Five-stage constant-current battery charging optimization method based on electrochemical-thermal coupling aging model

Abstract

The invention discloses a five-stage constant current battery charging optimization method based on an electrochemical-thermal coupling aging model, and relates to the technical field of lithium ion battery charging, and the technical scheme is characterized in that high-quality sample data is generated by constructing a physical model considering SEI film generation side reaction and lithium precipitation side reaction, and the high-quality sample data is analyzed; cooperative optimization of the charging time, the capacity retention ratio and the highest temperature is realized by combining CatBoost multi-target regression and Pareto leading edge search, and finally, the method has good engineering applicability by comprehensively considering subjective and objective importance and an output compromise solution.

Description

Technical Field

[0001] This invention relates to the field of lithium-ion battery charging technology, and more specifically, to a five-stage constant current battery charging optimization method based on an electrochemical-thermal coupled aging model. Background Technology

[0002] With the rapid development of electric vehicles and energy storage systems, the fast-charging performance of lithium-ion batteries has become a key research direction. Traditional charging protocols are mostly based on empirical design or single-objective optimization, making it difficult to achieve an effective balance between charging time, capacity retention, and safety. Especially under different aging conditions and temperature conditions, fixed charging strategies are prone to side reactions such as excessive growth of the SEI film and lithium plating, which can accelerate battery aging and even cause safety accidents. Therefore, there is an urgent need to develop charging methods that consider multiple performance aspects.

[0003] Current research on battery charging optimization by scholars both domestically and internationally mainly relies on equivalent circuit models and optimization algorithms to find optimal parameters. While this approach can achieve the optimization goal, it suffers from limitations in model accuracy, lack of electrochemical mechanism support, and insufficient physical interpretability. Therefore, this invention proposes a five-stage constant-current battery charging optimization method based on an electrochemical-thermal coupled aging model. Summary of the Invention

[0004] The above-mentioned technical objective of the present invention is achieved through the following technical solution: The first aspect of this invention provides a five-stage constant current battery charging optimization method based on an electrochemical-thermal coupled aging model, comprising the following steps: A Newman pseudo-two-dimensional electrochemical model was constructed in COMSOL multiphysics simulation software. Battery parameters were input, and the performance parameters of the battery model under five-stage constant current charging conditions were obtained. The optimal Latin hypercube sampling method was used to obtain samples of the charging rate of each stage in the five-stage constant current charging process. These samples were then input into the Newman pseudo-two-dimensional electrochemical model to obtain the charging time, capacity retention after n cycles, and maximum temperature during the process. This allowed for the construction of a database of multiple battery performance parameters under different charging conditions. Use Catboost to build a multi-objective agent model, train it, and obtain a joint prediction model; The TreeSHAP algorithm is used to build an interpreter for the joint prediction model and to quantify and interpret the three performance indicators of the battery respectively. The Pareto solution set is obtained by using a joint prediction model combined with an improved MOAPSO for multi-objective optimization. TOPSIS, which integrates subjective and objective weights, scores each solution in the Pareto solution set and outputs all performance parameters of the compromise solution and its TOPSIS score.

[0005] In conjunction with the first aspect, the present invention is further configured such that the five-stage constant current charging conditions are: 10%-30%SOC, 30%-50%SOC, 50-70%SOC, 70%-80%SOC, 80%-90%SOC, and the charging rate of each stage is 3C>R1>R2>R3>R4>R5>0.5C.

[0006] In conjunction with the first aspect, the present invention is further configured such that: the battery parameters include the SEI film formation reaction and the lithium deposition reaction, and a time acceleration factor is input into the stoichiometric coefficients; The local current density equation for the SEI film formation reaction is: ; In the formula, k 0,SEI It is the SEI reaction rate constant; U SEI It is the equilibrium potential for the SEI formation reaction, C S EC It is the concentration of EC on the graphite surface. R It is the gas constant. T It's temperature. F It is Faraday's constant. Where α is the cathode transfer coefficient and α is the active specific surface area. It is the electric potential of the solid phase (graphite). It is the potential of the electrolyte phase. j tot It is the total current density. R film It is the surface film resistance; The transfer current density of the lithium deposition reaction is: ; In the formula, i 0，lpl α is the exchange current density of lithium deposition, a is the specific surface area of ​​the electrode, and α is the specific current density of the electrode. c,lpl R is the cathode charge transfer coefficient of the lithium deposition reaction, T is the ideal gas constant, and T is the absolute temperature in Kelvin (K). It is the electric potential of the solid phase (graphite). It is the potential of the electrolyte phase, j tot It is the total current density, R film It is the surface film resistance.

