Model training method, battery SOC estimation method, electronic equipment and storage medium

Abstract

The invention provides a model training method, a battery SOC estimation method, electronic equipment and a storage medium, and can be applied to the technical field of batteries. The model training method comprises the following steps: constructing a deep learning model; the hyper-parameters of the deep learning model are optimized based on an enhanced litsea rotundifolia optimization algorithm, and the optimal hyper-parameters of the deep learning model are determined; training and testing the deep learning model based on a pre-constructed training set and test to obtain a battery SOC estimation model; wherein the battery SOC estimation model is used for obtaining a battery SOC estimation result based on the battery data of the target battery. According to the method, dynamic parameter adjustment is carried out on the deep learning model through the enhanced golden litunt optimization algorithm, the model adaptability can be improved, the battery SOC estimation precision of the battery SOC estimation model is remarkably improved, good generalization ability is shown, and higher estimation precision and higher robustness are achieved under the complex working condition.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of batteries, in particular to a model training method, a battery SOC estimation method, an electronic device and a storage medium. BACKGROUND

[0002] The battery SOC (State of Charge) is the state of charge of the battery, and the available energy of the battery can be directly reflected by the battery SOC. As one of the important carriers of electrochemical energy storage, the battery capacity will decrease in the long-term use process, and at the same time, the internal resistance will increase, which will significantly increase the heat generation of the battery in the use process and reduce the safety and stability of the battery. In order to ensure the safe and stable operation of the battery, the estimation of the battery SOC is crucial.

[0003] At present, the main methods to realize the estimation of the battery SOC can be summarized into two categories, namely the model method and the data-driven method. Among them, the model method is designed based on simplified assumptions, and the quality of the model directly affects the estimation accuracy of the battery SOC. The data-driven method does not need to consider the complex electrochemical characteristics of the battery, but constructs the relationship between the input parameters and the output parameters of the battery, so as to accurately estimate the SOC of the battery.

[0004] However, the traditional data-driven method (such as SVM and RF) does not need an accurate physical model, but relies on a large amount of historical data for training, has weak generalization ability for working conditions not included in the training set, and is difficult to capture the dynamic evolution characteristics of the battery, and is easy to accumulate errors in long-term prediction. SUMMARY

[0005] The present application provides a model training method, a battery SOC estimation method, an electronic device and a storage medium to solve the technical problem that the existing battery SOC estimation technology has poor estimation effect.

[0006] According to a first aspect of the present application, the present application provides a model training method, comprising:

[0007] constructing a deep learning model;

[0008] optimizing the hyperparameters of the deep learning model based on the enhanced golden jackal optimization algorithm, and determining the optimal hyperparameters of the deep learning model;

[0009] training and testing the deep learning model based on the pre-constructed training set and test set respectively, and obtaining a battery SOC estimation model; wherein the battery SOC estimation model is used to obtain a battery SOC estimation result based on the battery data of a target battery.

[0010] In an implementable embodiment, the hyperparameters of the deep learning model are optimized based on an enhanced golden jackal optimization algorithm, and the optimal hyperparameters of the deep learning model are determined, comprising:

[0011] initializing a population; wherein the population comprises a plurality of individuals, and the positions of the individuals are used to represent combinations of the hyperparameters, and the initial individuals are uniformly distributed in a search space;

[0012] When the iteration stop condition is not met, the positions of the individuals in the population are updated based on the fitness of the individuals in the population; wherein the iteration stop condition comprises at least one of the current iteration number being equal to the maximum iteration number or the positions of the individuals converging;

[0013] When the iteration stop condition is met, the position of the individual with the highest fitness in the population obtained at present is output as the optimal hyperparameters of the deep learning model.

[0014] In an implementable embodiment, the positions of the individuals in the population are updated based on the fitness of the individuals in the population, comprising:

[0015] Based on the fitness of the individuals in the population, a male golden jackal and a female golden jackal are determined; wherein the male golden jackal is the individual with the highest fitness, and the female golden jackal is the individual with the second highest fitness;

[0016] The escape energy of the prey in the population is updated; wherein the prey is an individual other than the male golden jackal and the female golden jackal in the population;

[0017] If the escape energy is greater than a preset threshold, the golden jackal is in an exploration stage, and the positions of the individuals in the population are updated based on an exploration strategy;

[0018] If the escape energy is less than a preset threshold, the golden jackal is in a hunting stage, and the positions of the individuals in the population are updated based on a hunting strategy.

[0019] In an implementable embodiment, the escape energy satisfies the following formula:

[0020]

[0021] wherein E represents the escape energy, represents the initial value of the escape energy of an individual, represents an energy decreasing function, r represents a random coefficient in the range of [0, 1], C1 represents a constant coefficient with a value of 1.5, t represents the current iteration number, and T represents the maximum iteration number.

[0022] In an implementable embodiment, the position of the individual in the population is updated based on an exploration strategy, including:

[0023] updating a first relative position and a second relative position of the individual with respect to the male and the female cheetahs based on a prey exploration model in which the male cheetah leads and the female cheetah follows;

[0024] obtaining an updated position of the individual based on an average position of the first relative position and the second relative position.

[0025] In an implementable embodiment, the position of the individual in the population is updated based on a hunting strategy, including:

[0026] updating a third relative position and a fourth relative position of the individual with respect to the male and the female cheetahs based on a prey hunting model in which the male and the female cheetahs hunt together;

[0027] obtaining an updated position of the individual based on an adaptive weight position of the third relative position and the fourth relative position.

