A method for predicting the state of hazardous waste (SOH) of lithium batteries based on Kirchhoff optimization algorithm-optimized KAN network

Abstract

This invention relates to a lithium battery state-of-the-art (SOH) prediction method based on Kirchhoff's algorithm (KOA) optimization of a KAN network. The invention involves acquiring raw operating data of the lithium battery; preprocessing the raw operating data to obtain standardized feature data; dividing the standardized feature data into training, validation, and test sets; initializing the core hyperparameters of the KAN network; adaptively optimizing the core hyperparameters using the Kirchhoff optimization algorithm (KOA) to generate the optimal hyperparameter combination; constructing the optimized KAN network based on the optimal hyperparameter combination; inputting the test set into the trained model to obtain the SOH prediction result; and outputting the prediction result. This invention avoids the subjectivity of manually setting hyperparameters based on experience, and improves the accuracy and efficiency of hyperparameter optimization through the synergistic effect of current allocation, voltage balancing, and mutation strategies.

Description

Technical Field

[0001] This invention relates to the field of lithium battery state detection technology, and is a lithium battery SOH prediction method based on Kirchhoff optimization algorithm to optimize KAN network. Background Technology

[0002] Lithium batteries are widely used in new energy vehicles, energy storage systems, portable electronic devices and other fields due to their advantages such as high energy density, long cycle life and low self-discharge rate.

[0003] State of Harm (SOH) is a core indicator for measuring the remaining capacity and performance degradation of lithium batteries. Its accurate prediction is of great significance for ensuring the safe operation of equipment and optimizing battery management strategies.

[0004] Existing methods for predicting the state of harm (SOH) of lithium batteries mainly include model-driven methods, data-driven methods, and hybrid methods.

[0005] Model-driven methods build models based on the internal electrochemical principles of batteries, but they suffer from problems such as complex parameter identification and poor adaptability to complex operating conditions. Traditional data-driven methods, such as neural networks and support vector machines, rely on human experience to set model hyperparameters, which can easily lead to insufficient prediction accuracy or overfitting.

[0006] In practical applications, lithium batteries operate in environments characterized by variable operating conditions, strong noise interference, and complex aging mechanisms. Existing SOH prediction models are prone to increased prediction errors and insufficient generalization ability under different temperatures, rates, and cycling stages, resulting in deviations between SOH estimation results and actual degradation trends, making it difficult to meet the requirements of high-precision battery health management.

[0007] As a novel type of neural network, KAN networks fit complex nonlinear mapping relationships through kernel functions and have strong feature learning capabilities. However, the selection of hyperparameters (kernel function type, number of network layers, number of neurons, etc.) directly affects the model performance. If the hyperparameters are not configured properly, it can easily lead to large prediction errors or unstable convergence.

[0008] Existing hyperparameter optimization algorithms, such as genetic algorithms and particle swarm optimization algorithms, suffer from insufficient balance between global search and local exploitation, and are prone to getting trapped in local optima. As a result, the obtained hyperparameter combinations are difficult to adapt to the nonlinear variation characteristics of SOH under complex working conditions, thus limiting the improvement of prediction accuracy of KAN networks.

[0009] Kirchhoff optimization algorithms are designed based on Kirchhoff's current law and voltage law in circuits to find optimal mechanisms. They have the characteristics of strong collaborative search ability and fast convergence speed, providing a new approach to solving the hyperparameter optimization problem of KAN networks. Summary of the Invention

[0010] This invention improves the accuracy and stability of SOH prediction by adaptively optimizing the hyperparameters of the KAN network using the KOA algorithm, and solves the problems of hyperparameter dependence on experience and insufficient performance of optimization algorithms in traditional methods. It discloses a lithium battery SOH prediction method based on Kirchhoff optimization algorithm to optimize the KAN network.

[0011] This invention provides the following technical solutions: A method for predicting the state of harm (SOH) of lithium batteries based on Kirchhoff's optimization algorithm and KAN network, the method comprising the following steps: Step 1: Obtain the raw operating data of the lithium battery; Step 2: Preprocess the raw operating data of the lithium battery to obtain standardized feature data; Step 3: Divide the standardized feature data into training set, validation set, and test set; Step 4: Initialize the core hyperparameters of the KAN network, apply Kirchhoff's optimization algorithm (KOA) to adaptively optimize the core hyperparameters, and generate the optimal hyperparameter combination; Step 5: Construct the optimized KAN network based on the best hyperparameter combination, train the model using the training set, and adjust the generalization ability using the validation set; Step 6: Input the test set into the trained model to obtain the SOH prediction results, calculate the evaluation index to assess the accuracy, and output the prediction results.