[0007] In conjunction with the first aspect, the present invention is further configured as follows: the step of using Catboost to construct a multi-objective surrogate model, training it, and obtaining a joint prediction model includes: dividing the training sample set into a training set and a test set; performing multi-objective robust normalization on the training set to obtain a normalized training set; constructing a five-fold cross-validation to obtain the optimal hyperparameters; using the optimal hyperparameters, setting the number of iteration rounds, and introducing an early stopping mechanism, stopping training if the validation set MultiRMSE does not decrease for 200 consecutive rounds; and retraining the CatBoost multi-output regressor using the normalized training set as input to obtain the joint prediction model.

[0008] In conjunction with the first aspect, the present invention is further configured such that: the multi-objective optimization is to minimize charging time, maximize capacity retention, and minimize maximum temperature.

[0009] In conjunction with the first aspect, the present invention is further configured such that: the scoring of each solution in the Pareto solution set based on the TOPSIS fusion of subjective and objective weights includes: The standardized matrix is ​​multiplied element-wise by the comprehensive weight vector to obtain the weighted matrix; Calculate the Euclidean distance from each candidate solution to the positive ideal solution, and calculate the Euclidean distance from each candidate solution to the negative ideal solution; then calculate the relative proximity. Select the candidate solution with the highest relative proximity as the compromise solution.

[0010] In conjunction with the first aspect, the present invention is further configured such that the method for calculating the comprehensive weight vector is: ; In the formula, α is the fusion coefficient, W1 is the first weight vector, and W2 is the second weight vector.

[0011] A second aspect of the present invention also provides an apparatus / device / system for optimizing five-stage constant current battery charging based on an electrochemical-thermal coupled aging model, comprising a memory, a processor, and a computer program stored in the memory, characterized in that the processor executes the computer program to implement the steps of the above method.

[0012] A third aspect of the present invention also provides a computer-readable storage medium having a computer program / instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0013] A fourth aspect of the present invention also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the above-described method.

[0014] In summary, the present invention has the following beneficial effects: This invention generates high-quality sample data by constructing a physical model that considers the side reactions of SEI film formation and lithium plating. It uses CatBoost multi-objective regression to construct a joint prediction model and SHAP for interpretability analysis to determine the degree of influence of the charging rate at each stage on battery performance. By combining CatBoost multi-objective regression and Pareto frontier search, it achieves synergistic optimization of charging time, capacity retention, and maximum temperature. Finally, by comprehensively considering the importance of subjective and objective factors, the output compromise solution has good engineering applicability. Attached Figure Description

[0015] Figure 1 This is a current-voltage change curve diagram from Embodiment 2 of the present invention; Figure 2 This is the accuracy evaluation standard for the proxy model in Embodiment 2 of the present invention; Figure 3 This is a comparison chart of actual and predicted values ​​in Embodiment 2 of the present invention; Figure 4 This is a swarm diagram of the charging time in Embodiment 2 of the present invention; Figure 5 This is a colony diagram showing the capacity retention rate in Embodiment 2 of the present invention; Figure 6 This is a bee colony diagram showing the maximum temperature in Embodiment 2 of the present invention; Figure 7 This is the Pareto front diagram in Embodiment 2 of the present invention; Figure 8 This is the compromise solution diagram selected by the subjective-objective fusion method in Embodiment 2 of the present invention; Figure 9 This is the compromise solution diagram selected by the subjective-objective fusion method in Embodiment 2 of the present invention. Detailed Implementation

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

[0017] Example 1: 1. Construct an electrochemical-thermal-aging coupled model A Newman pseudo-two-dimensional electrochemical model was constructed using the lithium-ion battery module in COMSOL multiphysics simulation software. In the model developer, under Component 1, the geometry option was selected to define the positive electrode, negative electrode, and separator length. Under the materials option, the positive electrode, negative electrode, and electrolyte materials were selected from the battery materials library and assigned to the respective domains of the lithium-ion battery.

