A parameter identification method for lithium battery electrochemical model based on machine learning classifier and heuristic algorithm

Abstract

A parameter identification method for a lithium battery electrochemical model based on a machine learning classifier and a heuristic algorithm, which relates to a parameter identification method for a lithium battery electrochemical model, and aims to solve the technical problem of low identification efficiency of the existing parameter identification method for a lithium battery electrochemical model. The method: sensitivity analysis is performed on the parameters of the electrochemical model and they are classified, and highly sensitive parameters are selected as identification objects; the heuristic algorithm is initialized, parameter set data for classifier training and corresponding convergence labels are collected, and the correlation between each parameter in the parameter set and the convergence of the electrochemical model is calculated; a classifier based on a machine learning algorithm is constructed, and the classifier is first trained with parameter set data with high correlation, and then the classifier is fine-tuned with parameter set data with low correlation; the heuristic algorithm is reinitialized, and it is iterated until the set number of times or the accuracy meets the conditions, and the accurate identification of the parameters is completed. It can be used in the field of electrochemistry.

Description

Technical Field

[0001] The invention relates to a parameter identification method of an electrochemical model, and belongs to the technical field of electrochemical energy storage. Background Art

[0002] The electrochemical model of lithium batteries can accurately describe the internal process of the battery, providing a theoretical basis for battery state estimation, life prediction and performance optimization. The electrochemical model involves many parameters with practical physical meanings, which have a crucial impact on the simulation accuracy of the model. However, it is time-consuming and laborious to obtain model parameters through actual testing, and some parameters cannot be obtained through direct measurement.

[0003] At present, heuristic algorithms are widely used for parameter identification of electrochemical models due to their global search capabilities and robustness. Common heuristic algorithms include particle swarm optimization (PSO), genetic algorithm (GA), simulated annealing (SA), and grey wolf optimization (GWO). These algorithms perform global searches by simulating the optimization process in nature and can find the best set of model parameters within a set search domain. However, there are parameter combinations in the search domain that cannot make the electrochemical model converge. The process of bringing these parameter combinations into the model for calculation is not only invalid, but also seriously affects the computational efficiency of the identification program. Summary of the invention

[0004] The present invention aims to solve the problem that there are parameter combinations in the search domain of the heuristic algorithm of the existing lithium battery electrochemical model parameter identification method that cannot make the model converge, thereby resulting in low identification efficiency, and proposes a parameter identification method for the lithium battery electrochemical model based on a machine learning classifier and a heuristic algorithm.

[0005] The parameter identification method of the lithium battery electrochemical model based on the machine learning classifier and the heuristic algorithm of the present invention is carried out according to the following steps:

[0006] Step 1: Perform sensitivity analysis on all parameters in the electrochemical model, classify the parameters according to their sensitivity, and select highly sensitive parameters that have a significant impact on the model simulation results as identification objects;

[0007] Step 2: Set the search domain of each highly sensitive parameter, initialize the heuristic algorithm, collect the parameter set data for classifier training and its corresponding convergence label, calculate the correlation between each parameter in the parameter set and the convergence of the electrochemical model according to the Pearson correlation coefficient, and divide the parameters in the highly sensitive parameter set into a high-correlation parameter set and a low-correlation parameter set;

[0008] Step 3: Construct a classifier based on a machine learning algorithm. First, use the data of a high-correlation parameter set to train the classifier, and then use the data of a low-correlation parameter set to fine-tune the classifier, thereby improving the generalization ability and robustness of the model and optimizing the training efficiency. Use a high-correlation parameter set to train the classifier, quickly learn the classification ability of the main features for model convergence, speed up the convergence of the classifier, reduce the interference of low-correlation parameters in the initial stage, and make the initial model of the classifier have higher accuracy. Use low-correlation parameters as supplementary information to further optimize the classifier by expanding the feature space or adjusting hyperparameters, which helps the model adapt to the complexity of practical applications, avoids overfitting of the model to high-correlation parameters, and allows the classifier to better handle noise and outliers, thereby improving robustness in actual data.

[0009] Step 4: Reinitialize the heuristic algorithm and enable the trained machine learning classifier after the parameter set is updated in each iteration. Eliminate the parameter set that may cause the model to not converge in the search domain of each iteration, and complete the vacated search space by introducing global perturbations. Finally, terminate the iteration after reaching the set number of iterations to complete the identification of the model parameters.

