, their changes indirectly reflect the changes in the battery health state, but this method is easily affected by the working conditions and the convergence of the estimation algorithm. The electrochemical model can simulate the electrochemical reaction process of the battery very well, but it has the problems of difficulty in identifying model parameters, complex calculations caused by a large number of partial differential equations, and poor generalization. The method based on the data-driven model does not need to consider the electrochemical reaction behavior and failure mechanism characteristics inside the battery, but only needs to analyze a large amount of data to accurately estimate the health state of the battery. However, this method lacks obvious physical meaning, depends on the quality of the data, has a large amount of calculation, and the current research is mostly focused on constant current charging and discharging conditions, while the user's discharge behavior is random, resulting in poor accuracy of lithium battery SOH estimation. Therefore, how to design a method that can effectively improve the accuracy and generalization of energy storage battery SOH estimation is a technical problem that needs to be solved urgently. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, the technical problem to be solved by the present invention is: how to provide a SOH estimation method for energy storage batteries based on electrochemical models and machine learning. First, features related to battery aging are extracted from the electrochemical model and experimental data. Secondly, the features are reduced in dimension by principal component analysis to obtain fused features. Finally, a battery aging model is established using a support vector machine to realize SOH estimation, thereby effectively improving the accuracy and generalization of SOH estimation of energy storage batteries (lithium batteries).

[0006] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0007] A method for estimating SOH of energy storage batteries based on electrochemical model and machine learning, comprising:

[0008] S1: Construct an electrochemical model of energy storage batteries;

[0009] S2: Parameter identification of the electrochemical model under the initial aging cycle;

[0010] S3: Extract model features related to battery aging based on the electrochemical model after parameter identification;

[0011] S4: Extracting data features related to battery aging based on the acquired experimental data;

[0012] S5: Perform correlation analysis, feature dimension reduction and feature fusion on model features and data features to obtain fusion features;

[0013] S6: Input the fused features into the trained support vector machine model and output the corresponding battery health status prediction value.

[0014] Preferably, in step S1, the electrochemical model of the energy storage battery is constructed by the following steps:

[0015] S101: Calculate the relationship between the lithium ion flow rate generated by the electrochemical reaction on the surface of the negative electrode and positive electrode active particles of the energy storage battery and the external current:

[0016]

[0017] Where: j r,n It represents the relationship between the lithium ion flow generated by the electrochemical reaction on the surface of the negative electrode active particles of the energy storage battery and the external current; r,p It represents the relationship between the lithium ion flow generated by the electrochemical reaction on the surface of the positive active particles of the energy storage battery and the external current; i represents the external current; L n Indicates the thickness of the negative electrode; A n Represents the area of ​​the negative electrode plate; an represents the specific surface area of ​​the negative electrode active particles; F represents the Faraday constant; L p Indicates the thickness of the positive electrode; A p Represents the area of ​​the positive electrode plate; a p Indicates the specific surface area of ​​the positive electrode active particles;

[0018] S102: The second-order Runge-Kutta method is used to obtain the surface lithium ion concentration of the positive and negative electrodes:

[0019]

[0020] Where: c surf,p and c surf,n Represents the lithium ion concentration on the particle surface of the positive electrode and the negative electrode respectively; D s,p and D s,n represents the solid phase diffusion coefficient of positive and negative electrodes;

[0021] S103: Calculating the electrode utilization of the positive electrode and the negative electrode based on the lithium ion concentration on the particle surface, and then calculating the open circuit voltage;

[0022] The calculation formula of electrode utilization is expressed as:

[0023]

[0024] Where: θ p ,θ n Represents the electrode utilization rate of the positive electrode and the negative electrode respectively; c smax,p and c smax,n Represent the maximum solid phase lithium ion concentration of the positive electrode and the negative electrode respectively;

[0025] The formula for the positive open circuit voltage is:

[0026] U p (θ p )=4.65-0.2076*tanh((θ p -0.4) / 0.06004)-0.06572*tanh((θ p -0.552) / 0.04231)

[0027] -0.1478*tanh((θ p -0.728) / 0.09524)+0.012814*tanh((θ p -0.4445) / 0.01732)

[0028] +0.006405*tanh((θ p -0.56) / 0.02347)-0.0728*tanh((θ p-0.90) / 0.06714)

[0029] -0.341*exp(223.8*(θ p -0.999))+0.004366*tanh((θ p -0.803) / 0.02835)

[0030] -0.005707*tanh((θ p -0.972) / 0.01034)+0.01604*exp(-((θ p -0.9927) / 0.003936) 2 )

[0031] +0.001*(-2.529*tanh((θ p -0.7) / 0.08168))-0.55

[0032] The formula for the negative electrode open circuit voltage is:

[0033] U n (θ n )=4.973-2.643*tanh((θ n -0.9832) / 0.01329)+5.621*tanh((θ n -0.6752) / 1.164)

[0034] -0.1455*tanh((θ n -0.5721) / 1.55)+4.688*tanh((θ n -0.472) / 0.2705)-0.5182

[0035] *tanh((θ n -0.4024) / 1.16)+2.935*exp(-28.9*θ n )+7.418*tanh((θ n -0.2756) / 0.204)

[0036] +14.46*tanh((θ n -0.1097) / 0.1035)-36.24*tanh((θ n -0.0624) / 0.4477)

[0037] +11.54*exp(-((θ n -0.03786) / 0.1157) 2 )

[0038] S104: The formula defining the liquid phase potential is:

[0039]

[0040] Where: φ e represents the liquid phase potential; A represents the electrode surface area; k eff Indicates effective conductivity; L p , L n and L sep Represent the thickness of positive electrode, negative electrode and separator respectively;

[0041] S105: j in step S101 r,n and j r,p Substituting into the Butler-Volmer equation we get:

[0042]

[0043] Where: k s,n and k s,p represents the electrochemical reaction rate constant of the negative electrode and the positive electrode; c e represents the concentration of lithium ions in the liquid phase, c e-s,n and c e-s,p represents the lithium ion concentration in the solid and liquid phases; α represents the transfer coefficient; R represents the gas constant; T represents the temperature constant; η p and η n denote the overpotentials of the positive and negative electrodes, respectively;

[0044] Among them, the auxiliary variable ξ p and n for:

[0045]

[0046] Calculate the overpotential generated by the positive and negative electrode reactions as:

[0047]

[0048] S106: Establish an electrochemical model to estimate the terminal voltage and SOC of the battery;

[0049] The terminal voltage calculation formula of the electrochemical model is expressed as:

[0050] U t =φ e +(U p -U n )+η p -η n +IR f ;

[0051] Where: U tIndicates terminal voltage; R f Indicates ohmic internal resistance; U p Indicates the positive open circuit voltage; U n Represents the negative electrode open circuit voltage; η p Represents the overpotential generated by the positive electrode reaction; η n Represents the overpotential generated by the negative electrode reaction; φ e represents liquid phase potential; I represents current;

[0052] The SOC of the electrochemical model is expressed by the electrode utilization of the positive electrode:

[0053]

[0054] Where: SOC(t) represents the SOC value; θ p0％ ,θ p100％ Respectively represent the surface electrode utilization rate of spherical particles when the battery is fully discharged and fully charged; θ p Indicates the electrode utilization rate of the positive electrode.