[0028] In an implementable embodiment, the method further includes:

[0029] training the deep learning model based on the position of the individual to determine a fitness of the individual; wherein the fitness is a prediction accuracy, an F1 score, a loss value, or a root mean square error of the deep learning model.

[0030] According to a second aspect of the present disclosure, the present disclosure provides a battery SOC estimation method, including:

[0031] obtaining battery data of a target battery;

[0032] inputting the battery data into a battery SOC estimation model to obtain a battery SOC estimation result output by the battery SOC estimation model; wherein the battery SOC estimation model is trained based on the model training method of any one of the first aspect.

[0033] According to a third aspect of the present disclosure, the present disclosure provides a model training device, including:

[0034] a model construction module for constructing a deep learning model;

[0035] a parameter optimization module for optimizing hyperparameters of the deep learning model based on an enhanced cheetah optimization algorithm to determine optimal hyperparameters of the deep learning model;

[0036] The model training module trains and tests the deep learning model based on a pre-constructed training set and test set respectively, to obtain a battery SOC estimation model; wherein the battery SOC estimation model is used to obtain a battery SOC estimation result based on battery data of a target battery.

[0037] According to a fourth aspect of the present application, the present application provides a battery SOC estimation device, comprising:

[0038] The data acquisition module is configured to acquire battery data of a target battery.

[0039] The SOC estimation module is configured to input the battery data into a battery SOC estimation model, and obtain a battery SOC estimation result output by the battery SOC estimation model; wherein the battery SOC estimation model is trained based on the model training method of any one of the first aspect.

[0040] According to a fifth aspect of the present application, the present application provides an electronic device, comprising a processor, and a memory connected with the processor in communication;

[0041] The memory stores computer execution instructions.

[0042] The processor executes the computer execution instructions stored in the memory, to implement the method of any one of the first aspect or the second aspect.

[0043] According to a sixth aspect of the present application, the present application provides a computer readable storage medium, wherein the computer readable storage medium stores computer execution instructions, and the computer execution instructions are executed to implement the method of any one of the first aspect or the second aspect.

[0044] According to a seventh aspect of the present application, the present application provides a computer program product, comprising a computer program, wherein the computer program is executed to implement the method of any one of the first aspect or the second aspect.

[0045] Compared with the prior art, the present application has the following beneficial effects:

[0046] This application provides a model training method, a battery SOC estimation method, an electronic device, and a storage medium. The Enhanced Golden Jackal Optimization Algorithm (IGJO) simulates the cooperative hunting behavior of a golden jackal group, enabling the algorithm to achieve fast convergence and strong global search capabilities. It can efficiently solve complex nonlinear problems in hyperparameter optimization of deep learning models, automatically adjusting and optimizing the hyperparameters of the deep learning model to avoid getting trapped in local optima, thus reducing the time and cost of manual parameter tuning. Simultaneously, using the IGJO algorithm for dynamic parameter tuning of the deep learning model can improve model adaptability, reduce the model's dependence on large amounts of labeled data, significantly improve the battery SOC estimation accuracy of the battery SOC estimation model, and exhibit good generalization ability. It demonstrates higher estimation accuracy and stronger robustness under complex operating conditions, contributing to improved battery management capabilities and further enhancing battery safety and stability. Attached Figure Description

[0047] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0048] Figure 1 A schematic flowchart illustrating a model training method provided in an embodiment of this application;

[0049] Figure 2 A flowchart illustrating a method for optimizing hyperparameters of a deep learning model based on an enhanced golden jackal optimization algorithm, provided in this application embodiment;

[0050] Figure 3 A schematic flowchart illustrating a battery SOC estimation method provided in an embodiment of this application;

[0051] Figure 4 This is a schematic diagram of the structure of a model training device provided in an embodiment of this application;

[0052] Figure 5 This is a schematic diagram of the structure of a battery SOC estimation device provided in an embodiment of this application;

[0053] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0054] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0055] The exemplary embodiments will be described in detail herein with reference to the attached drawings. The description of the exemplary embodiments is intended to apply to various alternative embodiments as well. The following description is not limited to the exemplary embodiments, but rather, is applicable to any apparatus and / or method which are adapted to carry out the aspects of the present application as described in the claims.

[0056] Battery SOC (State of Charge) is a key indicator of the current remaining capacity of the battery relative to its full charge state, usually presented in percentage form (such as 0% for completely empty, 100% for completely full), and can directly reflect the available energy of the battery. As one of the important carriers of electrochemical energy storage, the capacity of the battery will decline over time, and the internal resistance of the battery will also gradually increase. Since the battery internal resistance is a physical quantity that measures the degree of hindrance to the current inside the battery, an increase in internal resistance means that more energy will be lost in the form of heat during the charging and discharging process of the battery, resulting in a significant increase in the heat generated inside the battery, reducing its safety and stability.

[0057] Therefore, in order to ensure the safe and stable operation of the battery, the estimation of the battery SOC is crucial. By accurately estimating the SOC of the battery, the remaining capacity of the battery can be understood in real time, so as to reasonably plan the use of the battery and avoid further damage to the battery due to overcharging or overdischarging, thereby prolonging the service life of the battery. At the same time, accurate battery SOC estimation can also provide key data support for the battery management system, so that the system can adjust the charging and discharging strategy according to the real-time state of the battery, effectively control the heat generation of the battery, and ensure that the battery always operates within a safe and stable temperature range, greatly improving the reliability and safety of the entire energy storage system.