[0010] Preferably, in step 1, the electrochemical model of the lithium-ion battery is a P2D model and its derivative model, and all parameters of the model include: geometric parameters: L n , L se , L p , R n , R p , ε s,n , ε s,p , ε e,n , ε e,se , ε e,p ; Transport parameters: D s,n , D s,p , D e ,t + , σ n , σ p , κ se , κ SEI , brug; kinetic parameters: k n , k p , k SEI 、E a,kn 、E a,kp 、E a,Ds,n 、E a,Ds,p 、E a,kSEI , R film ; Concentration parameter: c n,max 、c p,max 、c e,0 ; Mechanical parameters: E n 、E p , νn , ν p ,Ω n ,Ω p ; Thermodynamic parameters: C p , λ, h. The meaning and units of each parameter are shown in Table 1.

[0011] Table 1 Meaning and units of model parameters

[0012]

[0013]

[0014] Preferably, in step 1, the parameter sensitivity analysis method is: formulate the value boundaries of all model parameters, divide the variation range of each parameter into N equidistant intervals, and evenly select N values ​​in each interval as parameter candidate values ​​for evaluating model performance during sensitivity analysis; when studying the sensitivity of a certain parameter, keep other parameters as parameter standard values ​​unchanged, and only change the value of the parameter under study; the method for selecting the parameter standard value is: when the upper and lower boundaries have the same order of magnitude, take the arithmetic mean as the parameter standard value; when the upper and lower boundaries have different orders of magnitude, first take the logarithm of the upper and lower boundary values, then calculate the arithmetic mean on a logarithmic scale, and finally take the antilogarithmic restoration of the result to use it as the parameter standard value; the sensitivity calculation formula is:

[0015]

[0016] Among them, SI is the parameter sensitivity, V i is the voltage simulation value when different parameters are taken, and V is the voltage simulation value when the parameters are taken as standard values.

[0017] Preferably, in step 1, parameters whose SI values ​​are greater than or equal to 0.05 are highly sensitive parameters that have a significant impact on the model simulation results, and parameters whose SI values ​​are less than 0.05 are low-sensitivity parameters.

[0018] Preferably, in step 2, the search domain of the highly sensitive parameter can directly use the parameter value boundary set during the sensitivity analysis, or the set search domain can be adjusted and corrected through preliminary experiments.

[0019] Preferably, in step 2, the heuristic algorithm used in the identification process is particle swarm optimization (PSO), genetic algorithm (GA), simulated annealing (SA) or grey wolf optimization algorithm (GWO).

[0020] Preferably, in step 2, the correlation between each parameter in the parameter set and the convergence of the electrochemical model is calculated by calculating the Pearson correlation coefficient, and the specific calculation method is:

[0021]

[0022] Where r is the relevant system; x i and are the i-th parameter value and the mean value of a parameter in the parameter set, y i and The ith value and mean are respectively the convergence and non-convergence labels of the parameter set encoded as numerical values. Through the calculation of the Pearson correlation coefficient, the parameters with |r|≥0.2 are included in the high correlation parameter set that is significantly correlated with the model convergence; the high correlation parameter set, and the parameters with |r|<0.2 are included in the low correlation parameter set that is less correlated with the model convergence.

[0023] Preferably, in step 3, the classifier based on the machine learning algorithm is the following algorithm model: support vector machine (SVM), k-nearest neighbor algorithm (kNN), decision tree (DT), random forest (RF), gradient boosting decision tree (GBDT), naive Bayes (NB) or artificial neural network (ANN).

[0024] Preferably, in step 3, the training process of the classifier based on the machine learning algorithm directly uses the numerical value of the parameter as the feature input without performing an additional feature extraction process.

[0025] Preferably, in step 4, the trained machine learning classifier is enabled before each iteration, and the parameter set generated during each iteration, that is, the position information in the search domain updated with the iteration, is input into the classifier to identify the parameter sets that can make the model converge and those that cannot make the model converge, eliminate the non-convergent parameter sets and complete the vacated search space.

[0026] Preferably, the specific method for supplementing the search space vacated after eliminating the non-convergent parameter set is: by introducing global perturbations, generating effective search points in the search domain of each iteration that can make the classifier recognize the convergent parameter set, ensuring that the space after elimination can be effectively supplemented to avoid local optimality. If the generated points are invalid, they are regenerated until the conditions are met. The specific calculation formula is:

[0027] x new =x certer +β·R·(x max -x min )

[0028] Among them, x new is the new search point generated by perturbation, x center is the center point of the current high-quality area, R is a random number that follows the normal distribution N(0,1), x max and x minare the upper and lower bounds of the current search space respectively, β is the global disturbance factor. Considering factors such as algorithm complexity and search efficiency, the value of β can be set to a fixed constant between 0.1 and 1.