[0055] Preferably, in step S2, the electrochemical model is parameter identified under the initial aging cycle by using a particle swarm optimization algorithm;

[0056] The processing steps of the particle swarm optimization algorithm include:

[0057] S201: Determine the model parameters that need to be identified;

[0058] S202: define each particle to represent a set of model parameters to be identified;

[0059] S203: Initialize the position and velocity of particles:

[0060] X=[c smax,n ,c smax,p ,R n ,R p ,k s,n ,k s,p ] T ;

[0061] V=[V csmax,n ,V csmax,p ,V Rn ,V Rp ,V ks,n ,V ks,p ] T ;

[0062] Where: c smax,p and c smax,n Represent the maximum solid phase lithium ion concentration of the negative electrode and the positive electrode, R n and R p Represent the radius of the negative and positive electrode solid particles, ks,n , and k s,p represent the electrochemical reaction rate constants of the negative and positive electrodes, respectively, and X and V represent the position and velocity of the particles, respectively;

[0063] S204: Define the minimum value of the root mean square error of the speed as a fitness function;

[0064] The formula is:

[0065]

[0066] Where: Fit represents the fitness function, V(t) and They are represented as the simulation terminal voltage and the measurement terminal voltage respectively;

[0067] S205: Calculate the fitness value of each particle; compare the fitness of the current particle position with its historical best position, and if the current position is better, update the individual best position;

[0068] S206: Find the particle with the best fitness among all particles and update the global best position;

[0069] S207: updating the speed and position of each particle according to the inertia weight, acceleration constant, and individual and global optimal positions;

[0070] The formula is:

[0071]

[0072] Where: X i (k) represents the position vector of the particle in the kth iteration; V i (k) represents the velocity vector of the particle in the kth iteration; P i (k) represents the historical optimal position of the particle in the kth iteration; G i (k) represents the historical optimal position of the group in the kth iteration; w represents the inertia weight; c1 and c2 represent the individual learning factor and the group learning factor respectively; r1 and r2 are random numbers in the range [0,1] respectively;

[0073] S208: Repeat steps S205 to S207 until a preset number of iterations is reached or a convergence condition is met, and output the model parameters corresponding to the optimal particles as the parameter identification results of the electrochemical model.

[0074] Preferably, in step S201, by using the control variable method, under the premise of ensuring that other parameters of the electrochemical model remain unchanged, the value of a certain parameter is changed to analyze the battery voltage characteristics of the electrochemical model respectively, and several parameters with the highest parameter sensitivity are screened out as model parameters that need to be identified.

[0075] Preferably, in step S3, the model features related to battery aging are extracted through the following steps:

[0076] S301: inputting experimental data of different working conditions under different aging cycles into the electrochemical model after parameter identification;

[0077] S302: Calculate the SOC simulation value of the energy storage battery through the electrochemical model; calculate the SOC error SOC through the SOC simulation value and the actual SOC value error ; Extract SOC error SOC error The mean and integral of are used as model features F1 and model features F2;

[0078] The formula is:

[0079] F1=mean(SOC error );

[0080]

[0081] S303: Calculate the voltage simulation value of the energy storage battery through the electrochemical model; calculate the voltage error U through the voltage simulation value and the actual voltage value error ; Extract terminal voltage error U error The mean and integral of are used as model features F3 and model features F4;

[0082] The formula is:

[0083] F3=mean(U error );

[0084]

[0085] Preferably, in step S4, data features related to battery aging are extracted through the following steps:

[0086] S401: Acquire experimental data;

[0087] S402: extracting the discharge time from the experimental data as the data feature F5;

[0088] The formula is:

[0089] F5=t end -t0;

[0090] Where: t end Indicates the end time of discharge; t0 indicates the start time of discharge;

[0091] S403: extracting net discharge electric energy from the experimental data as data feature F6;

[0092] The formula is:

[0093] F6=∫U d I d dt d -∫U c I c dt c ;

[0094] Where: U d and U c Respectively represent the voltage corresponding to the discharge current and the charging current; I d and I c Represent the discharge current and charge current respectively; t d and t c Represent the discharge time and charge time respectively;

[0095] S404: extracting the voltage change rate from the experimental data as data feature F7;

[0096] The formula is:

[0097]

[0098] Where: U end It represents the voltage at the end of discharge; U0 represents the voltage at the beginning of discharge; Δt represents the time from the beginning of discharge to the end of discharge.

[0099] Preferably, in step S5, the fusion feature is generated by the following steps:

[0100] S501: putting all model features and data features into a data set;

[0101] S502: Normalize all features in the data set;

[0102] The formula is:

[0103]

[0104] Where: x normalized represents the normalized features; x represents the original features; x max and x min Respectively represent the maximum and minimum values ​​of the features in the data set;

[0105] S503: Perform correlation analysis on the normalized features in the data set using the Pearson correlation coefficient;

[0106] The formula for correlation analysis is:

[0107]

[0108] Where: X,Yrepresents the Pearson correlation coefficient; cov(X,Y) represents the covariance matrix of two continuous variables X and Y, σ X and σ Y represent the standard deviation of X and Y respectively; E represents the expectation; μ X and μ Y are the means of X and Y respectively;

[0109] S504: performing principal component analysis dimensionality reduction on several target features with the highest correlation coefficients, and selecting the feature with the largest contribution rate as the fusion feature after dimensionality reduction;

[0110] The processing steps of principal component analysis dimensionality reduction include:

[0111] S5041: Centralize the target features;

[0112] The formula is:

[0113]

[0114] Where: represents the input features X1,X2,X3,...,X n The mean of y1, y2, …y n Represents the target features after centralization;

[0115] S5042: Calculate the covariance matrix based on the target features after centralization;

[0116] The formula is:

[0117]

[0118] Where: cov(y1,y2) represents the calculated covariance;

[0119] S5043: Solving the eigenvalues ​​and eigenvectors of the covariance matrix by eigenvalue decomposition;

[0120] S5044: Arrange the eigenvectors into a matrix from top to bottom according to the corresponding eigenvalues, and take the first k columns to form the principal component V k ;

[0121] S5045: Multiply the original data set by the principal component V k , project the features in the data set into the low-dimensional space to obtain the reduced-dimensional data, which is the contribution rate of the features in the principal component space;

[0122] The formula is:

[0123] W k =X·V k ;

[0124] Where: W k represents the data after dimensionality reduction; X represents the original data set;

[0125] S5046: Select the feature with the highest contribution rate as the fusion feature after dimensionality reduction.