[0012] Preferably, the adaptive search step size decreases non-linearly with the iteration process, as shown in the following formula:

[0013] in, For the first The search step size for the next iteration. The initial step size, The attenuation coefficient is... This represents the current iteration number. This represents the maximum number of iterations.

[0014] Preferably, the current allocation mechanism is used to guide individuals in the population to gather in high-quality areas, and the current weight is defined as:

[0015] in, For the first Individual in the first Current weights in the next iteration This represents the individual fitness value. For population size, To prevent division by zero of constants; Update individual positions based on current weights:

[0016] in, As a learning factor, For the first The position of the optimal individual.

[0017] Preferably, the voltage balance mechanism is used to maintain the cooperative search capability among individuals in the population. After the current allocation is updated, a population equilibrium perturbation is introduced, and the individual position update formula is as follows:

[0018] in, As the voltage balance factor, The numbers are uniformly distributed random numbers between 0 and 1. The location of an individual randomly selected from the population.

[0019] Preferably, the population diversity maintenance strategy involves resetting the population by mutation of individuals with poor fitness; let the first... The average fitness of the generation population is:

[0020] When an individual satisfies:

[0021] Then, it is mutated and reset, and the mutation formula is as follows:

[0022] in, , The first The lower and upper limits of the search range for each hyperparameter.

[0023] Preferably, the process of determining the optimal hyperparameter combination of the KAN network using the Kirchhoff optimization algorithm includes the following steps: Step S1: Set the population size, maximum number of iterations, and hyperparameter search range for KOA. Let the population size be... The maximum number of iterations is The KAN hyperparameter set is as follows:

[0024] The search scope is: ; Step S2: Initialize the population individuals, let the first individual be... The individual positions are: ,

[0025] Calculate individual fitness:

[0026] in, For true SOH, For KAN predicted values, This represents the number of training samples; Step S3: According to Kirchhoff's laws, update the population position through current distribution and voltage balance mechanisms, based on the adaptive step size:

[0027] Calculate the current weight:

[0028] Perform current allocation update:

[0029] Perform voltage balance update:

[0030] Step S4: Implement diversity preservation strategies for individuals with low fitness, when Perform mutation reset on the individual:

[0031] Step S5: Determine whether the maximum number of iterations or the precision threshold has been reached. If satisfied: or

[0032] Then the iteration terminates; Step S6: If the termination condition is met, output the optimal hyperparameter combination:

[0033] Otherwise Return to step 2 and repeat the iteration.

[0034] Preferably, the updated hyperparameters are subjected to boundary constraints, and the boundary constraint formula is as follows: .

[0035] Preferably, the preprocessing includes missing value imputation using linear interpolation and outlier removal using... Principles and data standardization are adopted. Normalization and key feature extraction: extract voltage peak value, current slope, and temperature change rate.

[0036] A computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement a lithium battery SOH prediction method based on Kirchhoff optimization algorithm to optimize a KAN network.

[0037] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement a lithium battery SOH prediction method based on Kirchhoff optimization algorithm to optimize KAN network.

[0038] The present invention has the following beneficial effects: This invention employs Kirchhoff optimization algorithm to optimize the hyperparameters of the KAN network, avoiding the subjectivity of setting hyperparameters based on human experience. Through the synergistic effect of current distribution, voltage balancing, and mutation strategies, the accuracy and efficiency of hyperparameter optimization are improved.

[0039] The data preprocessing step of this invention combines feature engineering to extract key time-series features, reducing the interference of redundant information on the model and enhancing the model's ability to capture the performance degradation law of lithium batteries.