[0029] Preferably, in step 4, the number of times is set to 50 to 1000 times, which is set according to the needs of the actual problem.

[0030] The beneficial effects of the present invention are:

[0031] The present invention uses a machine learning classifier in conjunction with a heuristic algorithm to identify parameters of a lithium battery electrochemical model. First, the highly sensitive parameters that need to be identified are determined by parameter sensitivity analysis, which reduces the dimension of parameter identification and thus reduces the difficulty of identification; a high-correlation parameter set is used for model training, and a low-correlation parameter set is used for model fine-tuning, which improves the generalization ability and robustness of the model and optimizes the training efficiency; compared with a convergent parameter set, the process of bringing a non-convergent parameter set into an electrochemical model for calculation is extremely slow, and the process is completely ineffective. By enabling a trained machine learning classifier in a heuristic algorithm, the advantages of a trained machine learning classifier are fast running speed and high classification accuracy, which is used to eliminate non-convergent parameter sets and use newly generated convergent parameter sets to supplement the algorithm's search space; therefore, compared with existing methods, the present invention significantly improves the efficiency of electrochemical model parameter identification, avoids the waste of computing resources, and improves the accuracy of identification to a certain extent. And the present invention has achieved good test results in the verification of actual battery test data. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 A flow chart of a parameter identification method for a lithium battery electrochemical model based on a machine learning classifier and a heuristic algorithm;

[0033] Figure 2 is the result of parameter sensitivity analysis in Example 1;

[0034] Figure 3 is the calculation result of the Pearson correlation coefficient between the highly sensitive parameter and the classification result of the classifier in Example 1;

[0035] Figure 4 This is a flow chart of the heuristic algorithm for introducing a machine learning classifier in Example 1;

[0036] Figure 5 The confusion matrix diagram of the verification result of the machine learning classifier of Example 1;

[0037] Figure 6 is the ROC curve of the verification result of the machine learning classifier in Example 1;

[0038] Figure 7The performance evaluation indicators of the machine learning classifier in Example 1 include accuracy, precision, recall and F 1 coefficient;

[0039] Figure 8 It is a simulation effect diagram of the parameter identification result in Example 1, where the solid line is the actual test data, the triangle is the simulation result under 0.5C, the circle is the simulation result under 1C, and the cross is the simulation result under 2C. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] Example 1: The electrochemical model used is a derivative model of the P2D model that takes into account the coupling of the battery aging process and multiple physical fields. This model has the most comprehensive description of the internal process of the battery, while also taking into account the influence of multiple physical fields. It contains many types of model parameters, so this example 1 uses this model to explain the present invention in detail. When used for parameter identification of the original P2D model or other simplified electrochemical models with fewer parameters, the implementation process is the same as the process described in this example, and is simpler. The parameter identification method of the lithium battery electrochemical model based on a machine learning classifier (support vector machine) and a heuristic algorithm (particle swarm algorithm) in this example is performed according to the following steps: Step 1. The mathematical representation of the model used in this example 1 is shown in Table 2:

[0042] Table 2: Mathematical representation of the model of Example 1

[0043]

[0044]

[0045] The parameter description and classification of the model used in Table 2 are shown in Table 3:

[0046] Table 3: Model parameter description and classification

[0047]

[0048]

[0049] The model contains 40 parameters, and their acquisition methods are listed in Table 3. A small number of parameters can be directly obtained through measurement or by asking the manufacturer, but most of the remaining parameters are difficult or impossible to obtain through experiments, and the model parameter values ​​need to be determined by identification. In order to comprehensively examine the influence of the parameters on the final model output results, sensitivity analysis was performed on all the above parameters. The calculation method of parameter sensitivity is as follows:

[0050] The value boundaries of all the above model parameters are formulated, as shown in Table 3. The range of variation of each parameter is divided into N equally spaced intervals, and N values ​​are evenly selected in each interval as parameter candidate values ​​for evaluating model performance during sensitivity analysis; when studying the sensitivity of a certain parameter, other parameters are kept unchanged as parameter standard values, and only the value of the parameter under study is changed; the method for selecting parameter standard values ​​is: if the upper and lower boundaries have the same order of magnitude, the arithmetic mean is taken as the parameter standard value; if the upper and lower boundaries have different orders of magnitude, first take the logarithm of the upper and lower boundary values, then calculate the arithmetic mean on a logarithmic scale, and finally take the antilogarithm of the result and use it as the parameter standard value. The sensitivity calculation formula is:

[0051]

[0052] Among them, SI is the parameter sensitivity, Vi is the voltage simulation value when the parameter takes different values, and V is the voltage simulation value when the parameter takes the standard value.