[0018] Select the separator option and enter the electrolyte volume fraction. Add a porous electrode option, selecting the positive electrode as the active area, and enter the corresponding parameters in particle intercalation and porous electrode reaction. Then add another porous electrode option, selecting the negative electrode as the active area, and enter the corresponding parameters in particle intercalation and porous electrode reaction. In the porous electrode-negative electrode option, add two porous electrode reactions. Input the following two equations into the two porous electrode reactions respectively to represent the other two side reactions occurring at the negative electrode of the lithium-ion battery: the SEI film formation reaction and the lithium deposition reaction. Enter the time acceleration factor in the stoichiometric coefficients.

[0019] Local current density of SEI film formation reaction ; In the formula, k 0,SEI It is the SEI reaction rate constant; U SEI It is the equilibrium potential for the SEI formation reaction, C S EC It is the concentration of EC on the graphite surface. R It is the gas constant. T It's temperature. F It is Faraday's constant. Where α is the cathode transfer coefficient and α is the active specific surface area. It is the electric potential of the solid phase (graphite). It is the potential of the electrolyte phase. j tot It is the total current density. R film It is the surface film resistance; Transfer current density of lithium deposition reaction ; In the formula, i 0，lpl α is the exchange current density of lithium deposition, a is the specific surface area of ​​the electrode, and α is the specific current density of the electrode. c,lpl R is the cathode charge transfer coefficient of the lithium deposition reaction, T is the ideal gas constant, and T is the absolute temperature in Kelvin (K). It is the electric potential of the solid phase (graphite). It is the potential of the electrolyte phase, j tot It is the total current density, R film It is the surface film resistance.

[0020] In the menu bar, select Add Component 2 and choose Solid Heat Transfer Field in the Physics Field. Define the geometry of the solid using the Geometry option, then select Solid Heat Transfer and add heat source and heat flux options. The heat source is the internal heat generated during the charging and discharging of the lithium-ion battery, mainly including reaction heat, ohmic heat, and polarization heat. Then select the heat flux and solid options respectively and input the corresponding parameters. Select Component 1 in the Model Developer and add a field in the main menu bar to add the event function to Component 1 to achieve five-stage constant current charging. Set boundary conditions: ground the negative terminal and input the electrode current at the positive terminal. After meshing, add the current distribution initialization option in the Study options to obtain the performance parameters of the battery model under five-stage constant current charging conditions.

[0021] 2. Optimal Latin hypercube sampling This invention employs a five-stage constant current charging method based on SOC segmentation. The segmentation criteria are 10%-30% SOC, 30%-50% SOC, 50-70% SOC, 70%-80% SOC, and 80%-90% SOC, with the charging rate of each stage being 3C>R1>R2>R3>R4>R5>0.5C. Sample points (charging method samples) of the charging rate of each stage in the five-stage constant current charging are obtained through optimal Latin hypercube sampling.

[0022] First, determine the input design parameters for optimal Latin hypercube sampling, including the number of sample points n, variable dimension d, variable range [low, high], maximum number of iterations, and cost function weights. Then, generate an initial ordered Latin hypercube design, ensuring that each row of variables is strictly ordered in descending order, and introduce random perturbations to avoid boundary clustering. Construct a multi-objective cost function, integrating the φ_p criterion, maximum blank space, average distance, coverage rate, and boundary coverage rate, using a weighted normalization form, and treating the boundary coverage rate as a penalty term, adding up to an additional penalty of 0.5 levels when the boundary coverage rate is below 80%.

[0023] Objective function = α•φ_p / φ_p_ref + β•max_blank / max_blank_ref γ•avg_dist / avg_dist_ref δ•coverage / coverage_ref+penalty Where α, β, γ, and δ are the weights of each optimization objective, penalty is the penalty term, and φ_p_ref, max_blank_ref, avg_dist_ref, and coverage_ref are normalization reference values.

[0024] The current design is perturbed using multiple perturbation strategies, including sparse region optimization, boundary optimization, coverage enhancement, average distance enhancement, and random jumps. Each perturbation strategy is assigned a trigger probability, and the probabilities of these five strategies are summed to 1. The Euclidean distance of all point pairs is calculated in one step using Python's spatial distance calculation module, providing a unified data source for the φ_p criterion, minimum distance, and average distance. A dynamic KD-Tree (spatial index structure) is constructed to return the nearest neighbor distance and coordinates of each sample point in real time. This is used for calculating the maximum blank index, identifying sparse regions, and directional movement of clustered points, ensuring that the optimization process has both a global statistical perspective and local topological accuracy. A simulated annealing framework is used to accept new designs, combined with a dynamic cooling strategy and restart mechanism to prevent premature stopping. Post-processing optimization is performed on the final design to identify sparse points, clustered points, and internal points, and local refinement is performed to further improve the performance indicators. Finally, an ordered Latin hypercube design that satisfies descending order constraints, boundary coverage ≥80%, and achieves multi-objective comprehensive optimality is output.