[0040] This invention constructs a closed-loop mechanism for hyperparameter optimization, model training, and accuracy verification, ensuring that the prediction results meet the accuracy requirements of practical applications and providing reliable technical support for lithium battery SOH monitoring. Attached Figure Description

[0041] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0042] Figure 1 This is shown as the overall process of the present invention; Figure 2 The diagram shows the optimization process of this invention. Figure 3 The image shown is a prediction result graph of the present invention. Figure 4 The image shown is a comparison chart of the accuracy of this invention. Figure 5 The diagram shows the technical effects of this invention. Figure 6 The diagram shown is a comparative illustration of the present invention. Detailed Implementation

[0043] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. 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.

[0044] The present invention will be described in detail below with reference to specific embodiments. Specific Implementation Example 1: according to Figures 1 to 6 As shown, the specific optimization technical solution adopted by the present invention to solve the above-mentioned technical problems is: The present invention relates to a lithium battery SOH prediction method based on Kirchhoff optimization algorithm to optimize KAN network.

[0046] This invention relates to the field of lithium battery state detection technology, aiming to solve the problem of large errors in current lithium battery health state prediction technologies and provide a prediction scheme with small errors and high accuracy. This invention discloses a lithium battery SOH (State of Health) prediction method based on Kirchhoff optimization algorithm-optimized KAN (Kolmogorov-Arnold Network) network, comprising the following steps: First, acquiring and preprocessing lithium battery operating data, then dividing the processed data into training, validation, and test sets; second, initializing the parameters of the KAN network to construct a basic KAN prediction model; third, using the mean squared error of prediction (MSE) as the fitness function, optimizing the parameters of the KAN network using the Kirchhoff optimization algorithm (KOA); next, substituting the optimized parameters into the KAN network to construct the optimized KAN prediction model and conducting training; finally, using the optimized model to predict the lithium battery SOH and verifying its performance. This invention introduces Kirchhoff optimization algorithm to adaptively determine the key parameters of KAN network (attention adjustment parameters, number of hidden layer nodes, inter-layer connection weights, etc.), avoiding problems such as local optima and poor generalization ability caused by manual parameter adjustment. By combining KOA and KAN, the nonlinear fitting advantage of KAN network and the parameter optimization ability of KOA algorithm are fully utilized, achieving high-precision and fast prediction of SOH of lithium battery.

[0047] This invention provides a method for predicting the state of hazardous waste (SOH) of lithium batteries based on Kirchhoff optimization algorithm-optimized KAN networks. The method includes the following steps: A method for predicting the state of harm (SOH) of lithium batteries based on Kirchhoff optimization algorithm-optimized KAN network includes the following steps: S1. Obtain raw operating data of the lithium battery; This invention uses a ternary lithium battery cycle dataset from a new energy laboratory. The tested lithium battery model is INR18650-2000 (cylindrical ternary lithium-ion battery). The charge-discharge cycle test was conducted in a constant temperature environment of 25℃±2℃, with a data sampling frequency of 1kHz. The nominal capacity of the lithium battery was 2Ah, and the nominal voltage was 3.7V. The total number of cycles was 1000 to simulate the performance degradation process of the lithium battery throughout its entire life cycle. Table 1 shows the detailed parameters of the tested lithium battery.

[0048] Table 1 Parameters of the tested lithium battery

[0049] In this step, raw operating data is simultaneously collected using voltage, current, and temperature sensors. The collected data includes voltage (V), current (A, positive for charging, negative for discharging), and temperature (°C) values ​​every 1ms, as well as the current cycle number. The collected data is stored in CSV format by cycle period, with each cycle generating a separate data file containing five columns: timestamp (ms), voltage (V), current (A), temperature (°C), and cycle number, as shown in Figure 5. Figure 6 shows the time-series curve of the raw voltage signal after preprocessing for the 500th cycle. The characteristics of voltage increase during charging and voltage decrease during discharging can be visually observed, providing a data foundation for subsequent feature extraction.

[0050] The state of harmonics (SOH) of lithium batteries is strongly correlated with the number of cycles. The core correlation formula is derived based on the capacity decay law as follows:

[0051] in, This indicates the actual capacity of the lithium battery during its current cycle (unit: Ah). Indicates the initial nominal capacity of the lithium battery (in this implementation case). Based on the cyclic test data in Table 1, the time at the 100th cycle was measured using the constant current discharge method. Substituting into the formula, we get On the 500th iteration Substituting into the formula, we get This provides a basis for subsequent model label construction.