[0053] Figure 2 The results of parameter sensitivity analysis are shown. The parameters are divided into two types according to their sensitivity: high-sensitivity parameters and low-sensitivity parameters. The highly sensitive parameters that have a significant impact on the model simulation results and are difficult to obtain directly through experiments are selected: anode solid phase volume fraction ε s,n , maximum lithium ion concentration at the anode c n,max , electrolyte conductivity κ se , cathode solid phase volume fraction ε s,p , anode reaction rate constant k n , anode solid phase diffusion coefficient D s,n , anode electrolyte volume fraction ε e,n , electrolyte diffusion coefficient D e , the lumped heat transfer coefficient h and the cathode solid phase diffusion coefficient D s,p as an object of identification.

[0054] Step 2: Set the search domain of each highly sensitive parameter to be identified. In this embodiment, the value range of each parameter during sensitivity analysis is directly used as the search domain, initialize the heuristic algorithm, collect parameter set data for classifier training and its corresponding convergence label, and calculate the correlation between each parameter in the parameter set and the convergence of the electrochemical model according to the Pearson correlation coefficient, and divide the parameters in the highly sensitive parameter set into a high-correlation parameter set and a low-correlation parameter set;

[0055] Among them, the method of initializing the heuristic algorithm to collect data for classifier training specifically refers to: using the particle swarm algorithm, setting the search range of each parameter, initializing the particle swarm algorithm, randomly generating the initial position of each particle in the search space, and each particle corresponds to a combination of a set of model parameters in the parameter set, thereby generating an initial parameter set. The initial parameter set is brought into the model for calculation, and it is determined whether each set of parameter combinations can make the model converge, and the convergence results are encoded into numerical values ​​and stored together with the corresponding parameter combinations. If the model can converge, it is recorded as "1", and if the model cannot converge, it is recorded as "0".

[0056] The method of calculating the correlation between parameters and model convergence according to the Pearson correlation coefficient specifically refers to: analyzing the correlation between each parameter and the classification result of the classifier through the Pearson correlation coefficient calculation formula, and the calculation formula is:

[0057]

[0058] Among them, x i and are the i-th parameter value and the mean value of a parameter in the parameter set, y i and The ith value and mean of the convergence and non-convergence labels of the parameter set are encoded as numerical values. By calculating the Pearson correlation coefficient, the parameter set with |r|≥0.2 that is significantly correlated with the model convergence is defined as a high-correlation parameter set, and the parameter set with |r|<0.2 that is less correlated with the model convergence is defined as a low-correlation parameter set.

[0059] Figure 3 The Pearson correlation coefficient calculation results of the highly sensitive parameters to be identified and the classification results of the classifier are shown. The parameters are divided into two categories according to the size of the |r| value: The high correlation parameter set is: the maximum lithium ion concentration of the anode c n,max , anode reaction rate constant k n , anode solid phase diffusion coefficient D s,n , anode electrolyte volume fraction ε e,n and the lumped heat transfer coefficient h; the low correlation parameter set is: anode solid volume fraction ε s,n , electrolyte conductivity κ se , cathode solid phase volume fraction εs,p , electrolyte diffusion coefficient D e and the cathode solid phase diffusion coefficient D s,p .

[0060] Step 3, construct a classifier based on the support vector machine algorithm. The specific method is: first select the Gaussian radial basis function as the kernel function of the support vector machine, map the data to a high-dimensional space, then set the type of the support vector machine to solve the binary classification problem, then set the initial hyperparameters of the support vector machine and perform hyperparameter optimization. The hyperparameters are the penalty coefficient C and the kernel function parameter γ. Use a high-correlation parameter set to train the classifier, and use a low-correlation parameter set to fine-tune the model.