[0025] The sample points generated by the optimal Latin hypercube design are input into the electrochemical-thermal coupled aging model constructed in COMSOL multiphysics software. The model is then input into the event function to perform five-stage constant current charging, thereby obtaining the charging time of lithium-ion batteries from SOC=10% to SOC=90%, the capacity retention rate after n cycles, and the maximum temperature during this process. A database of five-stage constant current charging methods and multiple battery performance characteristics under different charging conditions is constructed.

[0026] 3. Construct a multi-objective proxy model Data preparation and partitioning The data access module reads the charging method samples generated by the optimal Latin hypercube design and the corresponding three performance values ​​(charging time for lithium-ion batteries from SOC=10% to SOC=90%, capacity retention rate after n cycles, and maximum temperature during this process); the design matrix contains 5 columns of charging rate parameters, denoted as R1, R2, R3, R4, and R5; the three performance true values ​​are denoted as charging time Tc, capacity retention rate R... c and maximum temperature T m The data was split into training and test sets with a sample ratio of 8:2, and a fixed random seed was used to ensure the reproducibility of the experiment.

[0027] Multi-objective robust normalization For each dimension of the target variable in the training set, an independent robust normalization mapping is established. Specifically, a linear transformation centered on the median and scaled by the interquartile range, i.e., RobustScaler, is adopted to make the normalized data resilient to outliers. The three-dimensional target variables of the training set are mapped to the normalized space to obtain the normalized training target. The target variables of the test set are mapped again in subsequent steps to avoid information leakage.

[0028] Hyperparameter optimization based on Optuna A five-fold cross-validation loop is constructed. Within each fold, a joint model that can simultaneously output Tc, Rc, and Tm is established using CatBoost's multi-output regression mode and MultiRMSE as the loss function. The hyperparameter space to be optimized is defined, and the TPE Bayesian optimizer is used to perform 60 iterations of search within the space. The five-fold average RMSE is used as the fitness function, and the hyperparameter combination that minimizes the fitness is recorded as the optimal hyperparameter.

[0029] Final model training Using the optimal hyperparameters, the number of iterations was set to 6000, and an early stopping mechanism was introduced. If the validation set MultiRMSE did not decrease for 200 consecutive iterations, training was stopped. The CatBoost multi-output regressor was retrained using all training samples as input to obtain the joint prediction model.

[0030] Inverse normalization and error assessment The charging method data of the test set is input into the final joint prediction model, and the normalized predicted values ​​output by the model are restored to physical dimensions through inverse mapping to obtain the predicted values ​​of the test set. The following five error indicators are calculated: coefficient of determination R², mean square error MSE, root mean square error RMSE, mean absolute error MAE, mean absolute percentage error, and mean absolute error MAPE. The above indicators are output according to the target variable dimension to form a structured evaluation report. For each target variable dimension, a two-dimensional scatter plot of the true value versus the predicted value is plotted, and a y=x reference line is superimposed on the plot to visually observe the system bias.

[0031] SHAP-based global interpretation The TreeSHAP algorithm is used to build an interpreter for the trained model. The interpreter is used to calculate the SHAP value matrix of the training set samples to obtain the marginal contribution of each dimension of the target variable relative to the five rate parameters. Two visualizations are generated: a bar chart - sorted by average absolute SHAP value to show the global importance of each rate parameter to the target variable; and a bee colony plot - showing the coupling relationship between the SHAP value distribution and feature values, which is used to discover nonlinear sensitive intervals.

[0032] 4. Multi-objective optimization Model loading and preprocessing Load the pre-trained multi-objective regression model (CatBoostRegressor) above, which takes the input vector as input. The model outputs three target variables: ChargeTime (charging time), CapacityRet (capacity retention rate), and MaxTemp (maximum temperature). A corresponding normalizer is loaded to inversely normalize the model output to the actual physical quantities. A prediction function is constructed to map the input vector R to the three target values ​​[ChargeTime, CapacityRet, MaxTemp].