[0052] S2. Preprocess the raw operating data to obtain standardized feature data; To eliminate noise and format differences in the original data and ensure the quality of the model input, this step employs a four-step processing flow: "missing value imputation - outlier removal - data standardization - feature extraction." The operational logic and parameter settings for each step are as follows: Missing value imputation: For missing values ​​in the original temperature data caused by momentary sensor interruption (missing rate approximately 0.3%), linear interpolation is used to imput them. The interpolation formula is as follows:

[0053] in, For missing moments Temperature fill value, , These are the effective temperature values ​​adjacent to the missing time point, respectively. , These are the timestamps (in milliseconds) corresponding to the valid temperature values.

[0054] Outlier removal: based on The principle is to perform anomaly detection on voltage and current data. The anomaly determination formula is as follows:

[0055] in, For the original data points, This is the mean of the feature data. This represents the standard deviation of the characteristic data. In this implementation example, the voltage data... , Therefore, it exceeds Voltage values ​​within the specified range were identified as outliers; after screening, the proportion of outliers was approximately [percentage missing]. .

[0056] Data standardization: Min-Max normalization is used to map voltage, current, and temperature data to... For intervals, eliminating dimensional differences, the normalization formula is as follows:

[0057] in, For standardized data, The original data, , These are the minimum and maximum values ​​of the feature data, respectively. Taking voltage as an example, , Substituting these values ​​into the formula will standardize the voltage values.

[0058] Feature Extraction: Focusing on key characteristics of lithium battery performance degradation, four types of time-series features are extracted: voltage peak (maximum value during charging). Minimum value during discharge phase ), current slope (mean of absolute value of first derivative of current) ), rate of change of temperature (mean absolute value of the first derivative of temperature) After feature calculation, a 5-dimensional structure is formed. , , , The eigenvectors (number of cycles) and the preprocessed voltage timing curves are shown in Figure 6, which clearly show the characteristic differences in the charging and discharging stages.

[0059] S3. Divide the standardized feature data into training set, validation set and test set; To ensure the reliability of model training effectiveness and generalization ability evaluation, a stratified sampling method is adopted, i.e., the ratio of training set samples to validation set samples to test set samples is 7:1:2. In this implementation, the total number of samples is 1000 (corresponding to 1000 iterations). Substituting these into the formula yields a training set of 700 samples, a validation set of 100 samples, and a test set of 200 samples. During the partitioning process, a random seed (random_state=42) is set to ensure consistency in each partition and avoid model evaluation bias due to differences in data distribution. The training set is used for KAN network parameter learning, the validation set is used to adjust model regularization parameters to avoid overfitting, and the test set is used to verify the final SOH prediction accuracy. The dataset partitioning results can be further verified using a feature distribution histogram to ensure that the feature distribution of each subset is consistent with the total dataset.

[0060] S4. Initialize the core hyperparameters of the KAN network, and apply Kirchhoff's optimization algorithm (KOA) to adaptively optimize the core hyperparameters to generate the optimal hyperparameter combination; The KOA algorithm simulates the synergistic effect of Kirchhoff's Current Law (KCL) and Voltage Law (KVL) in a circuit. Each hyperparameter combination is treated as a "circuit node." The search direction is guided by current allocation, and node positions are updated by voltage balance, achieving adaptive optimization of the KAN network hyperparameters. The specific process and formulas are as follows: Hyperparameter initialization: The core hyperparameters of the KAN network include the kernel function type (candidate set: Gaussian kernel). polynomial kernel Sigmoid core The network consists of L (3-8 layers) and N (64-256 neurons). During initialization, 30 candidate hyperparameter combinations are generated by random sampling (i.e., the KOA population size). ).

[0061] Fitness calculation: The mean squared error (MSE) of the KAN network's predictions on the training set is used as the fitness value of the KOA algorithm. The MSE calculation formula is as follows:

[0062] in, The number of samples in the training set. Predict the SOH value for the KAN network. The value represents the true SOH value. A smaller fitness value indicates a better combination of hyperparameters.

[0063] Current allocation guidance (based on KCL): The current allocation ratio is calculated based on the fitness values ​​of each candidate hyperparameter combination, as shown in the following formula:

[0064] in, For the first The candidate hyperparameter combinations in the th... Current weights in the next iteration, For the first The fitness value of each combination (e.g., RMSE of the validation set). For population size, To be a very small positive number to avoid the denominator being 0 , The larger the value, the higher the priority of the combination in attracting the population.