[0058] The main methods for realizing battery SOC estimation can be summarized into two categories: model method and data-driven method. The common models in the model method are further divided into electrochemical model and equivalent circuit model. The electrochemical model describes the physical and chemical change mechanism inside the battery through partial differential equations, and the equivalent circuit model describes the dynamic response process of the battery through resistance, capacitance and inductance. The model method is designed based on simplified assumptions, and the quality of the model directly affects the estimation accuracy of the battery SOC. Although the electrochemical model method can accurately describe the internal reaction mechanism of the battery, it needs to establish a complex partial differential equation set, which involves a large number of parameters (such as ion concentration distribution, diffusion coefficient) that are difficult to measure in real time, resulting in heavy calculation burden, poor real-time performance, and sensitivity to parameter changes, making it difficult to adapt to battery aging and working condition fluctuations. The equivalent circuit model method simplifies the dynamic characteristics of the battery through an RC network, and has high calculation efficiency, but the model accuracy depends on the accuracy of parameter identification, and the parameters will change with temperature, SOC and cycle number, resulting in increased estimation error under complex working conditions, especially for large current charging and discharging or temperature change scenarios.

[0059] The data-driven method does not need to consider the complex electrochemical characteristics inside the battery, but establishes a SOC estimation model through data learning method, sets the battery as a "black box", and constructs the relationship between the input parameters and the output parameters of the battery, so as to accurately estimate the SOC of the battery. The data-driven technology for battery SOC estimation mainly relies on historical data to construct the mapping relationship between the input and the SOC, avoiding complex electrochemical modeling. In traditional machine learning methods, support vector machine (SVM) handles nonlinear problems through kernel function, and performs well in small sample scenarios; random forest (RF) integrates multiple decision trees, has strong noise resistance and feature importance evaluation; Gaussian process regression (GPR) provides probabilistic estimation based on Bayesian theory, and is suitable for quantifying uncertainty. However, traditional data-driven methods (such as SVM and RF) do not require accurate physical models, but rely on a large amount of historical data for training, have weak generalization ability for working conditions not included in the training set, and are difficult to capture the dynamic evolution characteristics of the battery, and are prone to accumulate errors in long-term prediction.

[0060] In view of the above technical problems, the present application provides a model training method, a battery SOC estimation method, an electronic device and a storage medium. The deep learning model is dynamically parameterized by an enhanced golden jackal optimization algorithm, which can improve the adaptability of the model, reduce the dependence of the model on a large amount of labeled data, significantly improve the battery SOC estimation accuracy of the battery SOC estimation model, and exhibit good generalization ability, higher estimation accuracy and stronger robustness under complex working conditions.

[0061] The technical solutions of the battery SOC estimation method provided in the application will be described in detail below through specific embodiments. It should be noted that the following embodiments can exist independently or in combination. For the same or similar content, it may not be repeated in different embodiments.

[0062] Figure 1 A flowchart of a model training method provided in an embodiment of the application is shown in Figure 1 In some embodiments, the flow of the model training method includes the following steps:

[0063] S101, constructing a deep learning model.

[0064] Specifically, the deep learning model includes a CNN model and a BiLSTM model. The CNN model is used to extract data space features of battery data such as voltage and current, and the BiLSTM model is used to capture time sequence dependence in both directions based on the data space features to obtain battery SOC estimation results.

[0065] The combination of the CNN model and the BiLSTM model can combine the efficiency of CNN local feature extraction and the advantage of BiLSTM comprehensive capture of forward and backward information, and can fuse the advantages of both to model the battery non-linear degradation precisely and improve the performance of the model in processing complex sequence data. From the estimation results, compared with the traditional single deep learning network model, the CNN-BiLSTM model can well adapt to the fusion of battery multi-dimensional information parameters as feature input

[0066] Optionally, the CNN model layer adopts a parallel convolution architecture to capture local detail features and global correlation features respectively.

[0067] S102, optimizing the hyperparameters of the deep learning model based on the enhanced golden jackal optimization algorithm to determine the optimal hyperparameters of the deep learning model.

[0068] The enhanced golden jackal optimization algorithm (Improved Golden Jackal Optimization, IGJO) can effectively avoid the local optimal trap in the optimization of the hyperparameters of the deep learning model by virtue of its group intelligence search mechanism, and quickly and accurately locate the optimal combination of the hyperparameters of the deep learning model.

[0069] Specifically, the hyperparameters of the model can be selected as optimization targets, such as learning rate, learning rate reduction factor, and iteration number. A suitable learning rate has a direct impact on the performance of the model. A high learning rate may cause oscillation of the model, and a low learning rate affects the convergence speed of the model. A suitable learning rate reduction factor can ensure that the model has a faster convergence speed in the early stage of training and the ability to fine-tune in the later stage.

[0070] S103, training and testing the deep learning model based on the pre-constructed training set and test set respectively, to obtain a battery SOC estimation model; wherein the battery SOC estimation model is used to obtain a battery SOC estimation result based on battery data of a target battery.