[0033] Objective function construction Define the objective functions to transform the prediction results into a standard multi-objective form with minimization directions: Objective 1: Minimize ChargeTime (charging time), Objective 2: Minimize -CapacityRet (i.e., maximize capacity retention), Objective 3: Minimize MaxTemp (maximum temperature). The output is a two-dimensional array of shape (N, 3), with each row corresponding to the three objective values ​​of a sample.

[0034] Particle swarm initialization Initialize particle swarm parameters: number of particles, number of iterations, and maximum storage limit. Generate a 5-dimensional input vector for each particle. The following conditions must be met: Value range: each dimension ∈ [0.5, 3.0]; strict monotonically decreasing constraint: If the generated vector does not satisfy monotonicity, it is forced to adjust to ensure monotonicity and the boundary is clipped. The particle velocity is initialized as a zero matrix with the same shape as the particle position.

[0035] Pareto archive initialization Calculate the objective function value of the initial particles to form the initial Pareto solution set. Construct an external archive to store non-dominated solutions: use Pareto dominance relations to determine the quality of solutions: if solution p is not inferior to solution q on all objectives, and is superior on at least one objective, then p is said to dominate q. All undominated solutions are stored in the archive, and dominated solutions are removed. If the archive size exceeds the upper limit, a crowding mechanism is used to truncate the archive, retaining the most evenly distributed solutions.

[0036] Iterative optimization process For each iteration, perform the following operations: update weights and coefficients, with inertia weight w = 0.5 + 0.5cos(πt / T), employing a cosine annealing strategy to enhance the balance between global exploration and local development. Learning factor. The weight of cognitive items decreases; learning factors The weight of social items increases progressively.

[0037] Global Optimal Selection: The crowding density of all solutions in the archive is calculated as the selection probability. A roulette wheel selection mechanism is used to choose a globally optimal solution from a sparsely distributed region in the archive, promoting diversity. Cauchy mutation is performed on particles with a probability of 0.1, followed by monotonically decreasing iterations to ensure monotonicity. Individual historical optimal solutions are updated only when the new solution dominates the original globally optimal solution. The target values ​​of all current particles are sent to the archive, and dominance removal is performed: all old solutions dominated by the new solution are deleted; and duplication removal is performed: duplicate solutions with identical target values ​​are removed; if the archive exceeds its capacity, the n most evenly distributed solutions are retained based on their crowding density.

[0038] Results Output and Visualization After the iteration is complete, all non-dominated solutions are extracted from the archive and input into the prediction model to obtain the true target value.

[0039] Construct the output data structure, including: input parameters: The output targets are: ChargeTime, CapacityRet, and MaxTemp. After deduplication, save the data as `pareto_front.csv`, keeping four decimal places. Plot a 3D Pareto front scatter plot with the following axes: X-axis: ChargeTime (charging time), Y-axis: CapacityRet (capacity retention rate), Z-axis: MaxTemp (maximum temperature). 5. Choosing a compromise solution based on subjectivity and objectivity Data acquisition and target direction setting Read the multi-objective optimization result dataset from the storage medium. The dataset contains multiple candidate solutions and their corresponding performance index vectors: charging time (ChargeTime), capacity retention rate (CapacityRet), and maximum temperature (MaxTemp). Set the optimization direction vector for each index, which includes maximizing (↑) or minimizing (↓). For example, the smaller the charging time and maximum temperature, the better, and the larger the capacity retention rate, the better.

[0040] Entropy weight method objective weighting module Construct a normalization matrix: Based on the direction of each indicator, perform linear normalization on each column of indicators to map them to the [0,1] interval; Calculate the information entropy value of each indicator: Based on the normalized indicator values, calculate the information entropy of each indicator to reflect the amount of information it contains; Calculate the objective weights: Calculate the difference coefficients of each indicator based on the information entropy, and then obtain the objective weight vector under the entropy weight method; output the first weight vector (denoted as W1).

[0041] Subjective weighting and consistency testing module of analytic hierarchy process Constructing the judgment matrix: Experts construct a pairwise comparison matrix based on the importance of the indicators; Weighting calculation unit: The subjective weight vector of each indicator is calculated using the geometric mean method; Consistency check unit: Calculate the consistency ratio CR; if CR≥0.1, the judgment matrix is ​​determined not to meet the consistency requirements, triggering a correction prompt or refusing to use the matrix; if CR<0.1, the subjective weight is accepted; output the second weight vector W2 and the consistency check result.