[0126] Preferably, in step S6, the support vector machine model is a support vector regression model.

[0127] Preferably, in step S6, the support vector regression model is trained by using a k-fold cross validation algorithm, and the training step includes:

[0128] S601: Obtain a data set including input features and target variables; divide the data set into k non-overlapping subsets;

[0129] S602: For each of the k subsets, perform the following steps:

[0130] S6021: Use the current subset as the test set and the other k-1 subsets as the training set;

[0131] S6022: Train a support vector regression model on the training set;

[0132] S6023: Evaluate the performance of the support vector regression model on the test set and record the performance indicators;

[0133] S6024: loop through S6021 to S6023, average the performance indicators obtained from k iterations, and obtain an overall evaluation of the support vector regression model performance;

[0134] S603: According to the result of k-fold cross validation, compare the average performance indicators under different parameter combinations; select the parameter combination with the best performance indicator as the optimal parameter of the support vector regression model;

[0135] S604: The support vector regression model with the optimal parameters is determined by training the entire data set until the model converges.

[0136] Preferably, in step S6023, the performance of the support vector regression model is evaluated by root mean square error and mean absolute error;

[0137] The formula is:

[0138]

[0139] Where: N represents the number of samples; represents the battery health status prediction value output by the support vector regression model; y m The real value representing the battery health status.

[0140] Compared with the prior art, the energy storage battery SOH estimation method based on electrochemical model and machine learning in the present invention has the following beneficial effects:

[0141] First, the present invention constructs an electrochemical model of an energy storage battery and performs parameter identification under an initial aging cycle to accurately describe the internal electrochemical behavior of the battery. The dynamic operating condition data under different aging cycles are input into the electrochemical model to obtain the voltage error and SOC error under different aging cycles, so that the model features closely related to battery aging can be obtained based on the model error; at the same time, data features are extracted from the experimental data, and the extracted model features and data features are subjected to correlation analysis, feature dimension reduction and feature fusion together, and the feature set used for SOH estimation is further screened and optimized, thereby improving the accuracy of SOH estimation of energy storage batteries (lithium batteries), making the prediction results closer to the actual health state of the battery, and better meeting the battery management system's demand for accurate SOH estimation. Then, the present invention reduces the number of features and the computational complexity through feature dimension reduction, and helps to remove redundant information and improve the generalization ability of the model. At the same time, the present invention uses a support vector machine (support vector regression, SVR) model to realize SOH estimation, and uses the advantages of SVR in dealing with nonlinear problems and avoiding overfitting to enhance the generalization performance of the model, so that the model can maintain stable prediction performance under different working conditions and different battery types, thereby improving the practicality and reliability of SOH estimation of energy storage batteries. Secondly, the present invention can achieve fast and real-time SOH estimation while ensuring accuracy through the application of feature fusion and SVR model. The SVR model has high computational efficiency when processing large data sets, while feature fusion reduces the number of features and reduces computational complexity. Finally, in order to address the problems of long electrochemical model parameter identification time and complex calculation process, the present invention only needs to identify the parameters of the electrochemical model under the initial cycle, without updating the electrochemical parameters under each cycle, and uses the electrochemical model with actual physical meaning to extract model features, thereby improving the accuracy and computational efficiency of energy storage battery SOH estimation. BRIEF DESCRIPTION OF THE DRAWINGS

[0142] In order to make the purpose, technical solution and advantages of the invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings, in which:

[0143] Figure 1 and Figure 2 They are the logic block diagram and principle diagram of the energy storage battery SOH estimation method respectively.

[0144] Figure 3 Schematic diagram of the electrochemical model.

[0145] Figure 4 Flowchart of parameter identification using particle swarm algorithm.

[0146] Figure 5 and Figure 6 Schematic diagrams of voltage error and SOC error respectively.

[0147] Figure 7 Flowchart for generating fusion features.

[0148] Figure 8 Schematic diagram of SOH estimation for the support vector regression model. DETAILED DESCRIPTION

[0149] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but only represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the scope of protection of the present invention.

[0150] The following is a further detailed description through specific implementation methods:

[0151] Example:

[0152] This embodiment discloses a method for estimating SOH of an energy storage battery based on an electrochemical model and machine learning.

[0153] like Figure 1 and Figure 2 As shown in the figure, the energy storage battery SOH estimation method based on electrochemical model and machine learning includes:

[0154] S1: Construct an electrochemical model of energy storage battery (lithium battery);

[0155] S2: Parameter identification of the electrochemical model under the initial aging cycle;

[0156] S3: Extract model features related to battery aging based on the electrochemical model after parameter identification;

[0157] S4: Extracting data features related to battery aging based on the acquired experimental data;

[0158] S5: Perform correlation analysis, feature dimension reduction and feature fusion on model features and data features to obtain fusion features;

[0159] S6: Input the fused features into the trained support vector machine model and output the corresponding battery health status prediction value.

[0160] First, the present invention constructs an electrochemical model of an energy storage battery and performs parameter identification under an initial aging cycle to accurately describe the internal electrochemical behavior of the battery. The dynamic operating condition data under different aging cycles are input into the electrochemical model to obtain the voltage error and SOC error under different aging cycles, so that the model features closely related to battery aging can be obtained based on the model error; at the same time, data features are extracted from the experimental data, and the extracted model features and data features are subjected to correlation analysis, feature dimension reduction and feature fusion together, and the feature set used for SOH estimation is further screened and optimized, thereby improving the accuracy of SOH estimation of energy storage batteries (lithium batteries), making the prediction results closer to the actual health state of the battery, and better meeting the battery management system's demand for accurate SOH estimation. Then, the present invention reduces the number of features and the computational complexity through feature dimension reduction, and helps to remove redundant information and improve the generalization ability of the model. At the same time, the present invention uses a support vector machine (support vector regression, SVR) model to realize SOH estimation, and uses the advantages of SVR in dealing with nonlinear problems and avoiding overfitting to enhance the generalization performance of the model, so that the model can maintain stable prediction performance under different working conditions and different battery types, thereby improving the practicality and reliability of SOH estimation of energy storage batteries. Secondly, the present invention can achieve fast and real-time SOH estimation while ensuring accuracy through the application of feature fusion and SVR model. The SVR model has high computational efficiency when processing large data sets, while feature fusion reduces the number of features and reduces computational complexity. Finally, in order to address the problems of long electrochemical model parameter identification time and complex calculation process, the present invention only needs to identify the parameters of the electrochemical model under the initial cycle, without updating the electrochemical parameters under each cycle, and uses the electrochemical model with actual physical meaning to extract model features, thereby improving the accuracy and computational efficiency of energy storage battery SOH estimation.