[0061] The specific method of training a classifier using a high-correlation parameter set is: divide the high-correlation parameter set data into a training set and a test set, set the hyperparameters of the machine learning classifier model, and optimize the hyperparameters using grid search, random search or other optimization methods, use the training set data to optimize the model parameters, minimize the loss function, and use k-fold cross validation to preliminarily evaluate the model performance. This step is to quickly establish the basic capabilities of the model, speed up the convergence of the classifier, reduce the interference of low-correlation parameters in the initial stage, and make the initial model of the classifier have a higher accuracy;

[0062] The method of fine-tuning the model using a low-correlation parameter set specifically refers to: concatenating the low-correlation parameter set with the high-correlation parameter set to form an expanded feature space, then introducing nonlinear feature interactions of high-order terms or combination terms to further expand the feature space, and finally using L1 regularization to select important features. For the expanded feature space, by adjusting the penalty coefficient C and kernel function parameter γ of the classifier, adapting the newly added features, improving the classification performance, and finally completing the training of the classifier. This step helps the model adapt to the complexity of practical applications, avoids overfitting of the model to high-correlation parameters, allows the classifier to better handle noise and outliers, and improves robustness in actual data.

[0063] Step 4. After completing the construction and training of the classifier, reinitialize the particle swarm algorithm and randomly generate the initial position of each particle in the set search space. The position of each particle corresponds to a combination of model parameters. Enable the trained classifier, input the position of each particle, that is, the combination of each set of parameters, into the classifier, and divide all parameter combinations generated in this iteration into convergent parameter sets and non-convergent parameter sets according to the output results of the classifier. Then retain the convergent parameter set and eliminate the non-convergent parameter set, and then introduce global perturbations to complete the search space vacated after eliminating the non-convergent parameter set.

[0064] The method of supplementing the vacated search space specifically refers to: by introducing global perturbations, generating effective search points in the search domain of each iteration that can make the classifier recognition result a convergence parameter set, ensuring that the space after elimination can be effectively supplemented to avoid local optimality. If the generated points are invalid, they are regenerated until the conditions are met. The specific calculation formula is:

[0065] x new =x center +β·R·(x max -x min )

[0066] Among them, x new is the new search point generated by perturbation, x center is the center point of the current high-quality area, β is the global disturbance factor, and in this embodiment, β is set to 0.2, R is a random number that obeys the normal distribution N(0,1), and x max and x min are the upper and lower bounds of the current search space respectively.

[0067] After using the classifier to adjust and optimize the search space of the particle swarm algorithm, the fitness of the parameter set represented by each particle is calculated to obtain the parameter set of the optimal solution, and the parameter set of the next iteration is updated according to the speed and position update method of the particle swarm algorithm. In the next iteration, the classifier is first used to eliminate the non-convergent parameter set in the updated parameter set and fill the vacated search space, and the cyclic iterative operation is performed according to this rule.

[0068] After the set number of iterations reaches 100, the iteration is terminated, the parameter identification of the model is completed, and the obtained parameters are the final identification results.

[0069] Figure 4 Flowchart of the heuristic algorithm for introducing the machine learning classifier in step 4.

[0070] Figure 5 is a confusion matrix diagram of the verification result of the machine learning classifier of Example 1; Figure 5 It can be seen intuitively that the classifier has a good classification effect, and according to Figure 5 The confusion matrix diagram can calculate various indicators for evaluating the performance of the classifier.

[0071] Figure 6 is the ROC curve of the verification result of the machine learning classifier in Example 1; Figure 6 The area under the ROC curve is the AUC value. The AUC value of the classifier is 0.9637, which is very close to the perfect classifier with an AUC value of 1, indicating that the classifier in this embodiment has a good classification effect.

[0072] Figure 7The performance evaluation indicators of the machine learning classifier in Example 1 include accuracy, precision, recall and F 1 Coefficient; from Figure 7 It can be seen that the accuracy, precision, recall and F of the classifier 1 The coefficients are all greater than 90%, indicating that the classifier has excellent performance.

[0073] Figure 8 The simulation effect diagram of the parameter identification result in Example 1, the solid line in the figure is the actual test data, the triangle is the simulation result under 0.5C, the circle is the simulation result under 1C, and the cross is the simulation result under 2C. Figure 8 It can be seen that bringing the parameter identification results into the model for simulation can obtain good terminal voltage simulation effects, verifying the accuracy of the parameters.