[0042] Comprehensive weight fusion module Set the fusion coefficient α∈[0,1] to adjust the ratio of objective weights to subjective weights; calculate the comprehensive weight vector: Output the final weight vector w for subsequent evaluation.

[0043] Ideal solution mapping and standardization Define the physical ideal solution and the physical anti-ideal solution: where the positive ideal solution is the optimal numerical combination in the physical sense of each index; and the negative ideal solution is the worst numerical combination in the physical sense of each index. Mapping unit: Maps the physical ideal solution and anti-ideal solution to the normalized space to obtain the normalized positive ideal solution and negative ideal solution; performs the same normalization process on the candidate solution set to generate the standardized decision matrix Z.

[0044] TOPSIS Compromise Solution Calculation Module The standardized matrix is ​​multiplied element-wise by the comprehensive weight vector to obtain the weighted matrix; Calculate the Euclidean distance from each candidate solution to the positive ideal solution and the Euclidean distance from each candidate solution to the negative ideal solution; then calculate the relative proximity. Select the candidate solution with the highest relative proximity as the compromise solution; output all performance parameters of this compromise solution and its TOPSIS score.

[0045] Example 2: Following the steps above, a Newman pseudo-two-dimensional electrochemical model was constructed in COMSOL multiphysics simulation software. Two side reactions—solid electrolyte interface (SEI) growth and lithium deposition—were embedded into the porous electrode reaction at the negative electrode and coupled with the solid heat transfer field. A five-stage constant current charging method was implemented by adding event functionality. The SOC operating range was limited to 10%–90%; the segmented charging method was implemented at 10%-30% SOC, 30%-50% SOC, and 50-70% SOC. The charging is divided into segments based on 70%-80% SOC and 80%-90% SOC. The charging rate is limited to the range of 0.5C to 3C, and the charging rate of each stage is arranged as follows: .

[0046] Finally, the output results are as follows: Figure 1The current-voltage change curve shown is shown.

[0047] Based on the optimal Latin hypercube design described above, 300 sample points are generated, which must strictly satisfy... The degree of improvement in the sample points provided by the optimized Latin hypercube design compared to the original Latin hypercube design can be determined. Improvement in φ_p value: 55.4% (from 18.9481 to 8.4430) Minimum distance improvement: 125.0% (from 0.0553 to 0.1244) Average distance improvement: 47.1% (from 1.2165 to 1.7897) Maximum improvement in blank margin: 39.1% (from 0.4842 to 0.2950) Range coverage improved by 11.9% (from 48.8% to 54.7%). Border coverage improvement: 350.0% (from 20.0% to 90.0%) The 300 sample points were then fed back into the electrochemical-thermal-aging coupled model constructed above to obtain the charging time for different five-stage constant current charging, the capacity retention after 1500 cycles, and the highest temperature during this process. Based on the process of constructing the multi-objective surrogate model above, after constructing the multi-objective surrogate model using Catboost, the following can be obtained: Figure 2 The accuracy evaluation criteria for the proxy model are shown.

[0048] Plot an Actual vs. Predicted scatter plot on the test set, overlaid with a red dashed line for y=x. The comparison of actual and predicted values ​​is shown below. Figure 3 As shown.

[0049] After training the multi-objective model Catboost, the SHAP framework based on Shapley additive interpretation is further introduced to quantify the three performance metrics of the battery. The global feature importance ranking can be obtained, and plotting it as a scatter plot yields a beehive graph, as shown below. Figure 4 , 5 As shown in Figure 6, the positive and negative effects of high / low eigenvalues ​​on charging time extension or capacity decay are visually demonstrated.

[0050] A multi-objective optimization using an improved MOAPSO is employed to obtain the Pareto solution set. The optimization objectives are to minimize charging time, maximize capacity retention, and minimize maximum temperature, yielding the Pareto front as follows: Figure 7 .

[0051] Using the above compromise solution based on subjectivity and objectivity, the weights obtained from the entropy weight method and the analytic hierarchy process (AHP) are linearly weighted to obtain the weights after integrating subjective and objective evaluations. The constructed AHP judgment matrix is ​​as follows: ].

[0052]

[0053] Where α and β are the weights corresponding to the entropy weight method and the analytic hierarchy process, respectively. Here, α = 0.5 and β = 0.5 are taken.

[0054] Finally, based on TOPSIS, each solution in the Pareto solution set is scored, and a compromise solution is selected.