[0065] Voltage balance update (based on KVL): Candidate hyperparameter combinations (node ​​locations) are updated using an adaptive weighting factor, as shown in the following formula: (Current distribution updated, optimized development)

[0066] (Voltage balance update, collaborative exploration)

[0067] in, For the first After the nth iteration The positions of each combination (hyperparameter vector). For the first Position at the next iteration Let the position be the globally optimal combination in the t-th iteration. As a learning factor, For voltage balance factor, A random number between 0 and 1. This represents the position of an individual randomly selected from the population. The adaptive step size decreases with each iteration.

[0068] Parameters were adjusted for the bottom 30% of candidate combinations based on fitness ranking. This adjustment aimed to maintain population diversity and prevent the optimization process from getting stuck in local optima. The parameter adjustment was calculated using the following method: Perform a random reset on the bottom 30% of individuals based on fitness ranking:

[0069] Boundary constraints to prevent out-of-bounds access:

[0070] Set the optimization stopping condition: when the number of parameter adjustments reaches the preset maximum number of times. Or the mean square error corresponding to the globally optimal parameters ( )satisfy When the parameter adjustment reaches the 85th iteration, the hyperparameter optimization is stopped. In this implementation case, the stopping condition is met when the parameter adjustment is performed for the 85th time. At this point, the optimal hyperparameter combination output is "Gaussian kernel function + 5-layer network + 128 neurons". The convergence curve of the KOA algorithm is shown in Figure 2. It can be clearly observed from the curve that as the number of parameter adjustments increases, the fitness value gradually decreases and eventually tends to stabilize, indicating that the optimization process is effective and the parameter combination has reached the ideal state.

[0071] S5. Construct an optimized KAN network based on the best hyperparameter combination, train the model using the training set, and adjust the generalization ability using the validation set. The KAN network is constructed based on the optimal hyperparameter combination of S4 output, and adopts a three-layer architecture of "input layer - at least one hidden layer - output layer". The input layer is used to receive multidimensional feature vectors, and the output layer is used to output the target prediction value. Unlike the connection method of "scalar weight matrix + fixed activation function" used by traditional neural networks, the KAN network introduces a learnable one-dimensional function as the connection weight on each connection between adjacent neurons. This allows the input features to be mapped by the corresponding one-dimensional function before being passed to the next layer. The outputs of each connection function are summed and aggregated at the target neuron to form the output of that neuron. In this way, the combination of multiple one-dimensional function connections can approximate multidimensional nonlinear mapping relationships, thereby improving the network's ability to express complex nonlinear feature relationships.

[0072] The training process incorporates a validation set to adjust generalization capability, with the following specific settings: Network structure: Number of neurons in the input layer (Corresponding to 5 features), the hidden layer has 5 layers and the number of neurons in each layer is... (Using Gaussian kernel function, kernel function formula) Number of neurons in the output layer (Output SOH prediction value).

[0073] Training parameters: Adam optimizer (learning rate) Momentum parameters , MSE loss function, batch size Maximum number of training rounds .

[0074] Generalization ability adjustment: An early stopping strategy is adopted. Training is stopped when the validation set MSE no longer decreases for 5 consecutive epochs to avoid overfitting. In this implementation, training stops when the validation set MSE drops to 0.0006 and remains stable for 5 consecutive epochs after the 60th epoch. At this point, the model's performance on the validation set is satisfactory. The loss curve during the model training process can be observed with the help of Figure 2. Figure 3 shows the comparison of the MSE of the training set and the validation set with the changes of the rounds, which can intuitively determine whether the model is overfitting.

[0075] S6. Input the test set into the trained model to obtain the SOH prediction results, calculate the evaluation index to assess the accuracy, and output the prediction results that meet the accuracy requirements. The test set of 200 samples is input into the KAN network trained on S5 to obtain the SOH prediction value. The prediction accuracy is evaluated using three metrics: mean squared error (MSE), mean absolute error (MAE), and coefficient of determination (R²). in, The number of samples in the test set. This represents the mean of the true SOH values ​​in the test set.