[0161] In order to better introduce the technical solution of the present invention, this embodiment is described through the following parts.

[0162] 1. Electrochemical Model

[0163] In the specific implementation process, Figure 3 As shown, the electrochemical model of the energy storage battery is constructed through the following steps:

[0164] S101: Ignore the diffusion of solid lithium ion concentration along the electrode, assume that the solid spherical particles distributed along the x direction in each electrode are indifferent, and the solid phase diffusion of lithium ions occurs in a representative spherical particle. For a position with coordinate x' in the negative electrode region, we have:

[0165]

[0166] Where: i e (x') represents the liquid phase current density at x'; a n represents the specific surface area of ​​the negative electrode active particles; j r,n It is the lithium ion flow generated by the electrochemical reaction on the surface of the negative electrode active particles; F represents the Faraday constant;

[0167] From the boundary conditions of the negative electrode liquid phase current, we know that:

[0168]

[0169] Therefore, the relationship between the lithium ion flow rate generated by the electrochemical reaction on the surface of the negative electrode active particles of the energy storage battery and the external current is calculated as follows:

[0170]

[0171] Where: i represents the external current; L n Indicates the thickness of the negative electrode; A n Represents the area of ​​the negative electrode plate; a n represents the specific surface area of ​​the negative electrode active particles; F represents the Faraday constant;

[0172] Similarly, for a position with coordinate x' in the positive region, we have:

[0173]

[0174] Where: a p represents the specific surface area of ​​the positive electrode active particles; j r,p It is the lithium ion flow generated by the electrochemical reaction on the surface of the positive electrode active particles;

[0175] From the boundary conditions of the positive electrode liquid phase current, we know that:

[0176]

[0177] Therefore, the relationship between the lithium ion flow generated by the electrochemical reaction on the surface of the positive electrode active particles of the energy storage battery and the external current is calculated as follows:

[0178]

[0179] Where: i represents the external current; L p Indicates the thickness of the positive electrode; A p Represents the area of ​​the positive electrode plate; a p represents the specific surface area of ​​the positive electrode active particles; F represents the Faraday constant;

[0180] S102: The pseudo-two-dimensional model is the most classic electrochemical model. It contains 4 partial differential equations, 1 algebraic equation, and 1 terminal voltage calculation equation. The equations are severely coupled, and the partial differential equations are difficult to calculate. Therefore, the pseudo-two-dimensional model needs to be simplified. The solid phase diffusion equation is expressed by a complex partial differential equation, so the finite difference method is used to divide the active particle radius into m segments, and the partial differential equation is converted into an ordinary differential equation for calculation to obtain the state space expression of the system;

[0181] The formula is:

[0182]

[0183] Where: R represents the radius of the solid phase particle; m is 10; D s is the solid phase diffusion coefficient; The state vector representing the lithium ion concentration in the solid particles;

[0184] Taking the negative electrode solid phase particles as an example, the equation of state of the system is expressed as:

[0185]

[0186] In short:

[0187]

[0188] Among them, C n is the state vector of lithium ion concentration of active particles, A n and B n are the corresponding coefficient matrices in the state space expression respectively.

[0189] The second-order Runge-Kutta method is used to solve the high-order system and obtain the lithium ion concentration on the particle surface of the positive and negative electrodes:

[0190]

[0191] Where: c surf,p and c surf,n Represents the lithium ion concentration on the particle surface of the positive electrode and the negative electrode respectively; D s,p and D s,n represents the solid phase diffusion coefficient of positive and negative electrodes;

[0192] S103: Calculating the electrode utilization of the positive electrode and the negative electrode based on the lithium ion concentration on the particle surface, and then calculating the open circuit voltage;

[0193] The calculation formula of electrode utilization is expressed as:

[0194]

[0195] Where: θp ,θ n Represents the electrode utilization rate of the positive electrode and the negative electrode respectively; c smax,p and c smax,n Represent the maximum solid phase lithium ion concentration of the positive electrode and the negative electrode respectively;

[0196] The formula for the positive open circuit voltage is:

[0197] U p (θ p )=4.65-0.2076*tanh((θ p -0.4) / 0.06004)-0.06572*tanh((θ p -0.552) / 0.04231)

[0198] -0.1478*tanh((θ p -0.728) / 0.09524)+0.012814*tanh((θ p -0.4445) / 0.01732)

[0199] +0.006405*tanh((θ p -0.56) / 0.02347)-0.0728*tanh((θ p -0.90) / 0.06714)

[0200] -0.341*exp(223.8*(θ p -0.999))+0.004366*tanh((θ p -0.803) / 0.02835)

[0201] -0.005707*tanh((θ p -0.972) / 0.01034)+0.01604*exp(-((θ p -0.9927) / 0.003936) 2 )

[0202] +0.001*(-2.529*tanh((θ p -0.7) / 0.08168))-0.55

[0203] The formula for the negative electrode open circuit voltage is:

[0204] U n (θ n )=4.973-2.643*tanh((θ n -0.9832) / 0.01329)+5.621*tanh((θn -0.6752) / 1.164)

[0205] -0.1455*tanh((θ n -0.5721) / 1.55)+4.688*tanh((θ n -0.472) / 0.2705)-0.5182

[0206] *tanh((θ n -0.4024) / 1.16)+2.935*exp(-28.9*θ n )+7.418*tanh((θ n -0.2756) / 0.204)

[0207] +14.46*tanh((θ n -0.1097) / 0.1035)-36.24*tanh((θ n -0.0624) / 0.4477)

[0208] +11.54*exp(-((θ n -0.03786) / 0.1157) 2 )

[0209] S104: The formula defining the liquid phase potential is:

[0210]

[0211] Where: φ e represents the liquid phase potential; A represents the electrode surface area; k eff Indicates effective conductivity; L p , L n and L sep Represent the thickness of positive electrode, negative electrode and separator respectively;

[0212] S105: j in step S101 r,n and j r,p Substituting into the Butler-Volmer equation we get:

[0213]

[0214] Where: k s,n and k s,p represents the electrochemical reaction rate constant of the negative electrode and the positive electrode; c e represents the concentration of lithium ions in the liquid phase, c e-s,n and c e-s,p represents the lithium ion concentration in the solid and liquid phases; α represents the transfer coefficient; R represents the gas constant; T represents the temperature constant; ηp and η n denote the overpotentials of the positive and negative electrodes, respectively;