[0071] Wherein, after obtaining the optimal combination of hyperparameters of the deep learning model through the enhanced golden jackal optimization algorithm, the deep learning model is trained to be a battery SOC estimation model for obtaining a battery SOC estimation result based on battery data of a target battery, which significantly improves the prediction accuracy and convergence speed of the model.

[0072] Specifically, after training, a battery SOC estimation model based on the IGJO-CNN-BiLSTM model is obtained

[0073] In this embodiment, the enhanced golden jackal optimization algorithm (IGJO) simulates the cooperative hunting behavior of a golden jackal group, so that the algorithm has fast convergence and strong global search capability, can efficiently solve complex nonlinear problems in hyperparameter optimization of deep learning models, and can automatically adjust and optimize the hyperparameters of the deep learning model to avoid being trapped in local optimum, thereby reducing the time and cost of manual parameter adjustment. At the same time, dynamic parameter adjustment of the deep learning model with the enhanced golden jackal optimization algorithm can also improve the adaptability of the model, reduce the dependence of the model on a large amount of labeled data, significantly improve the battery SOC estimation accuracy of the battery SOC estimation model, and exhibit good generalization ability. In complex working conditions, it has higher estimation accuracy and stronger robustness, which helps to improve the battery management capability, thereby further improving the safety and stability of the battery.

[0074] In Figure 1 The model training method shown in the figure needs to optimize the hyperparameters of the deep learning model based on the enhanced golden jackal optimization algorithm. The following describes the model training method in combination with Figure 2 The technical scheme of the above model training method further introduces the content of optimizing the hyperparameters of the deep learning model based on the enhanced golden jackal optimization algorithm.

[0075] Figure 2 A flowchart of a method for optimizing the hyperparameters of a deep learning model based on an enhanced golden jackal optimization algorithm according to an embodiment of the present application is shown in Figure 2 In some embodiments, the method for optimizing the hyperparameters of the deep learning model based on the enhanced golden jackal optimization algorithm includes the following steps:

[0076] S201, initialize the population; wherein the population includes multiple individuals, and the position of an individual is used to represent a combination of hyperparameters. The initial individuals are uniformly distributed in the search space.

[0077] Specifically, in the initialization stage, the individuals of the initial population are uniformly distributed in the search space. Wherein, the position of the initial individual satisfies the following formula:

[0078]

[0079] Wherein, represents the initial position of the individual, and respectively represent the upper and lower boundaries of the search space, and rand represents a uniformly distributed random value in [0, 1].

[0080] Specifically, the position matrix of the population satisfies the following formula:

[0081]

[0082] Wherein, Prey represents the position matrix of the population, n and d respectively represent the number of individuals in the population and the dimension of the hyperparameter variable to be optimized.

[0083] Specifically, the fitness matrix of the population satisfies the following formula:

[0084]

[0085] Wherein, FOA represents the fitness matrix of the population, and f represents the fitness function.

[0086] Optionally, for obtaining the fitness of the individual in the population, specifically includes: training the deep learning model based on the position of the individual to determine the fitness of the individual; wherein, the fitness is the prediction accuracy, F1 score, loss value or root mean square error of the deep learning model.

[0087] Wherein, since the position of the individual is used to represent the combination of hyperparameters, the fitness of the individual is obtained by substituting the position of the individual into the deep learning model for training.

[0088] Specifically, the prediction accuracy refers to the proportion of the number of samples correctly predicted by the model in the test data to the total number of samples, which directly reflects the overall classification ability of the model. As the fitness, it can directly measure the correctness of the model prediction result, and quickly screen out hyperparameter combinations with high classification accuracy.

[0089] Specifically, the F1 score is the harmonic mean of precision (the proportion of samples actually positive among samples predicted positive) and recall (the proportion of samples correctly predicted among samples actually positive), which is used to balance the performance of the model in accuracy and coverage. As the fitness, it can consider the recognition ability and coverage ability of the model to the minority class at the same time, avoiding ignoring the minority class due to the pursuit of high accuracy.

[0090] Specifically, the loss value refers to a difference value between a model prediction value and a true value, and is used to guide the adjustment of model parameters by optimizing a loss function (such as cross-entropy loss), and can be highly consistent with the model training target as fitness, and can dynamically reflect the influence of hyperparameters on the convergence speed and final performance of the model.

[0091] Specifically, the root mean square error (RMSE) refers to the square root of the average value of the square of the difference between the model prediction value and the true value, and is used to quantify the discrete degree of the prediction result of the regression model, and the smaller the value, the closer the model prediction is to the true value. As fitness, it can be sensitive to outliers and retains the square error characteristic, and is suitable for regression tasks that require strict control of prediction error.

[0092] S202, when the iteration stop condition is not met, the position of the individual in the population is updated based on the fitness of the individual in the population; wherein the iteration stop condition includes at least one of the current iteration number equaling the maximum iteration number or the position of the individual converging.

[0093] Wherein, when the iteration stop condition is not met, the position of the population is iteratively updated to continuously optimize.

[0094] Optionally, the position of the individual in the population is updated based on the fitness of the individual in the population, specifically including:

[0095] Step 1, based on the fitness of the individual in the population, determine the male and female gold cats; wherein the male gold cat is the individual with the highest fitness, and the female gold cat is the individual with the second highest fitness.

[0096] Wherein, the fitness of the individual in the population is sorted from high to low, then the individual with the highest fitness, i.e. the optimal individual in the population, is the male gold cat; the individual with the second highest fitness, i.e. the suboptimal individual in the population, is the female gold cat.