[0055] like Figure 8 As shown, a compromise solution can be obtained through comprehensive scoring. On the Pareto front, the red dot represents the compromise solution selected using the subjective-objective fusion method.

[0056] Applying the above five-stage constant current charging method to the electrochemical-thermal coupled aging model, the simulation results are as follows: t=1311.7, capacity retention rate=88.642%, T max =30.017. Simultaneously, comparing it with 2C constant current charging, at t=1425, the capacity retention rate is 88.513%, T max =30.715, with improvements in all aspects of performance.

[0057] Alternatively, the decision matrix of the analytic hierarchy process can be constructed as follows: ],like Figure 9 As shown, a compromise solution can be obtained through comprehensive scoring: .

[0058] Applying the above five-stage constant current charging method to the electrochemical-thermal coupled aging model, the simulation results are: t=2934.1, capacity retention = 93.053%, T max =27.402. Comparing this to 0.5C constant current charging, t=5770.2, capacity retention = 93.396%, T max =27.562. As can be seen, the charging time was greatly shortened while the maximum temperature was also slightly reduced, even with a slight decrease in capacity retention rate, proving the feasibility of this method.

[0059] Example 3: A five-stage constant current battery charging optimization method based on an electrochemical-thermal coupled aging model includes the following steps: A Newman pseudo-two-dimensional electrochemical model was constructed in COMSOL multiphysics simulation software. Battery parameters were input, and the performance parameters of the battery model under five-stage constant current charging conditions were obtained. The optimal Latin hypercube sampling method was used to obtain the rate of each stage in the five-stage constant current charging process. The samples were then input into the Newman pseudo-two-dimensional electrochemical model to obtain the charging time, capacity retention after n cycles, and maximum temperature during the process. A database of the five-stage constant current charging method and multiple battery performance parameters under different conditions was constructed. Use Catboost to build a multi-objective agent model, train it, and obtain a joint prediction model; The TreeSH1AP algorithm is used to build an interpreter for the joint prediction model and to quantify and interpret the three performance indicators of the battery respectively. The joint prediction model uses an improved MOAPSO for multi-objective optimization to obtain the Pareto solution set; The Pareto solution set is scored using TOPSIS, and all performance parameters of the compromise solution and its TOPSIS score are output.

[0060] In conjunction with the first aspect, the present invention is further configured such that the five-stage constant current charging conditions are: 10%-30%SOC, 30%-50%SOC, 50-70%SOC, 70%-80%SOC, 80%-90%SOC, and the charging rate of each stage is 3C>R1>R2>R3>R4>R5>0.5C.

[0061] In conjunction with the first aspect, the present invention is further configured such that: the battery parameters include the SEI film formation reaction and the lithium deposition reaction, and a time acceleration factor is input into the stoichiometric coefficients; The local current density equation for the SEI film formation reaction is: ; In the formula, k 0,SEI It is the SEI reaction rate constant; U SEI It is the equilibrium potential for the SEI formation reaction, C S EC It is the concentration of EC on the graphite surface. R It is the gas constant. T It's temperature. F It is Faraday's constant, α c,SEI Where α is the cathode transfer coefficient and α is the active specific surface area. It is the electric potential of the solid phase (graphite). It is the potential of the electrolyte phase. j tot It is the total current density.R film It is the surface film resistance; The transfer current density of the lithium deposition reaction is: ; In the formula, i 0，lpl α is the exchange current density of lithium deposition, a is the specific surface area of ​​the electrode, and α is the specific current density of the electrode. c,lpl R is the cathode charge transfer coefficient of the lithium deposition reaction, T is the ideal gas constant, and T is the absolute temperature in Kelvin (K). It is the electric potential of the solid phase (graphite). It is the potential of the electrolyte phase, j tot It is the total current density, R film It is the surface film resistance.

[0062] The present invention also provides a device / equipment / system for optimizing five-stage constant current battery charging based on an electrochemical-thermal coupled aging model, comprising a memory, a processor, and a computer program stored in the memory, characterized in that the processor executes the computer program to implement the steps of the above method.

[0063] The present invention also provides a computer-readable storage medium having a computer program / instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0064] The present invention also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the above-described method.

