Lithium battery health state prediction method combining BWO optimization and GPR

Abstract

The invention discloses a lithium battery health state prediction method combining BWO optimization and GPR, and belongs to the field of battery health monitoring. According to the method, through combination of optimization of a whale optimization algorithm and a Gaussian process regression model, the accuracy and robustness of prediction are further improved, input noise and complex data can be effectively processed, the generalization ability and prediction precision of the model are remarkably improved through a multi-model integration and secondary learning mode, and the method is suitable for various application scenes and has wide application prospects. The invention particularly relates to a lithium battery management system in a mobile energy storage power supply.

Description

Technical Field

[0001] The present invention relates to the technical field of battery health monitoring, and in particular to a lithium battery health status prediction method combining BWO optimization and GPR. Background Art

[0002] Lithium batteries have been widely used in electric vehicles, mobile devices, and energy storage systems due to their high energy density, long life, and good environmental adaptability. In order to ensure the safe use of batteries and extend their life, accurate estimation of the battery's state of health (SOH) has become one of the key tasks in the battery management system (BMS). At present, commonly used SOH estimation methods include prediction methods based on empirical models, equivalent circuit models, and data-driven models.

[0003] Methods based on empirical models and equivalent circuit models need to rely on a large amount of experimental data and complex electrochemical models, which makes it difficult to adapt to the battery aging behavior under different working conditions in practical applications. In contrast, data-driven models have gradually become the mainstream research direction due to their advantages of not relying on physical models and being able to handle complex nonlinear relationships. Gaussian process regression (GPR), as a common machine learning model, is widely used in SOH estimation because it can handle small sample data and has high prediction accuracy. However, traditional GPR models are prone to reduced prediction accuracy when faced with complex input noise. In addition, the hyperparameter selection of the model in the prior art usually relies on manual parameter adjustment or simple search algorithms, and cannot fully utilize the optimization algorithm to improve the model performance. Some researchers have proposed a GPR model based on genetic algorithm optimization to improve the accuracy of SOH estimation. However, the optimization methods in the prior art still have the problems of insufficient generalization ability and poor adaptability when processing complex battery data. Summary of the invention

[0004] The present invention provides a lithium battery health status prediction method combining BWO optimization and GPR, provides an efficient and reliable SOH prediction method, and provides strong support for the safe operation and extended service life of lithium batteries.

[0005] The embodiment of the present invention provides a lithium battery health status prediction method combining BWO optimization and GPR, comprising the following steps:

[0006] Step 1, obtaining historical charge and discharge data of the lithium battery, and preprocessing the historical charge and discharge data;

[0007] Step 2, randomly generate an initial beluga population, each beluga individual corresponds to a set of hyperparameter combinations of the GPR model;

[0008] Step 3: Use the BWO algorithm to optimize the key hyperparameters of the GPR model and maximize the model performance by combining the global search mechanism and the random search mechanism.

[0009] Step 4: Based on the optimized key hyperparameters, train the GPR model and the BWO-GPR model, use the predicted outputs of the GPR model and the BWO-GPR model as input features, train the second-layer integrated model, and use the GPR regression model in the second-layer model to integrate the predicted results of the BWO-GPR model and the GPR model to form the final stacked integrated model, and use the stacked integrated model to predict the health status of the lithium battery.

[0010] Optionally, in one embodiment of the present invention, step 1 specifically includes:

[0011] Obtain historical charging and discharging data of lithium batteries, including historical voltage and current operation data;

[0012] The historical charge and discharge data are cleaned, normalized and feature extracted, outliers and duplicate data in the historical charge and discharge data are removed, missing values ​​are supplemented, features of different magnitudes are converted into the same scale, and healthy features are extracted therefrom: fixed voltage rise time, constant current charging time and discharge efficiency. The fixed voltage rise time refers to the time required for the lithium battery to rise from one fixed voltage to another fixed voltage during the charging process; the constant current charging time refers to the time required for the lithium battery to enter the constant current stage during the charging process; the discharge efficiency refers to the ratio of the effective energy that the lithium battery can release during the discharge process to its nominal capacity.