[0097] Step 2, update the escape energy of the prey in the population; wherein the prey is other individuals in the population except the male and female gold cats.

[0098] Wherein, in the enhanced gold cat optimization algorithm, the escape energy of the prey is a core parameter for controlling the algorithm to switch from global exploration (exploration phase) to local development (pursuit phase).

[0099] Specifically, the escape energy satisfies the following formula:

[0100]

[0101] Wherein, E represents the escape energy, represents the initial value of the individual escape energy, represents the energy decreasing function, r represents a random coefficient in the range of [0, 1], C1 represents a constant coefficient with a value of 1.5, t represents the current iteration number, and T represents the maximum iteration number.

[0102] wherein, for making the escape energy non-linearly decrease from C1 to 0.

[0103] wherein, throughout the iteration process, by constructing the energy decreasing function, so that non-linearly decreases from 1.5 to 0, which enhances the convergence ability of the optimization algorithm.

[0104] Step 3, if the escape energy is greater than the preset threshold, the golden cat is in the exploration stage, and the positions of individuals in the population are updated based on the exploration strategy.

[0105] wherein, when the prey escape energy is sufficient, the golden cat turns to global search to find new prey and avoids falling into local optimum.

[0106] Specifically, the preset threshold is set to 1.

[0107] Optionally, the positions of individuals in the population are updated based on the exploration strategy, including:

[0108] Step 31, based on the prey exploration model of male golden cat leading and female golden cat following, the first relative position and the second relative position of the individual with the male golden cat and the female golden cat are updated.

[0109] wherein, in the stage of exploring prey, the prey exploration model simulates the ability of golden cat hunting and tracking prey. When the golden cat hunts, generally the male golden cat leads the hunting, while the female golden cat follows closely behind. Therefore, according to the hunting characteristics of the golden cat, it is concluded that the update of the individual position simulated in the prey exploration model satisfies the following formula:

[0110]

[0111] wherein, t represents the current iteration number, Prey(t) is the position vector of the prey in the tth iteration, represents the position of the male golden cat, represents the position of the female golden cat, represents the first relative position of the prey and the male golden cat in the tth iteration, represents the second relative position of the prey and the female golden cat; rl represents a random number based on levy distribution, which simulates the randomness of the prey escaping in the counter-hunting when representing the distance between the golden cat and the prey.

[0112] Specifically, rl satisfies the following formula:

[0113]

[0114] wherein, LF(y) represents a Levy flight function, which satisfies the following formula:

[0115]

[0116] wherein, μ and v respectively represent random numbers in the range of (0, 1), and β represents a default constant, which is set to 1.5.

[0117] Optionally, in the optimization process, the ordinary GJO causes the golden jackal population to approach the current optimal fitness value position led by the male golden jackal, which is easy to cause the optimization algorithm to fall into local optimum. Therefore, the Gaussian mutation strategy is introduced for the calculation of the male golden jackal position in the exploration stage. The Gaussian mutation increases the disturbance and expands the range of local search of the golden jackal, and enhances the ability of the algorithm to search for the potential best fitness.

[0118] Specifically, the Gaussian mutation satisfies the following formula:

[0119]

[0120] wherein, α represents a random number between 0 and 1.

[0121] Correspondingly, the first relative position obtained based on the Gaussian mutation satisfies the following formula:

[0122]

[0123] Step 32, obtaining the updated position of the individual based on the average position of the first relative position and the second relative position.

[0124] Specifically, the updated position of the individual satisfies the following formula:

[0125]

[0126] wherein, Y(t+1) represents the position of the individual after t+1 iterations.

[0127] Step 4, if the escape energy is less than the preset threshold, the golden jackal is in the hunting stage, and the position of the individual in the population is updated based on the hunting strategy.

[0128] wherein, when the prey energy is exhausted, the golden jackal initiates local attack to approach the optimal solution.

[0129] Optionally, updating the position of the individual in the population based on the hunting strategy includes:

[0130] Step 41, updating the third relative position and the fourth relative position of the individual with the male golden jackal and the female golden jackal respectively based on the prey hunting model of the prey hunted by the male golden jackal and the female golden jackal.

[0131] wherein the escape energy of the prey decreases with the increase of the iteration number under the continuous pursuit of the dhole. Meanwhile, in order to simulate the situation that the dhole encounters obstacles on the way to approach the prey, and to prevent the ideal situation that the dhole quickly approaches the prey, the rl variable is added to the mathematical model of the male and female dholes in the attack stage, so that the update of the individual position simulated in the prey encirclement model satisfies the following formula according to the hunting characteristics of the dhole:

[0132]

[0133] wherein t represents the current iteration number, Prey(t) is the position vector of the prey in the tth iteration, represents the position of the male dhole, represents the position of the female dhole, represents the third relative position between the prey and the male dhole in the tth iteration, represents the fourth relative position between the prey and the female dhole; rl represents a random number based on the levy distribution, which simulates the randomness of the prey escaping in the counter-encirclement when representing the distance between the dhole and the prey.

[0134] Step 42, obtaining the updated position of the individual based on the adaptive weight position of the third relative position and the fourth relative position.

[0135] Specifically, the updated position of the individual satisfies the following formula:

[0136]

[0137] wherein Y(t+1) represents the position of the individual after the (t+1)th iteration, β represents the update coefficient, t is the current iteration number, and T represents the maximum iteration number, and represent the coefficients.