[0215] Among them, the auxiliary variable ξ p and n for:

[0216]

[0217] Calculate the overpotential generated by the positive and negative electrode reactions as:

[0218]

[0219] S106: Finally, an electrochemical model is established to estimate the terminal voltage and SOC of the battery;

[0220] The terminal voltage calculation formula of the electrochemical model is expressed as:

[0221] U t =φ e +(U p -U n )+η p -η n +IR f ;

[0222] Where: U t Indicates terminal voltage; R f Indicates ohmic internal resistance; U p Indicates the positive open circuit voltage; U n Represents the negative electrode open circuit voltage; η p Represents the overpotential generated by the positive electrode reaction; η n Represents the overpotential generated by the negative electrode reaction; φ e represents liquid phase potential; I represents current;

[0223] The SOC of the electrochemical model is expressed by the electrode utilization of the positive electrode:

[0224]

[0225] Where: SOC(t) represents the SOC value; θ p0％ ,θ p100％ Respectively represent the surface electrode utilization rate of spherical particles when the battery is fully discharged and fully charged; θ p Indicates the electrode utilization rate of the positive electrode.

[0226] The present invention extracts model features based on the electrochemical model and simplifies the electrochemical model through the finite difference method. Compared with the equivalent circuit model and the data-driven method, the electrochemical model can describe the electrochemical reaction process inside the battery, can more accurately describe the aging state of the battery, and improve the accuracy of SOH estimation. However, the traditional pseudo-two-dimensional model contains four partial differential equations, an algebraic equation and an equation for calculating the terminal voltage. The equations are severely coupled, the partial differential equations are difficult to calculate, the electrochemical parameters are as many as more than 30, and they are difficult to obtain, and are also affected by temperature and aging. Therefore, the present invention simplifies the electrochemical model.

[0227] 2. Parameter Identification

[0228] In the specific implementation process, the particle swarm optimization algorithm is used to identify the parameters of the electrochemical model under the initial aging cycle.

[0229] Combination Figure 4 As shown in Figure 2, the processing steps of the particle swarm optimization algorithm include:

[0230] S201: Determine the model parameters that need to be identified;

[0231] In this embodiment, by using the control variable method, under the premise of ensuring that other parameters of the electrochemical model remain unchanged, the value of a certain parameter is changed to analyze the battery voltage characteristics of the electrochemical model respectively, and several parameters with the highest parameter sensitivity are screened out as model parameters that need to be identified.

[0232] S202: define each particle to represent a set of model parameters to be identified;

[0233] S203: Initialize the position and velocity of particles:

[0234] X=[c smax,n ,c smax,p ,R n ,R p ,rk n ,rk p ] T ;

[0235] V=[V csmax,n ,V csmax,p ,V Rn ,V Rp ,V rkn ,V rkp ] T ;

[0236] Where: c smax,p and c smax,n Represent the maximum solid phase lithium ion concentration of the negative electrode and the positive electrode, R n and R pRepresent the radius of the negative and positive electrode solid particles, k s,n , and k s,p represent the electrochemical reaction rate constants of the negative and positive electrodes, respectively, and X and V represent the position and velocity of the particles, respectively;

[0237] S204: Define the minimum value of the root mean square error of the speed as a fitness function;

[0238] The formula is:

[0239]

[0240] Where: Fit represents the fitness function, V(t) and They are represented as the simulation terminal voltage and the measurement terminal voltage respectively;

[0241] In this embodiment, necessary parameters of the particle swarm optimization algorithm may be set as needed, including population size, number of iterations, inertia weight, acceleration constant, boundary conditions, etc. The parameter settings of this part are completed with reference to existing means.

[0242] S205: Calculate the fitness value of each particle; compare the fitness of the current particle position with its historical best position, and if the current position is better, update the individual best position;

[0243] S206: Find the particle with the best fitness among all particles and update the global best position;

[0244] S207: updating the speed and position of each particle according to the inertia weight, acceleration constant, and individual and global optimal positions;

[0245] The formula is:

[0246]

[0247] Where: X i (k) represents the position vector of the particle in the kth iteration; V i (k) represents the velocity vector of the particle in the kth iteration; P i (k) represents the historical optimal position of the particle in the kth iteration; G i (k) represents the historical optimal position of the group in the kth iteration; w represents the inertia weight; c1 and c2 represent the individual learning factor and the group learning factor respectively; r1 and r2 are random numbers in the range [0,1] to increase the randomness of the search;

[0248] S208: Repeat steps S205 to S207 until a preset number of iterations is reached or a convergence condition is met, and output the model parameters corresponding to the optimal particles as the parameter identification results of the electrochemical model.

[0249] In this embodiment, the final parameter identification results are shown in Table 1.

[0250] Table 1 High sensitivity parameter identification results at initial aging point

[0251]

[0252] The present invention only performs parameter identification on the model parameters under the initial aging cycle, and does not need to update the model parameters under each cycle, thereby reducing the amount of calculation and improving the calculation efficiency of the energy storage battery SOH estimation.

[0253] 3. Model feature extraction

[0254] In the specific implementation process, the experimental data of different working conditions under different aging cycles are input into the electrochemical model. Since the model parameters are not updated, there will be errors in the SOC and terminal voltage output by the model. As the battery ages, the model error will gradually increase. Taking the FUDS working condition as an example, Figure 5 and Figure 6 shown.

[0255] Specifically, the model features related to battery aging are extracted through the following steps:

[0256] S301: inputting experimental data of different working conditions under different aging cycles into the electrochemical model after parameter identification;

[0257] S302: Calculate the SOC simulation value of the energy storage battery through the electrochemical model; calculate the SOC error SOC through the SOC simulation value and the actual SOC value error ; Extract SOC error SOC error The mean and integral of are used as model features F1 and model features F2;

[0258] The formula is:

[0259] F1=mean(SOC error );

[0260]

[0261] S303: Calculate the voltage simulation value of the energy storage battery through the electrochemical model; calculate the voltage error U through the voltage simulation value and the actual voltage value error ; Extract terminal voltage error U error The mean and integral of are used as model features F3 and model features F4;

[0262] The formula is:

[0263] F3=mean(U error );

[0264]

[0265] 4. Data Feature Extraction

[0266] In the specific implementation process, the data features related to battery aging are extracted through the following steps:

[0267] S401: Acquire experimental data;

[0268] In this embodiment, the process of obtaining experimental data includes: conducting a battery aging test on a 1.1Ah lithium iron phosphate battery at 25°C, including a capacity test, an OCV test, and different dynamic working condition test experiments, wherein the dynamic working conditions include DST working condition, FUDS working condition, UDDS working condition, BJDST working condition, and US06 working condition. An accelerated aging experiment is conducted at 15°C, and the test ends when the failure threshold is reached.