[0065] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A five-stage constant current battery charging optimization method based on electrochemical-thermal coupled aging model, characterized by: It comprises the following steps: A Newman pseudo two-dimensional electrochemical model is constructed in COMSOL multi-physical field simulation software, battery parameters are input, and performance parameters of the battery model under five-stage constant current charging conditions are obtained; Charging mode samples are obtained by optimal Latin hypercube sampling, and are input into the Newman pseudo two-dimensional electrochemical model to obtain charging time, capacity retention rate after n cycles, and maximum temperature during the process, thereby constructing a database of multiple performances of the battery under different charging conditions; A multi-objective proxy model is constructed using Catboost, and the charging mode samples and the database of multiple performances of the battery are used for training and optimization to obtain a joint prediction model; A TreeSH1AP algorithm is used to establish an interpreter for the joint prediction model, and the three performance indicators of the battery are quantitatively explained respectively; A Pareto solution set is obtained by multi-objective optimization based on the joint prediction model and the improved MOAPSO; The comprehensive weight of each performance is obtained by combining the entropy weight method and the analytic hierarchy process, and each solution in the Pareto solution set is scored based on TOPSIS, and the overall performance parameters and the TOPSIS score of the compromise solution are output.

2. The battery charge optimization method of claim 1, wherein: The five-stage constant current charging conditions are: 10%-30% SOC, 30%-50% SOC, 50-70% SOC, 70%-80% SOC, and 80%-90% SOC, and the charging rate of each stage is 3C>R1>R2>R3>R4>R5>0.5C.

3. The five stage constant current battery charge optimization method of claim 1, wherein: The battery parameters include the generation reaction of the SEI film and the lithium deposition reaction, and the time acceleration factor is input in the stoichiometric coefficient; The local current density equation of the generation reaction of the SEI film is: ； wherein k 0,SEI is the SEI reaction rate constant; U SEI is the equilibrium potential of the SEI formation reaction, C S EC is the concentration of EC at the graphite surface, R is the gas constant, T is the temperature, F is the Faraday constant, is the cathode transfer coefficient, a is the active specific surface area, is the potential of the solid phase (graphite), is the potential of the electrolyte phase, j tot is the total current density, R film is the surface film resistance; The transfer current density of the lithium deposition reaction is: ； wherein i 0，lpl is the exchange current density for lithium deposition, a is the specific surface area of the electrode, a c,lpl is the cathodic charge transfer coefficient for the lithium deposition reaction, R is the ideal gas constant, T: absolute temperature in Kelvin (K), is the potential of the solid phase (graphite), is the potential of the electrolyte phase, j tot is the total current density, R film is the surface film resistance.

4. The battery charge optimization method of claim 1, wherein: The multi-objective proxy model is constructed using Catboost, trained, and a joint prediction model is obtained, which comprises: dividing the training sample set into a training set and a test set, performing multi-objective robust normalization on the training set to obtain a normalized training set, constructing a five-fold cross-validation to obtain optimal hyperparameters, setting the number of iterations using the optimal hyperparameters, and introducing an early stopping mechanism; if the MultiRMSE of the validation set does not decrease for 200 consecutive rounds, stop training; use the normalized training set as input to retrain the CatBoost multi-output regressor to obtain the joint prediction model.

5. The battery charge optimization method of claim 1, wherein: The multi-objective optimization is to minimize the charging time, maximize the capacity retention rate, and minimize the maximum temperature.

6. The battery charge optimization method of claim 1, wherein: The scoring of each solution in the Pareto solution set based on TOPSIS comprises: The standardized matrix is multiplied by the comprehensive weight vector element by element to obtain a weighted matrix; The Euclidean distance of each candidate solution to the positive ideal solution is calculated, the Euclidean distance of each candidate solution to the negative ideal solution is calculated, and the relative closeness is calculated. The candidate solution with the maximum relative closeness is selected as the compromise solution.

7. The battery charge optimization method of claim 6, wherein: The comprehensive weight vector calculation method is: ； In the formula, a is a fusion coefficient, W1 is a first weight vector, and W2 is a second weight vector.

8. An apparatus / device / system for five-stage constant current battery charging optimization based on electro-thermal coupled aging model, comprising a memory, a processor and a computer program stored on the memory, characterized in that, The processor executes the computer program to implement the steps of the method of any one of claims 1-7.

9. A computer readable storage medium having stored thereon computer programs / instructions, characterized in that, The computer program / instructions, which when executed by the processor, implement the steps of the method of any one of claims 1-7.

10. A computer program product comprising computer programs / instructions, characterized in that, The computer program / instructions, which when executed by the processor, implement the steps of the method of any one of claims 1-7.

Images