[0138] Specifically, it is found through testing that when and the coefficients are 0.45 and 0.35 respectively, the convergence speed and accuracy of the model are improved.

[0139] wherein since the direct average method of male and female dholes is adopted, the characteristics of individuals in the population are ignored, and the importance of individuals with high fitness in decision-making cannot be reflected, therefore, in the encirclement stage, the influence weight of the male dhole on the position is improved through the adaptive weight position, so as to further improve the accuracy of the hyperparameter optimization.

[0140] S203, when the iteration stop condition is met, the position of the individual with the highest fitness in the current obtained population is output as the optimal hyperparameter of the deep learning model.

[0141] Wherein, after the enhanced golden jackal optimization algorithm optimization is ended, the individual with the highest fitness in the population, that is, the position of the finally determined male golden jackal, is taken as the output to obtain the optimal hyperparameters of the deep learning model.

[0142] In the embodiment, the golden jackal optimization (GJO) algorithm is a meta-heuristic optimization algorithm developed by simulating the hunting and social behavior of golden jackals. The algorithm has the advantages of simple principle, few parameters, excellent performance in unknown search space, etc. In view of the problem that traditional GJO is prone to falling into local optimal solution as a meta-heuristic algorithm, the enhanced golden jackal optimization algorithm IGJO in the embodiment improves the GJO algorithm by constructing an energy decreasing function and a self-adaptive weight fusion method to further improve the optimization performance of the optimization algorithm.

[0143] Figure 3 A flowchart of a battery SOC estimation method provided in the embodiment is shown in Figure 3 In some embodiments, the flow of the battery SOC estimation method includes the following steps:

[0144] S301, obtaining battery data of a target battery.

[0145] S302, inputting the battery data into a battery SOC estimation model to obtain a battery SOC estimation result output by the battery SOC estimation model; wherein the battery SOC estimation model is trained by the model training method described above.

[0146] In the embodiment, the battery SOC estimation model obtained by parameter tuning based on the enhanced golden jackal optimization algorithm can significantly improve the estimation accuracy and efficiency of the battery SOC, thereby helping to improve the battery management capability and further improve the safety and stability of the battery.

[0147] The battery SOC estimation method of the embodiment will be further described below in combination with specific tests:

[0148] Specifically, three battery operating conditions are set for testing:

[0149] Condition 1: The battery is placed in a low-temperature environment of 0℃ constant temperature, and periodic charge and discharge tests are carried out at a constant low rate;

[0150] Condition 2: The battery is placed in a high-temperature environment of 50℃ constant temperature, and periodic charge and discharge tests are carried out at a constant high rate;

[0151] Condition 3: The battery is placed in a constant temperature oven, the temperature of the constant temperature oven is randomly changed between 0~50℃, and the battery charge and discharge and the charge and discharge rate are randomly switched for testing.

[0152] Specifically, the model 1 and the model 2 applied in the test are two common conventional deep learning models.

[0153] Specifically, for the test scheme one, specifically includes:

[0154] Firstly, the voltage, current and temperature data of the single battery under different working conditions are collected by using high-precision data acquisition equipment, and the battery charging and discharging state is recorded synchronously. In view of the high-frequency noise and peak interference existing in the original data, the MATLAB moving average filtering method is used for denoising processing, which effectively retains more than 95% of the useful signal energy and reduces the interference of "noise" on the model.

[0155] Secondly, the preprocessed data is normalized to the interval [0, 1], and the parallel convolution architecture is adopted in the CNN layer to capture local detail features and global correlation features respectively. The learning rate, learning rate reduction factor and iteration number are selected as the optimization target in the BiLSTM model, and finally the enhanced golden jackal optimization algorithm (IGJO) is introduced for hyperparameter optimization processing.

[0156] As shown in Table 1, the IGJO-CNN-BiLSTM model determines the hyperparameters through the adaptive optimization of IGJO and the serial combination of CNN and BiLSTM, which improves the ability of the model to mine and fuse multi-dimensional parameters as input features. In the test of various working conditions, the IGJO-CNN-BiLSTM model shows good SOC estimation effect.

[0157] Table 1 IGJO-CNN-BiLSTM model estimation battery SOC precision evaluation index table based on different working conditions

[0158]

[0159] Table 2 Battery SOC estimation precision evaluation index table based on different models

[0160]

[0161] As shown in Table 2, compared with the traditional deep learning model, the IGJO-CNN-BiLSTM model has different degrees of improvement in the three evaluation parameters. The RMSE decreases to 1.55%, the MAE decreases to 1.28%, and the correlation coefficient R2 increases to 0.9975.

[0162] Specifically, for the test scheme two, specifically includes:

[0163] Firstly, the voltage, current and temperature data of each single battery in the battery pack under complex working conditions are synchronously collected by using a distributed multi-channel data acquisition system, and the charging and discharging state, charging and discharging cycle number and connection state between each single battery of the battery pack are recorded.

[0164] Due to the mutual influence between each single battery in the battery pack during operation and the data being easily disturbed by environmental factors, leading to large data fluctuations, a hybrid denoising method based on wavelet transform and Kalman filtering is adopted for the periodic interference, channel crosstalk and random noise in the original data. The data is decomposed into different frequency domains by wavelet transform to remove high-frequency noise, and then Kalman filtering is used for dynamic estimation of each single battery data to effectively suppress channel crosstalk, retain more than 98% of useful signal energy, improve data accuracy and stability, and reduce the influence of interference on the model.