[0013] Optionally, in one embodiment of the present invention, step 2 specifically includes:

[0014] The initial beluga population is randomly generated, and each beluga individual X i Represents a group Each beluga individual corresponds to a set of hyperparameter combinations of the GPR model, namely a set of length scale l and signal variance

[0015] In the GPR model, the kernel function is used to define the similarity between data points, where the expression of the radial basis kernel function is:

[0016]

[0017] Among them, x i is the i-th data point, x j is the jth data point.

[0018] Optionally, in one embodiment of the present invention, step 3 specifically includes:

[0019] (1) Fitness evaluation: Calculate the fitness value of each beluga individual and measure the model performance by maximizing the log-marginal likelihood function;

[0020] (2) Position update: The standard white whale predation behavior and Levy flight random search in the BWO algorithm are used to update the position of the white whale, so that the fitness value of the new position gradually increases. The specific position update formula is:

[0021] X i (t+1)=X i (t)+α(X best (t)-X i (t))+β·Levy(λ)

[0022] Among them, X i (t) is the position of the i-th white whale at the t-th iteration, that is, the current hyperparameter combination; X best (t) is the current global optimal hyperparameter combination; α and β are parameters for controlling the step size; Levy (λ) is the Levy flight function used for random search;

[0023] (3) Fitness evaluation and selection: For the newly generated positions, i.e., the new hyperparameter combinations, the fitness values ​​are calculated, and the beluga individuals with higher fitness values ​​are retained to enter the next generation of iterations;

[0024] (4) Iterate until convergence: Continue iterating until the predetermined maximum number of iterations T is reached max Or the change in fitness value is lower than the preset threshold, and finally the optimal hyperparameter combination is obtained;

[0025] Among them, the BWO algorithm transitions from the exploration stage to the development stage according to the balance coefficient Bf, and its model is:

[0026] B f =B 0 (1-T / 2T max )

[0027] Among them, B 0 is the basic coefficient of random change, T is the current iteration number, T max is the maximum number of iterations, each iteration B 0 Randomly changes between (0,1), when the balance coefficient B f >0.5, it enters the search phase. When the balance coefficient B f ≤0.5, it enters the development stage. As the number of iterations T increases, B f The fluctuation range of decreases from (0,1) to (0,0.5), indicating that the probability changes significantly from the search stage to the development stage, and the probability of the development stage increases with the continuous increase of the number of iterations T.

[0028] The lithium battery health status prediction method combining BWO optimization and GPR of the embodiment of the present invention realizes high-precision prediction of the health status of lithium batteries by combining the Beluga Optimization Algorithm (BWO) and the Gaussian Process Regression model and adopting a stacked integrated secondary learning method. The BWO-optimized GPR model effectively solves the problems of weak noise data processing capabilities and difficulty in hyperparameter selection of traditional models, improves the adaptability and robustness of the model, and the stacked integration strategy further enhances the accuracy and robustness of the prediction results through multi-model fusion. Therefore, the present invention can not only provide reliable SOH estimation in a variety of application scenarios, but also effectively improve the safety and efficiency of lithium battery management systems.

[0029] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0031] Figure 1 A flowchart of a lithium battery health status prediction method combining BWO optimization and GPR according to an embodiment of the present invention;

[0032] Figure 2 This is a comparison diagram of the predicted SOH curve and the actual SOH curve of an embodiment of the present invention. DETAILED DESCRIPTION