[0076] Accuracy determination: based on preset standards ( and In this implementation case The accuracy requirements are met, and the final SOH prediction result is output. Figure 3 shows a comparison between the predicted and actual values ​​on the test set, indicating a high degree of fit. To further verify the superiority of the Kirchhoff optimization algorithm (KOA) used in this invention for hyperparameter optimization of KAN networks, the lithium battery SOH prediction accuracy (MSE, R²) and algorithm convergence time were used as evaluation indicators. Comparative tests were conducted with the Jackal Optimization Algorithm (COA) and the Alpha Evolutionary Algorithm (αEA), with the hyperparameter settings of the three algorithms kept consistent (population size 30, maximum number of iterations 100, and the same hyperparameter search range). The comparison results are as follows: From the perspective of SOH prediction accuracy, the three optimized KAN networks were used to make predictions on the test set (200 samples), and the calculated results were... and As shown in Table 2. Among them, the KOA-KAN model... Minimum (0.0008) Maximum (0.96), compared to the COA-KAN model. Decreased by 18.4% An improvement of 0.02, compared to the αEA-KAN model. Reduced by 27.3%, The improvement of 0.03 indicates that the optimized hyperparameter combination of the KOA algorithm can better leverage the nonlinear fitting ability of the KAN network and more accurately capture the SOH decay law of lithium batteries.

[0077] Table 2 Comparison of SOH prediction accuracy of KAN network for different optimization algorithms

[0078] From the perspective of algorithm convergence performance, the convergence time (total time from initializing the population to meeting the termination condition) required for the three algorithms to complete the hyperparameter optimization of the KAN network is shown in Table 3. The KOA algorithm has the shortest average convergence time, with a time of 152s in the hyperparameter optimization phase of the training set, which is 23.2% shorter than the COA algorithm (198s) and 32.4% shorter than the αEA algorithm (225s). Even in the complete optimization process including model training, the total time of the KOA algorithm is still lower than the other two algorithms. This is due to the cooperative search mechanism of KOA based on Kirchhoff's laws, which can quickly guide the population to gather towards the optimal hyperparameter combination and reduce ineffective iterations.

[0079] Table 3. Comparison of the time taken by different optimization algorithms to complete the hyperparameter optimization of the KAN network (unit: s)

[0080] In summary, the KOA algorithm possesses the dual advantages of "high accuracy" and "fast convergence" in the hyperparameter optimization of KAN networks for lithium battery SOH prediction: on the one hand, it achieves accurate hyperparameter optimization through current allocation and voltage balancing mechanisms, thereby improving the accuracy of SOH prediction; on the other hand, it reduces redundant iterations and shortens the optimization time through adaptive weights and mutation strategies, making it more suitable for model optimization scenarios for lithium battery SOH prediction than the Jackal Optimization and Alpha Evolution algorithms.

[0081] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement a lithium battery SOH prediction method based on Kirchhoff optimization algorithm to optimize KAN network.

[0082] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement a lithium battery SOH prediction method based on Kirchhoff optimization algorithm to optimize KAN network.

[0083] Data preprocessing mechanism: To address the issues of missing values, outliers, and dimensional differences in the original operating data of lithium batteries, linear interpolation is used to fill in missing values, outliers are removed based on the 3σ principle, data standardization is achieved through Min-Max normalization, and time-series features such as voltage peak value, current slope, and temperature change rate are extracted to provide a high-quality data foundation for model input.

[0084] The hyperparameter optimization mechanism of Kirchhoff's optimization algorithm is as follows: a population guidance mechanism is designed by simulating Kirchhoff's current law, which determines the guiding priority of high-quality individuals to the population through current allocation values, ensuring that the algorithm searches towards the optimal solution; an individual position update mechanism is constructed based on Kirchhoff's voltage law, which combines adaptive weight factors to dynamically balance global exploration and local exploitation capabilities, gradually reducing the global search intensity and improving the local search accuracy as the iteration process progresses; a mutation strategy is implemented for low-fitness individuals, which maintains population diversity through random perturbation and avoids the algorithm from getting trapped in local optima.

[0085] KAN network modeling and accuracy verification mechanism: A KAN network is constructed based on the optimal hyperparameter combination output by the KOA algorithm. The model is trained using the training set and its generalization ability is adjusted using the validation set. Mean squared error (MSE) is employed. Mean absolute error ( ) and coefficient of determination ( As an evaluation indicator, when it meets the following conditions... and If the model accuracy meets the standard, the prediction result is output; otherwise, the process returns to the hyperparameter optimization step to re-optimize.