[0165] Secondly, the preprocessed data is normalized to the [0, 1] interval, and a hierarchical parallel convolution architecture is adopted in the CNN layer according to the characteristics of the battery pack data. The first layer of parallel convolution extracts the local detail features of each single battery data, and the second layer of convolution fuses the local features of each single battery to capture the global correlation features of the battery pack and the mutual influence relationship between the single batteries. In the BiLSTM model, the learning rate, learning rate drop factor and number of hidden layer neurons are selected as the optimization target. Combined with the enhanced golden jackal optimization algorithm (IGJO), the efficiency and accuracy of hyperparameter optimization are improved, so as to improve the accuracy of the model in estimating the SOC of the battery pack.

[0166] Table 3 Battery SOC estimation accuracy evaluation index table based on different models

[0167]

[0168] As shown in Table 3, under the battery working condition 3, when the current (I), voltage (V) and temperature (T) are used as feature data, the model of the application has a significant advantage in battery SOC estimation. The root mean square error (RMSE) is only 0.0231, which is significantly lower than 0.0432 of model 1 and 0.0484 of model 2. The mean absolute error (MAE) is 0.0208, which is much smaller than 0.0317 of model 1 and 0.0377 of model 2, indicating that the deviation between the predicted value and the true value is extremely small. At the same time, the determination coefficient (R²) of the model of the application reaches 0.9891, which is higher than 0.9751 of model 1 and 0.9725 of model 2, and the fitting degree of data is better, which can more accurately reflect the relationship between the battery state and the feature data, and the comprehensive performance is better than that of the comparative models.

[0169] Specifically, for test scheme three, it specifically includes:

[0170] First, the charging and discharging state of the battery pack as a whole is recorded in view of the characteristics of the lead-acid battery pack. Considering that the lead-acid battery has a significant voltage hysteresis effect during charging and discharging, the data presents nonlinear fluctuation characteristics. Therefore, in view of the voltage hysteresis noise existing in the original data, a suitable denoising method is selected for noise removal. For example, EMD is used to decompose the data into multiple intrinsic mode functions (IMF), and remove the IMF components related to noise; CEEMDAN is used to further decompose and reconstruct the remaining IMF components, effectively suppress the voltage hysteresis noise and concentration fluctuation interference, retain more than 97% of the useful signal energy, enhance the smoothness and reliability of the data, and reduce the influence of interference on the model.

[0171] Second, the preprocessed data is normalized, and different scale convolution kernels are used in the first layer of CNN to extract multi-scale features of the data in parallel; the second layer of residual connection structure fuses features of different scales to capture the nonlinear change characteristics of the lead-acid battery pack data and the dynamic characteristics in the charging and discharging process. In the BiLSTM model, the learning rate, the learning rate drop factor and the forgetting gate threshold are selected as the optimization target. Considering the memory effect and self-discharge characteristics of the lead-acid battery, a suitable forgetting gate threshold can better process the historical information in the long sequence data. Finally, the enhanced golden jackal optimization algorithm (IGJO) is combined to perform global search to find the optimal region of the hyperparameters, improve the efficiency and accuracy of hyperparameter optimization, and thus improve the accuracy of the model in estimating the SOC of the lead-acid battery pack.

[0172] Figure 4 is a structural schematic diagram of a model training device provided by an embodiment of the present application, referring to Figure 4 The model training device includes various functional modules for implementing the foregoing model training method, and any functional module can be realized by software and / or hardware.

[0173] In some embodiments, the model training device 400 includes a model construction module 401, a parameter optimization module 402, and a model training module 403. Among them:

[0174] The model construction module 401 is configured to construct a deep learning model.

[0175] The parameter optimization module 402 is configured to optimize the hyperparameters of the deep learning model based on the enhanced golden jackal optimization algorithm, and determine the optimal hyperparameters of the deep learning model.

[0176] The model training module 403 trains and tests the deep learning model based on the pre-constructed training set and test set respectively, and obtains a battery SOC estimation model; wherein the battery SOC estimation model is configured to obtain a battery SOC estimation result based on battery data of a target battery.

[0177] In some embodiments, the parameter optimization module 402 is specifically configured to:

[0178] initialize a population; wherein the population comprises a plurality of individuals, and a position of each individual is used to represent a combination of hyperparameters, and the initial individuals are uniformly distributed in a search space;

[0179] when the iteration stop condition is not met, perform an operation of updating the position of each individual in the population based on the fitness of the individual in the population; wherein the iteration stop condition comprises at least one of the current iteration number being equal to the maximum iteration number or the position of the individual converging;

[0180] when the iteration stop condition is met, output the position of the individual with the highest fitness in the current obtained population as the optimal hyperparameters of the deep learning model.

[0181] In some embodiments, the parameter optimization module 402 is specifically further configured to:

[0182] determine a male and a female based on the fitness of the individual in the population; wherein the male is the individual with the highest fitness, and the female is the individual with the second highest fitness;

[0183] update the escape energy of the prey in the population; wherein the prey is other individual in the population except the male and the female;

[0184] if the escape energy is greater than a preset threshold, the jackal is in an exploration stage, and the position of each individual in the population is updated based on an exploration strategy;

[0185] if the escape energy is less than the preset threshold, the jackal is in a hunting stage, and the position of each individual in the population is updated based on a hunting strategy.