[0033] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0034] Existing lithium battery health state estimation methods are mainly divided into two categories: physical model-based and data-driven. However, the physical model-based method relies on complex electrochemical and equivalent circuit models. Although its physical meaning is clear, it has poor adaptability when dealing with the diverse use environments of different types of batteries, and it is difficult to meet the accuracy requirements in practical applications. On the other hand, although the data-driven method performs well in processing complex battery data, most existing methods still have deficiencies in processing noisy data, improving model generalization capabilities and real-time prediction performance, and are difficult to adapt to the complex and changeable use scenarios of lithium batteries. In addition, the hyperparameter selection and optimization of the model still rely on manual parameter adjustment or simple search algorithms, which affects the accuracy and efficiency of the prediction. In response to the above problems, the present invention provides a lithium battery health state prediction method combining BWO optimization and GPR, which further improves the accuracy and robustness of the prediction by combining the White Whale optimization algorithm optimization with the Gaussian process regression model. The method can effectively handle input noise and complex data, and significantly improve the generalization ability and prediction accuracy of the model through multi-model integration and secondary learning. It is suitable for a variety of application scenarios, especially lithium battery management systems in mobile energy storage power supplies. By optimizing the model's hyperparameters and improving its adaptability, an efficient and reliable SOH prediction method is provided, which provides strong support for the safe operation and extended service life of lithium batteries.

[0035] The present invention proposes a lithium battery SOH estimation model based on the integration of BWO-optimized GPR and traditional GPR stacking. This method combines the global search capability of the Beluga optimization algorithm BWO and the small sample data processing capability of the GPR model, significantly improving the accuracy and robustness of lithium battery SOH estimation. The main steps are as follows:

[0036] Step 1: Obtain historical charge and discharge data of the lithium battery and pre-process the historical charge and discharge data.

[0037] First, data preparation: collect historical charging and discharging data of lithium batteries, including operating data of voltage and current. Secondly, perform data preprocessing: clean, normalize and extract features of the data, that is, remove outliers and duplicate data in the data, supplement missing values, convert features of different magnitudes into the same scale, and extract healthy features from them: fixed voltage rise time, constant current charging time and discharge efficiency. Fixed voltage rise time refers to the time required for the battery to rise from one fixed voltage to another during the charging process; constant current charging time refers to the time required for the battery to reach the constant current stage during the charging process; discharge efficiency refers to the ratio of the effective energy that the battery can release during the discharge process to its nominal capacity.

[0038] Step 2: Randomly generate an initial beluga population, where each beluga individual corresponds to a set of hyperparameter combinations of the GPR model.

[0039] In an embodiment of the present invention, step 2 specifically includes:

[0040] The initial beluga population is randomly generated, and each beluga individual X i Represents a group Each beluga individual corresponds to a set of hyperparameter combinations of the GPR model, namely a set of length scale l and signal variance

[0041] In the GPR model, the kernel function is used to define the similarity between data points. One of the common kernel functions is the radial basis kernel function (RBF kernel). The expression of the radial basis kernel function is:

[0042]

[0043] Among them, x i is the i-th data point, x j is the jth data point.

[0044] Step 3: Use the BWO algorithm to optimize the key hyperparameters of the GPR model and maximize the model performance by combining the global search mechanism and the random search mechanism.

[0045] The BWO algorithm is used to optimize the key hyperparameters of the GPR model. By combining the global search mechanism with the random search mechanism, the hyperparameter space is effectively explored to maximize the model performance.

[0046] In an embodiment of the present invention, step 3 specifically includes:

[0047] (1) Fitness evaluation: Calculate the fitness value of each beluga individual and measure the model performance by maximizing the log-marginal likelihood function;

[0048] (2) Position update: The standard white whale predation behavior and Levy flight random search in the BWO algorithm are used to update the position of the white whale, so that the fitness value of the new position gradually increases. The specific position update formula is:

[0049] X i (t+1)=X i (t)+α(X best (t)-X i (t))+β·Levy(λ)

[0050] Among them, X i (t) is the position of the i-th white whale at the t-th iteration, that is, the current hyperparameter combination; X best (t) is the current global optimal hyperparameter combination; α and β are parameters for controlling the step size; Levy (λ) is the Levy flight function used for random search;

[0051] (3) Fitness evaluation and selection: For the newly generated positions, i.e., the new hyperparameter combinations, the fitness values ​​are calculated, and the beluga individuals with higher fitness values ​​are retained to enter the next generation of iterations;