[0086] The above description is merely a preferred embodiment of a lithium battery SOH prediction method based on Kirchhoff's optimization algorithm to optimize a KAN network. The scope of protection for this lithium battery SOH prediction method based on Kirchhoff's optimization algorithm to optimize a KAN network is not limited to the above embodiments. All technical solutions falling within this conceptual framework are within the scope of protection of this invention. It should be noted that for those skilled in the art, any improvements and variations made without departing from the principles of this invention should also be considered within the scope of protection of this invention.

Claims

1. A lithium battery SOH prediction method based on Kirchhoff optimization algorithm to optimize KAN network, characterized by: The method includes the following steps: Step 1: Obtain the raw operating data of the lithium battery; Step 2: Preprocess the raw operating data of the lithium battery to obtain standardized feature data; Step 3: Divide the standardized feature data into training set, validation set, and test set; Step 4: Initialize the core hyperparameters of the KAN network, apply Kirchhoff's optimization algorithm (KOA) to adaptively optimize the core hyperparameters, and generate the optimal hyperparameter combination; Step 5: Construct the optimized KAN network based on the best hyperparameter combination, train the model using the training set, and adjust the generalization ability using the validation set; Step 6: Input the test set into the trained model to obtain the SOH prediction results, calculate the evaluation index to assess the accuracy, and output the prediction results.

2. The method according to claim 1, characterized in that: The adaptive search step size decreases non-linearly with the iteration process, as shown in the following formula: in, For the first The search step size for the next iteration. The initial step size, The attenuation coefficient is... This represents the current iteration number. This represents the maximum number of iterations.

3. The method according to claim 2, characterized in that: The current allocation mechanism is used to guide individuals in the population to gather in high-quality areas. The current weight is defined as follows: in, For the first Individual in the first Current weights in the next iteration This represents the individual fitness value. For population size, To prevent division by zero of constants; Update individual positions based on current weights: in, As a learning factor, For the first The position of the optimal individual.

4. The method according to claim 3, characterized in that: The voltage balance mechanism is used to maintain the cooperative search ability among individuals in the population. After the current allocation is updated, a population equilibrium perturbation is introduced. The individual position update formula is as follows: in, As the voltage balance factor, The numbers are uniformly distributed random numbers between 0 and 1. The location of an individual randomly selected from the population.

5. The method according to claim 4, characterized in that: Population diversity maintenance strategies involve resetting individuals with poor fitness through mutation; assuming the first... The average fitness of the generation population is: When an individual satisfies: Then, it is mutated and reset, and the mutation formula is as follows: in, , The first The lower and upper limits of the search range for each hyperparameter.

6. The method according to claim 5, characterized in that: The process of determining the optimal hyperparameter combination of a KAN network using Kirchhoff's optimization algorithm includes the following steps: Step S1: Set the population size, maximum number of iterations, and hyperparameter search range for KOA. Let the population size be... The maximum number of iterations is The KAN hyperparameter set is as follows: The search scope is: ； Step S2: Initialize the population individuals, let the first individual be... The individual positions are: ， Calculate individual fitness: in, For true SOH, For KAN predicted values, This represents the number of training samples; Step S3: According to Kirchhoff's laws, update the population position through current distribution and voltage balance mechanisms, based on the adaptive step size: Calculate the current weight: Perform current allocation update: Perform voltage balance update: Step S4: Implement diversity preservation strategies for individuals with low fitness, when Perform mutation reset on the individual: Step S5: Determine whether the maximum number of iterations or the precision threshold has been reached. If satisfied: or Then the iteration terminates; Step S6: If the termination condition is met, output the optimal hyperparameter combination: Otherwise Return to step 2 and repeat the iteration.

7. The method according to claim 6, characterized in that: The updated hyperparameters are then subjected to boundary constraints, and the boundary constraint formulas are as follows: 。 8. The method according to claim 7, characterized in that: Preprocessing includes missing value imputation using linear interpolation and outlier removal using... Principles and data standardization are adopted. Normalization and key feature extraction: extract voltage peak value, current slope, and temperature change rate.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the method as claimed in any one of claims 1-8.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the method of any one of claims 1-8.

Images