[0074] In this embodiment, the performance of the machine learning classifier and the accuracy verification results of the final parameter identification results are as follows: Figures 5 to 8 As shown, the good performance of the machine learning classifier in this embodiment and the accuracy of the parameter identification results are demonstrated.

[0075] Although the present invention is described with reference to specific implementations (using support vector machines in machine learning to build classifiers and particle swarm algorithms in heuristic algorithms as examples) in this Example 1, it should be understood that these embodiments are merely examples of the principles and applications of the present invention. Therefore, it should be understood that many modifications can be made to the exemplary embodiments, and other arrangements can be designed without departing from the spirit and scope of the present invention as defined by the appended claims.

Claims

1. A parameter identification method for a lithium battery electrochemical model based on a machine learning classifier and a heuristic algorithm, characterized in that: The method proceeds as follows: Step 1: Perform sensitivity analysis on all parameters in the electrochemical model, classify the parameters according to their sensitivity, and select highly sensitive parameters that have a significant impact on the model simulation results as identification objects; Step 2: Set the search domain of each highly sensitive parameter, initialize the heuristic algorithm, collect the parameter set data for classifier training and its corresponding convergence label, calculate the correlation between each parameter in the parameter set and the convergence of the electrochemical model according to the Pearson correlation coefficient, and divide the parameters in the highly sensitive parameter set into a high-correlation parameter set and a low-correlation parameter set; Step 3: Build a classifier based on the machine learning algorithm. First, use the data of the high-correlation parameter set to train the classifier, and then use the data of the low-correlation parameter set to fine-tune the classifier, so as to improve the generalization ability and robustness of the model and optimize the training efficiency. Step 4: Reinitialize the heuristic algorithm, enable the trained machine learning classifier after the parameter set is updated in each iteration, eliminate the parameter set that may cause the model to not converge in the search domain of each iteration, and complete the vacated search space by introducing global perturbations. Finally, terminate the iteration when the set number of iterations is reached to complete the identification of the model parameters. The specific method for supplementing the search space vacated after eliminating the non-convergent parameter set is as follows: by introducing global perturbations, effective search points are generated in the search domain of each iteration to enable the classifier to identify the convergent parameter set, ensuring that the space after elimination can be effectively supplemented to avoid local optimality; if the generated points are invalid, they are regenerated until the conditions are met; the specific calculation formula is: x new =x center +β·R·(x max -x min ) Among them, x new is the new search point generated by perturbation, x center is the center point of the current high-quality area, R is a random number that follows the normal distribution N(0,1), x max and x min are the upper and lower bounds of the current search space respectively, β is the global perturbation factor, and the value of β is set to a fixed constant between 0.1 and 1.

2. The parameter identification method of a lithium battery electrochemical model based on a machine learning classifier and a heuristic algorithm according to claim 1, characterized in that: The electrochemical model of the lithium-ion battery described in step 1 is the P2D model and its derivative model. All parameters of the model include: Geometric parameters: L n , anode thickness, μm; L se , membrane thickness, μm; L p , cathode thickness, μm; R n , anode particle radius, μm; R p , cathode particle radius, μm; ε s,n , anode solid volume fraction, dimensionless; ε s,p , cathode solid volume fraction, dimensionless; ε e,n , anode electrolyte volume fraction, dimensionless; ε e,se , separator electrolyte volume fraction, dimensionless; ε e,p , cathode electrolyte volume fraction, dimensionless; Transport parameter: D s,n , anode solid phase diffusion coefficient, unit is m 2 s -1 ; D s,p , cathode solid phase diffusion coefficient, unit is m 2 s -1 ; D e , electrolyte diffusion coefficient, unit is m 2 s -1 ; t + , lithium ion transfer number, dimensionless; σ n , anode conductivity, anode conductivity, unit is Sm -1 ; σ p , cathode conductivity, anode conductivity, unit is Sm -1 ; K se , electrolyte conductivity, unit is Sm -1 ; κ SEI , SEI conductivity, unit is Sm -1 ; brug, Bruggeman coefficient; Kinetic parameter: k n , the anodic reaction rate constant, in m 2.5 mol -0.5 s -1 ; k p , cathode reaction rate constant, in m 2.5 mol -0.5 s -1 ; k SEI , SEI growth rate constant, unit is m 2.5 mol -0.5 s -1 ; E a,kn , the activation energy of the anode reaction, in kJ mol -1 ; E a,kp , cathode reaction activation energy, in kJ mol -1 ; E a,Ds,n , anode solid phase diffusion activation energy, unit is kJ mol -1 ; E a,Ds,p , cathode solid phase diffusion activation energy, unit is kJ mol -1 ; E a,kSEI , SEI growth activation energy, in kJ mol -1 ; R film , anode film resistance, unit is Ωm 2 ; Concentration parameter: c n,max , the maximum lithium ion concentration at the anode, in mol m -3 ; c p,max , the maximum lithium ion concentration at the cathode, in mol m -3 ; c e,0 , initial lithium ion concentration of the electrolyte, in mol m -3 ; Mechanical parameters: E n , Young's modulus of anode particles, in GPa; E p , the unit of Young’s modulus of cathode particles is GPa; ν n , Poisson's ratio of anode particles, dimensionless; ν p , Poisson’s ratio of cathode particles, dimensionless; Ω n , partial molar volume of anode particles, unit is m 3 mol -1 ; Ω p , partial molar volume of cathode particles, unit is m 3 mol -1 ; Thermodynamic parameters: C p , lumped specific heat capacity, unit is Jkg -1 K -1 ; λ, lumped thermal conductivity, in W m -2 K -1 ; h, lumped heat transfer coefficient, in W m -2 K -1 .