[0052] (4) Iterate until convergence: Continue iterating until the predetermined maximum number of iterations T is reached max Or the change in fitness value is lower than the preset threshold, and finally the optimal hyperparameter combination is obtained;

[0053] Among them, the BWO algorithm transitions from the exploration stage to the development stage according to the balance coefficient Bf, and its model is:

[0054] B f =B 0 (1-T / 2T max )

[0055] Among them, B 0 is the basic coefficient of random change, T is the current iteration number, T max is the maximum number of iterations, each iteration B 0 Randomly changes between (0,1), when the balance coefficient B f >0.5, it enters the search phase. When the balance coefficient B f ≤0.5, it enters the development stage. As the number of iterations T increases, B f The fluctuation range of decreases from (0,1) to (0,0.5), indicating that the probability changes significantly from the search stage to the development stage, and the probability of the development stage increases with the continuous increase of the number of iterations T.

[0056] Step 4: Based on the optimized key hyperparameters, train the GPR model and the BWO-GPR model, use the predicted outputs of the GPR model and the BWO-GPR model as input features, train the second-layer integrated model, and use the GPR regression model in the second-layer model to integrate the predicted results of the BWO-GPR model and the GPR model to form the final stacked integrated model, and use the stacked integrated model to predict the health status of the lithium battery.

[0057] The lithium battery health status prediction method combining BWO optimization and GPR of the present invention is described in detail below through specific embodiments.

[0058] exist Figure 1In the figure, the process starts with data collection and preprocessing, and then the data is divided into training and test sets. Next, two different GPR models are trained, including a BWO-optimized GPR model and a normal GPR model. The outputs of these two models are then used as inputs to the second-layer GPR model for secondary learning. Finally, the model performance is evaluated on the test set, and the health status estimation results of the lithium battery are output. The flowchart clearly shows the complete steps from data processing to model training and result output, reflecting the core idea of ​​using stacked integration and optimization algorithms for SOH prediction.

[0059] The cycle operation data of a 179Ah CBOMHW3NA battery at room temperature of 25°C was used. The first 50% of the collected data was used for model training, and the last 50% was used for model verification.

[0060] The parameters of the studied lithium batteries are shown in Table 1. These parameters include battery capacity, voltage range, cycle life, operating temperature range, etc., which are important bases for experimental design and result analysis.

[0061] Table 1 Lithium battery parameters

[0062]

[0063] from Figure 2 It can be seen that the proposed stacking integrated model performs extremely well in the prediction of the state of health (SOH) of lithium batteries. By comparison, it can be found that the prediction curve of the model of the present invention is highly consistent with the actual SOH curve, especially in the early and middle stages of battery degradation, the model prediction value almost completely fits the actual value, which shows that the model has high accuracy in capturing the battery degradation trend. In addition, the shaded part in the figure represents the confidence interval of the prediction result. It can be observed that the confidence interval is narrow, indicating that the model of the present invention can not only accurately predict the SOH value of the battery, but also show good control ability in terms of prediction uncertainty. This is especially important because in practical applications, the uncertainty of the prediction will directly affect the decision-making process of the battery management system (BMS). The reasonable range of the confidence interval further demonstrates the robustness and reliability of the present model under different working conditions.

[0064] In addition, through the evaluation index calculation, the root mean square error (RMSE) of the model of the present invention is 0.0021809, and the mean absolute error (MAE) is 0.0016394, indicating that the estimation error of the model for the lithium battery SOH is extremely small and has very high accuracy. In addition, the determination coefficient (R 2) reaches 0.99483, which is close to 1, indicating that the model has an excellent fitting effect and can accurately capture the changing trend of SOH. This shows that the model can maintain a high prediction accuracy under different conditions and has strong generalization ability and robustness.

[0065] According to the lithium battery health status prediction method combining BWO optimization and GPR proposed in the embodiment of the present invention, the accuracy and robustness of the prediction are further improved by combining the White Whale optimization algorithm optimization and the Gaussian process regression model. This method can effectively handle input noise and complex data, and significantly improve the generalization ability and prediction accuracy of the model through multi-model integration and secondary learning. It is suitable for a variety of application scenarios, especially lithium battery management systems in mobile energy storage power supplies.