[0269] S402: extracting the discharge time from the experimental data as the data feature F5;

[0270] The formula is:

[0271] F5=t end -t0;

[0272] Where: t end Indicates the end time of discharge; t0 indicates the start time of discharge;

[0273] S403: extracting net discharge electric energy from the experimental data as data feature F6;

[0274] The formula is:

[0275] F6=∫U d I d dt d -∫U c I c dt c

[0276] Where: U d and U c Respectively represent the voltage corresponding to the discharge current and the charging current; I d and I c Represent the discharge current and charge current respectively; t d and t c Represent the discharge time and charge time respectively;

[0277] S404: extracting the voltage change rate from the experimental data as data feature F7;

[0278] The formula is:

[0279]

[0280] Where: U end It represents the voltage at the end of discharge; U0 represents the voltage at the beginning of discharge; Δt represents the time from the beginning of discharge to the end of discharge.

[0281] In addition to extracting model features that can reflect the internal structure of the battery, the present invention also extracts data features that can reflect the external structure of the battery based on experimental data. Compared with traditional data-driven methods, the method of the present invention simultaneously extracts features that can reflect the internal and external features of the battery, making the information obtained by the machine learning model more comprehensive and complete, improving the model's sensitivity to changes in battery performance, and enhancing prediction accuracy and robustness.

[0282] 5. Fusion Feature Generation

[0283] In the specific implementation process, the features extracted based on the model and data have a strong correlation with battery aging, but the differences between the features under different working conditions are large. Figure 7 As shown, the fusion features are generated through the following steps:

[0284] S501: putting all model features and data features into a data set;

[0285] S502: The magnitude of features under different working conditions is different, so all features in the data set need to be normalized;

[0286] The formula is:

[0287]

[0288] Where: x normalized represents the normalized features; x represents the original features; x max and x min Respectively represent the maximum and minimum values ​​of the features in the data set;

[0289] S503: Perform correlation analysis on the normalized features in the data set using the Pearson correlation coefficient;

[0290] The formula for correlation analysis is:

[0291]

[0292] Where: X,Y represents the Pearson correlation coefficient; cov(X,Y) represents the covariance matrix of two continuous variables X and Y, σ X and σ Y represent the standard deviation of X and Y respectively; E represents the expectation; μ X and μ Y are the means of X and Y respectively;

[0293] S504: performing principal component analysis (PCA) dimensionality reduction on several (five) target features with the highest correlation coefficients, and selecting the first principal component with the largest contribution rate as the fusion feature after dimensionality reduction;

[0294] The processing steps of principal component analysis dimensionality reduction include:

[0295] S5041: Centralize the target features;

[0296] The formula is:

[0297]

[0298] Where: represents the input features X1,X2,X3,...,X n The mean of y1, y2, …y n Represents the target features after centralization;

[0299] S5042: Calculate the covariance matrix based on the target features after centralization;

[0300] The formula is:

[0301]

[0302] Where: cov(y1,y2) represents the calculated covariance;

[0303] S5043: Solving the eigenvalues ​​and eigenvectors of the covariance matrix by eigenvalue decomposition;

[0304] S5044: Arrange the eigenvectors into a matrix from top to bottom according to the corresponding eigenvalues, and take the first k columns to form the principal component V k ;

[0305] S5045: Multiply the original data set by the principal component V k , project the features in the data set into the low-dimensional space to obtain the reduced-dimensional data, which is the contribution rate of the features in the principal component space;

[0306] The formula is:

[0307] W k =X·V k ;

[0308] Where: W k represents the features after dimensionality reduction; X represents the original data set;

[0309] S5046: Select the feature with the highest contribution rate as the fusion feature after dimensionality reduction.

[0310] 6. Support Vector Regression Model

[0311] In this embodiment, the support vector machine model selects a support vector regression model (SVR).

[0312] The working logic of the support vector regression model is expressed as:

[0313] 1) Map the original data set into a high-dimensional feature space through a mapping function Φ(x), and perform linear regression in the feature space:

[0314] f(x)=wΦ(x)+b;

[0315] Where: w represents the hyperplane normal vector; b represents the hyperplane parameter;

[0316] 2) Define the ε linear insensitive loss function:

[0317]

[0318] Where: f(x) represents the predicted value of the battery health state; y represents the true value of the battery health state; ε is a threshold that determines the tolerance of the model to error;

[0319] 3) Based on the idea of ​​maximizing the support vector machine interval and minimizing the loss, the support vector regression model is expressed as:

[0320]

[0321] Where: C represents the penalty factor; ζ i and Respectively represent the degree to which each sample exceeds the upper and lower interval bands; m represents the number of samples;

[0322] 4) Introduce the Lagrangian function and convert it into a dual form, and the regression function is obtained as follows:

[0323]

[0324] Where: a i and Dual variables, is the offset, x i represents the eigenvector, x j The center of the kernel function, K(x i ,x j ) represents the radial basis kernel function; σ represents the width parameter of the function, which controls the radial range of the kernel function.

[0325] Unlike other complex neural network models, the present invention does not require a large data set for training, nor does it require a deep network architecture and high-performance computing resources. It can achieve efficient and accurate estimation of the battery health status using a simple machine learning algorithm.

[0326] 7. Model Training

[0327] In the specific implementation process, Figure 8 As shown, the support vector regression model is trained using the k-fold cross validation algorithm. The training steps include:

[0328] S601: Obtain a data set including input features and target variables; divide the data set into k non-overlapping subsets;

[0329] S602: For each of the k subsets, perform the following steps:

[0330] S6021: Use the current subset as the test set and the other k-1 subsets as the training set;

[0331] S6022: Train a support vector regression model on the training set;

[0332] S6023: Evaluate the performance of the support vector regression model on the test set and record the performance indicators;

[0333] S6024: loop through S6021 to S6023, average the performance indicators obtained from k iterations, and obtain an overall evaluation of the support vector regression model performance;

[0334] S603: According to the result of k-fold cross validation, the average performance indicators under different parameter combinations are compared; the parameter combination with the best performance indicator (such as the smallest MSE and the largest R2) is selected as the optimal parameter of the support vector regression model;

[0335] S604: The support vector regression model with the optimal parameters is determined by training the entire data set until the model converges.

[0336] Specifically, the performance of the support vector regression model was evaluated by the root mean square error (RMSE) and mean absolute error (MAE);

[0337] The formula is:

[0338]

[0339] Where N represents the number of samples; represents the battery health status prediction value output by the support vector regression model; y m The real value representing the battery health status.