3. A parameter identification method for a lithium battery electrochemical model based on a machine learning classifier and a heuristic algorithm according to claim 1 or 2, characterized in that: The parameter sensitivity analysis method described in step 1 is: formulate the value boundaries of all model parameters, N = 5 to 20; when studying the sensitivity of a certain parameter, keep other parameters as parameter standard values ​​and only change the value of the parameter being studied; the method for selecting the parameter standard value is: if the upper and lower boundaries have the same order of magnitude, take the arithmetic mean as the parameter standard value; if the upper and lower boundaries have different orders of magnitude, first take the logarithm of the upper and lower boundary values, then calculate the arithmetic mean on the logarithmic scale, and finally take the antilogarithm of the result to restore it as the parameter standard value; the sensitivity calculation formula is: Among them, SI is the parameter sensitivity, V i is the voltage simulation value when different parameters are taken, and V is the voltage simulation value when the parameters take standard values.

4. A parameter identification method for a lithium battery electrochemical model based on a machine learning classifier and a heuristic algorithm according to claim 1 or 2, characterized in that: In step 2, the heuristic algorithm used in the identification process is particle swarm optimization, genetic algorithm, simulated annealing or grey wolf optimization algorithm.

5. The parameter identification method of a lithium battery electrochemical model based on a machine learning classifier and a heuristic algorithm according to claim 1 or 2, characterized in that: In step 2, the correlation between each parameter in the parameter set and the convergence of the electrochemical model is calculated by calculating the Pearson correlation coefficient. The specific calculation method is: Among them, r is the correlation coefficient; x i and are the i-th parameter value and the mean value of a parameter in the parameter set, y i and They are the i-th value and mean after the convergence and non-convergence labels of the parameter set are encoded as numerical values; through the calculation of the Pearson correlation coefficient, the parameters with |r|≥0.2 are included in the high-correlation parameter set that is significantly correlated with the convergence of the model; the high-correlation parameter set, and the parameters with |r|<0.2 are included in the low-correlation parameter set that has a lower correlation with the convergence of the model.

6. A parameter identification method for a lithium battery electrochemical model based on a machine learning classifier and a heuristic algorithm according to claim 1 or 2, characterized in that: In step 3, the classifier based on the machine learning algorithm is the following algorithm model: support vector machine, k-nearest neighbor algorithm, decision tree, random forest, gradient boosting decision tree, naive Bayes or artificial neural network.

7. The parameter identification method of a lithium battery electrochemical model based on a machine learning classifier and a heuristic algorithm according to claim 1 or 2, characterized in that: The training process of the classifier based on the machine learning algorithm directly uses the parameter value as the feature input without performing additional feature extraction process.

8. The parameter identification method of a lithium battery electrochemical model based on a machine learning classifier and a heuristic algorithm according to claim 1 or 2, characterized in that: In step 4, before each iteration, the trained machine learning classifier is enabled, and the parameter set generated during each iteration, that is, the position information in the search domain updated with the iteration, is input into the classifier to identify the parameter sets that can make the model converge and those that cannot make the model converge, eliminate the non-convergent parameter sets and complete the vacated search space.

Images