[0066] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or N embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0067] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "N" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0068] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present invention belong.

Claims

1. A lithium battery health status prediction method combining BWO optimization and GPR, characterized in that: The following steps are involved: Step 1, obtaining historical charge and discharge data of the lithium battery, and preprocessing the historical charge and discharge data; Step 2, randomly generate an initial beluga population, each beluga individual corresponds to a set of hyperparameter combinations of the GPR model; Step 3: Use the BWO algorithm to optimize the key hyperparameters of the GPR model and maximize the model performance by combining the global search mechanism and the random search mechanism. Step 4: Based on the optimized key hyperparameters, train the GPR model and the BWO-GPR model, use the predicted outputs of the GPR model and the BWO-GPR model as input features, train the second-layer integrated model, and use the GPR regression model in the second-layer model to integrate the predicted results of the BWO-GPR model and the GPR model to form the final stacked integrated model, and use the stacked integrated model to predict the health status of the lithium battery.

2. The method according to claim 1, characterized in that Step 1 specifically includes: Obtain historical charging and discharging data of lithium batteries, including historical voltage and current operation data; The historical charge and discharge data are cleaned, normalized and feature extracted, outliers and duplicate data in the historical charge and discharge data are removed, missing values ​​are supplemented, features of different magnitudes are converted into the same scale, and healthy features are extracted therefrom: fixed voltage rise time, constant current charging time and discharge efficiency. The fixed voltage rise time refers to the time required for the lithium battery to rise from one fixed voltage to another fixed voltage during the charging process; the constant current charging time refers to the time required for the lithium battery to enter the constant current stage during the charging process; the discharge efficiency refers to the ratio of the effective energy that the lithium battery can release during the discharge process to its nominal capacity.

3. The method according to claim 1, characterized in that Step 2 specifically includes: The initial beluga population is randomly generated, and each beluga individual X i Represents a group Each beluga individual corresponds to a set of hyperparameter combinations of the GPR model, namely a set of length scale l and signal variance In the GPR model, the kernel function is used to define the similarity between data points, where the expression of the radial basis kernel function is: Among them, x i is the i-th data point, x j is the jth data point.

4. The method according to claim 1, characterized in that: Step 3 specifically includes: (1) Fitness evaluation: Calculate the fitness value of each beluga individual and measure the model performance by maximizing the log-marginal likelihood function; (2) Position update: The standard white whale predation behavior and Levy flight random search in the BWO algorithm are used to update the position of the white whale, so that the fitness value of the new position gradually increases. The specific position update formula is: X i (t+1)=X i (t)+α(X best (t)-X i (t))+β·Levy(λ) Among them, X i (t) is the position of the i-th white whale at the t-th iteration, that is, the current hyperparameter combination; X best (t) is the current global optimal hyperparameter combination; α and β are parameters for controlling the step size; Levy (λ) is the Levy flight function used for random search; (3) Fitness evaluation and selection: For the newly generated positions, i.e., the new hyperparameter combinations, the fitness values ​​are calculated, and the beluga individuals with higher fitness values ​​are retained to enter the next generation of iterations; (4) Iterate until convergence: Continue iterating until the predetermined maximum number of iterations T is reached max Or the change in fitness value is lower than the preset threshold, and finally the optimal hyperparameter combination is obtained; Among them, the BWO algorithm transitions from the exploration stage to the development stage according to the balance coefficient Bf, and its model is: B f =B0(1-T / 2T max ) Among them, B0 is the basic coefficient of random change, T is the current number of iterations, and T max is the maximum number of iterations. In each iteration, B0 changes randomly between (0,1). f >0.5, it enters the search phase. When the balance coefficient B f ≤0.5, it enters the development stage. As the number of iterations T increases, B f The fluctuation range of decreases from (0,1) to (0,0.5), indicating that the probability changes significantly from the search stage to the development stage, and the probability of the development stage increases with the continuous increase of the number of iterations T.

Images