[0340] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit the technical solution. Those skilled in the art should understand that those modifications or equivalent substitutions of the technical solution of the present invention that do not depart from the purpose and scope of the technical solution should be included in the scope of the claims of the present invention.

Claims

1. A method for estimating SOH of energy storage batteries based on electrochemical model and machine learning, characterized in that: include: S1: Construct an electrochemical model of energy storage batteries; S2: Parameter identification of the electrochemical model under the initial aging cycle; S3: Extract model features related to battery aging based on the electrochemical model after parameter identification; S4: Extracting data features related to battery aging based on the acquired experimental data; S5: Perform correlation analysis, feature dimension reduction and feature fusion on model features and data features to obtain fusion features; S6: Input the fused features into the trained support vector machine model and output the corresponding battery health status prediction value.

2. The method for estimating SOH of an energy storage battery based on an electrochemical model and machine learning as claimed in claim 1, characterized in that: In step S1, the electrochemical model of the energy storage battery is constructed by the following steps: S101: Calculate the relationship between the lithium ion flow rate generated by the electrochemical reaction on the surface of the negative electrode and positive electrode active particles of the energy storage battery and the external current: Where: j r,n It represents the relationship between the lithium ion flow generated by the electrochemical reaction on the surface of the negative electrode active particles of the energy storage battery and the external current; r,p It represents the relationship between the lithium ion flow generated by the electrochemical reaction on the surface of the positive active particles of the energy storage battery and the external current; i represents the external current; L n Indicates the thickness of the negative electrode; A n Represents the area of ​​the negative electrode plate; a n represents the specific surface area of ​​the negative electrode active particles; F represents the Faraday constant; L p Indicates the thickness of the positive electrode; A p Represents the area of ​​the positive electrode plate; a p Represents the specific surface area of ​​the positive electrode active particles; S102: The second-order Runge-Kutta method is used to obtain the surface lithium ion concentration of the positive and negative electrodes: Where: c surf,p and c surf,n Represents the lithium ion concentration on the particle surface of the positive electrode and the negative electrode respectively; D s,p and D s,n represents the solid phase diffusion coefficient of positive and negative electrodes; S103: Calculating the electrode utilization of the positive electrode and the negative electrode based on the lithium ion concentration on the particle surface, and then calculating the open circuit voltage; The calculation formula of electrode utilization is expressed as: Where: θ p ,θ n Represents the electrode utilization of the positive electrode and the negative electrode respectively; c smax,p and c smax,n Represent the maximum solid phase lithium ion concentration of the positive electrode and the negative electrode respectively; The formula for the positive open circuit voltage is: U p (θ p )=4.65-0.2076*tanh((θ p -0.4) / 0.06004)-0.06572*tanh((θ p -0.552) / 0.04231)-0.1478*tanh((θ p -0.728) / 0.09524)+0.012814*tanh((θ p -0.4445) / 0.01732)+0.006405*tanh((θ p -0.56) / 0.02347)-0.0728*tanh((θ p -0.90) / 0.06714)-0.341*exp(223.8*(θ p -0.999))+0.004366*tanh((θ p -0.803) / 0.02835)-0.005707*tanh((θ p -0.972) / 0.01034)+0.01604*exp(-((θ p -0.9927) / 0.003936) 2 )+0.001*(-2.529*tanh((θ p -0.7) / 0.08168))-0.55 The formula for the negative electrode open circuit voltage is: U n (θ n )=4.973-2.643*tanh((θ n -0.9832) / 0.01329)+5.621*tanh((θ n -0.6752) / 1.164)-0.1455*tanh((θ n -0.5721) / 1.55)+4.688*tanh((θ n -0.472) / 0.2705)-0.5182*tanh((θ n -0.4024) / 1.16)+2.935*exp(-28.9*θ n )+7.418*tanh((θ n -0.2756) / 0.204)+14.46*tanh((θ n -0.1097) / 0.1035)-36.24*tanh((θ n -0.0624) / 0.4477)+11.54*exp(-((θ n -0.03786) / 0.1157) 2 ) S104: The formula defining the liquid phase potential is: Where: φ e represents the liquid phase potential; A represents the electrode surface area; k eff Indicates effective conductivity; L p , L n and L sep Represent the thickness of positive electrode, negative electrode and separator respectively; S105: j in step S101 r,n and j r,p Substituting into the Butler-Volmer equation we get: Where: k s,n and k s,p represents the electrochemical reaction rate constant of the negative electrode and the positive electrode; c e represents the concentration of lithium ions in the liquid phase, c e-s,n and c e-s,p represents the lithium ion concentration in the solid and liquid phases; α represents the transfer coefficient; R represents the gas constant; T represents the temperature constant; η p and η n denote the overpotentials of the positive and negative electrodes, respectively; Among them, the auxiliary variable ξ p and n for: Calculate the overpotential generated by the positive and negative electrode reactions as: S106: Establish an electrochemical model to estimate the terminal voltage and SOC of the battery; The terminal voltage calculation formula of the electrochemical model is expressed as: U t =φ e +(U p -U n )+η p -η n +IR f ; Where: U t Indicates terminal voltage; R f Indicates ohmic internal resistance; U p Indicates the positive open circuit voltage; U n Represents the negative electrode open circuit voltage; η p Represents the overpotential generated by the positive electrode reaction; η n Represents the overpotential generated by the negative electrode reaction; φ e represents liquid phase potential; I represents current; The SOC of the electrochemical model is expressed by the electrode utilization of the positive electrode: Where: SOC(t) represents the SOC value; θ p0％ ,θ p100％ Respectively represent the surface electrode utilization rate of spherical particles when the battery is fully discharged and fully charged; θ p Indicates the electrode utilization rate of the positive electrode.

3. The method for estimating SOH of an energy storage battery based on an electrochemical model and machine learning as claimed in claim 1, characterized in that: In step S2, the electrochemical model is parameter identified under the initial aging cycle by using a particle swarm optimization algorithm; The processing steps of the particle swarm optimization algorithm include: S201: Determine the model parameters that need to be identified; S202: define each particle to represent a set of model parameters to be identified; S203: Initialize the position and velocity of particles: X=[c smax,n ,c smax,p ,R n ,R p ,k s,n ,k s,p ] T ; V=[V csmax,n ,V csmax,p ,V Rn ,V Rp ,V ks,n ,V ks,p ] T ; Where: c smax,p and c smax,n Represent the maximum solid phase lithium ion concentration of the negative electrode and the positive electrode, R n and R p Represent the radius of the negative and positive electrode solid particles, k s,n , and k s,p represent the electrochemical reaction rate constants of the negative and positive electrodes, respectively, and X and V represent the position and velocity of the particles, respectively; S204: Define the minimum value of the root mean square error of the speed as a fitness function; The formula is: Where: Fit represents the fitness function, V(t) and They are represented as the simulation terminal voltage and the measurement terminal voltage respectively; S205: Calculate the fitness value of each particle; compare the fitness of the current particle position with its historical best position, and if the current position is better, update the individual best position; S206: Find the particle with the best fitness among all particles and update the global best position; S207: updating the speed and position of each particle according to the inertia weight, acceleration constant, and individual and global optimal positions; The formula is: Where: X i (k) represents the position vector of the particle in the kth iteration; V i (k) represents the velocity vector of the particle in the kth iteration; P i (k) represents the historical optimal position of the particle in the kth iteration; G i (k) represents the historical optimal position of the group in the kth iteration; w represents the inertia weight; c1 and c2 represent the individual learning factor and the group learning factor respectively; r1 and r2 are random numbers in the range [0,1] respectively; S208: Repeat steps S205 to S207 until a preset number of iterations is reached or a convergence condition is met, and output the model parameters corresponding to the optimal particles as the parameter identification results of the electrochemical model.

4. The method for estimating SOH of an energy storage battery based on an electrochemical model and machine learning as claimed in claim 3, characterized in that: In step S201, by using the control variable method, under the premise of ensuring that other parameters of the electrochemical model remain unchanged, the value of a certain parameter is changed to analyze the battery voltage characteristics of the electrochemical model respectively, and several parameters with the highest parameter sensitivity are selected as model parameters that need to be identified.

5. The method for estimating SOH of an energy storage battery based on an electrochemical model and machine learning as claimed in claim 1, characterized in that: In step S3, the model features related to battery aging are extracted through the following steps: S3 01: Input the experimental data of different working conditions under different aging cycles into the electrochemical model after parameter identification; S302: Calculate the SOC simulation value of the energy storage battery through the electrochemical model; calculate the SOC error SOC through the SOC simulation value and the actual SOC value error ; Extract SOC error SOC error The mean and integral of are used as model features F1 and model features F2; The formula is: F1=mean(SOC error ); S303: Calculate the voltage simulation value of the energy storage battery through the electrochemical model; calculate the voltage error U through the voltage simulation value and the actual voltage value error ; Extract terminal voltage error U error The mean and integral of are used as model features F3 and model features F4; The formula is: F3=mean(U error ); 6. The method for estimating SOH of an energy storage battery based on an electrochemical model and machine learning as claimed in claim 1, characterized in that: In step S4, data features related to battery aging are extracted through the following steps: S401: Acquire experimental data; S402: extracting the discharge time from the experimental data as the data feature F5; The formula is: F5=t end -t0; Where: t end Indicates the discharge end time; t0 represents the discharge start time; S403: extracting net discharge electric energy from the experimental data as data feature F6; The formula is: F6=∫U d I d dt d -∫U c I c dt c ; Where: U d and U c Respectively represent the voltage corresponding to the discharge current and the charging current; I d and I c Represent the discharge current and charge current respectively; t d and t c Represent the discharge time and charge time respectively; S404: extracting the voltage change rate from the experimental data as data feature F7; The formula is: Where: U end It represents the voltage at the end of discharge; U0 represents the voltage at the beginning of discharge; Δt represents the time from the beginning of discharge to the end of discharge.

7. The method for estimating SOH of an energy storage battery based on an electrochemical model and machine learning as claimed in claim 1, characterized in that: In step S5, the fusion features are generated by the following steps: S5 01: Put all model features and data features into a dataset; S502: Normalize all features in the data set; The formula is: Where: x normalized represents the normalized features; x represents the original features; x max and x min Respectively represent the maximum and minimum values ​​of the features in the data set; S503: Perform correlation analysis on the normalized features in the data set using the Pearson correlation coefficient; The formula for correlation analysis is: Where: X,Y represents the Pearson correlation coefficient; cov(X,Y) represents the covariance matrix of two continuous variables X and Y, σ X and σ Y represent the standard deviation of X and Y respectively; E represents the expectation; μ X and μ Y are the means of X and Y respectively; S504: performing principal component analysis dimensionality reduction on several target features with the highest correlation coefficients, and selecting the feature with the largest contribution rate as the fusion feature after dimensionality reduction; The processing steps of principal component analysis dimensionality reduction include: S5041: Centralize the target features; The formula is: Where: represents the input features X1,X2,X3,...,X n The mean of y1, y2, …y n Represents the target features after centralization; S5042: Calculate the covariance matrix based on the target features after centralization; The formula is: Where: cov(y1,y2) represents the calculated covariance; S5043: Solving the eigenvalues ​​and eigenvectors of the covariance matrix by eigenvalue decomposition; S5044: Arrange the eigenvectors into a matrix from top to bottom according to the corresponding eigenvalues, and take the first k columns to form the principal component V k ; S5045: Multiply the original data set by the principal component V k , project the features in the data set into the low-dimensional space to obtain the reduced-dimensional data, which is the contribution rate of the features in the principal component space; The formula is: W k =X·V k ; Where: W k represents the data after dimensionality reduction; X represents the original data set; S5046: Select the feature with the highest contribution rate as the fusion feature after dimensionality reduction.

8. The method for estimating SOH of an energy storage battery based on an electrochemical model and machine learning as claimed in claim 1, characterized in that: In step S6, the support vector machine model is a support vector regression model.

9. The method for estimating SOH of an energy storage battery based on an electrochemical model and machine learning as claimed in claim 8, characterized in that: In step S6, the support vector regression model is trained by using a k-fold cross validation algorithm, and the training steps include: S601: Obtain a data set including input features and target variables; divide the data set into k non-overlapping subsets; S602: For each of the k subsets, perform the following steps: S6021: Use the current subset as the test set and the other k-1 subsets as the training set; S6022: Train a support vector regression model on the training set; S6023: Evaluate the performance of the support vector regression model on the test set and record the performance indicators; S6024: loop through S6021 to S6023, average the performance indicators obtained from k iterations, and obtain an overall evaluation of the support vector regression model performance; S603: According to the result of k-fold cross validation, compare the average performance indicators under different parameter combinations; select the parameter combination with the best performance indicator as the optimal parameter of the support vector regression model; S604: The support vector regression model with the optimal parameters is determined by training the entire data set until the model converges.

10. The method for estimating SOH of an energy storage battery based on an electrochemical model and machine learning as claimed in claim 9, characterized in that: In step S6023, the performance of the support vector regression model is evaluated by the root mean square error and the mean absolute error; The formula is: Where: N represents the number of samples; represents the battery health status prediction value output by the support vector regression model; y m The real value representing the battery health status.

Images