[0186] In some embodiments, the escape energy satisfies the following formula:

[0187]

[0188] wherein E represents the escape energy, represents an initial value of the escape energy of the individual, represents an energy decreasing function, r represents a random coefficient in the range of [0, 1], C1 represents a constant coefficient with a value of 1.5, t represents the current iteration number, and T represents the maximum iteration number.

[0189] In some embodiments, the parameter optimization module 402 is specifically further configured to:

[0190] update the first relative position and the second relative position of each individual with the male and the female based on a prey exploration model of the male leading and the female following;

[0191] obtain the updated position of the individual based on the average position of the first relative position and the second relative position.

[0192] In some embodiments, the parameter optimization module 402 is specifically further configured to:

[0193] update the third relative position and the fourth relative position of the individual with the male dhole and the female dhole respectively based on a hunting encirclement model of a prey hunted by the male dhole and the female dhole together;

[0194] obtain the updated position of the individual based on the adaptive weight position of the third relative position and the fourth relative position.

[0195] In some embodiments, the apparatus 400 further comprises a fitness obtaining module 404, which is specifically configured to:

[0196] train the deep learning model based on the position of the individual to determine the fitness of the individual; wherein the fitness is a prediction accuracy, an F1 score, a loss value or a root mean square error of the deep learning model.

[0197] The model training apparatus 400 provided by the embodiments of the present application is used to execute the technical solutions provided by the foregoing model training method embodiments, and has similar implementation principles and technical effects to those in the foregoing method embodiments, which will not be described here in detail.

[0198] Figure 5 is a structural schematic diagram of a battery SOC estimation apparatus provided by an embodiment of the present application, referring to Figure 5 The battery SOC estimation apparatus comprises various functional modules for implementing the foregoing battery SOC estimation method, and any functional module can be realized by software and / or hardware.

[0199] In some embodiments, the battery SOC estimation apparatus 500 comprises a data obtaining module 501 and a SOC estimation module 502. Wherein:

[0200] The data obtaining module 501 is configured to obtain battery data of a target battery;

[0201] The SOC estimation module 502 is configured to input the battery data into a battery SOC estimation model to obtain a battery SOC estimation result output by the battery SOC estimation model; wherein the battery SOC estimation model is trained based on the model training method of any one of the first aspect.

[0202] The battery SOC estimation apparatus 500 provided by the embodiments of the present application is used to execute the technical solutions provided by the foregoing battery SOC estimation method embodiments, and has similar implementation principles and technical effects to those in the foregoing method embodiments, which will not be described here in detail.

[0203] It should be noted that the division of each module of the above device is only a logical functional division, and all or part of it can be integrated into a physical entity or physically separated when actually implemented. And these modules can all be implemented in the form of software called by a processing element, or all in the form of hardware, or part of the modules in the form of software called by a processing element and part of the modules in the form of hardware. For example, the model construction module 401 can be a separately established processing element, or can be integrated in a certain chip of the above device, in addition, it can also be stored in the form of program code in the memory of the above device, and the function of the above model construction module 401 is called and executed by a certain processing element of the above device. The implementation of other modules is similar. In addition, all or part of these modules can be integrated together or independently implemented. The processing element here can be an integrated circuit with signal processing capability. In the implementation process, each step of the above method or each module can be completed by the integrated logic circuit of the hardware in the processor element or the instruction in the form of software.

[0204] Figure 6 A structural schematic diagram of an electronic device provided by an embodiment of the present application is shown in FIG. 6. Figure 6 The electronic device 600 includes a processor 601 and a memory 602 in communication with the processor 601.

[0205] The memory 602 stores computer execution instructions.

[0206] The processor 601 executes the computer execution instructions stored in the memory 602 to implement the technical solutions of the above model training method or battery SOC estimation method.

[0207] In the electronic device 600 described above, the memory 602 and the processor 601 are directly or indirectly electrically connected to realize data transmission or interaction. For example, these elements can be electrically connected to each other through one or more communication buses or signal lines, such as through bus connection. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc., but it does not mean that there is only one bus or one type of bus. The memory 602 stores computer execution instructions for implementing the foregoing model training method or battery SOC estimation method, including at least one software function module stored in the memory 602 in the form of software or firmware, and the processor 601 executes various function applications and data processing by running the software program and module stored in the memory 602.

[0208] The memory 602 at least includes one type of readable storage medium, not limited to a Random Access Memory (RAM), a Read Only Memory (ROM), a Programmable Read-Only Memory (PROM), an Erasable Programmable Read-Only Memory (EPROM), an Electric Erasable Programmable Read-Only Memory (EEPROM), etc. Among them, the memory 602 is used to store programs, and the processor 601 executes the programs after receiving execution instructions. Further, the software program and module in the memory 602 described above can also include an operating system, which can include various software components and / or drivers for managing system tasks (such as memory management, storage device control, power management, etc.), and can communicate with various hardware or software components to provide an operating environment for other software components.

[0209] The processor 601 can be an integrated circuit chip with a processing capability of signals. The processor 601 described above can be a general processor, including a central processing unit (CPU), a network processor (NP), a digital signal processor (DSP), an application specific integrated circuit (ASIC), and the like. The processor can implement or execute the disclosed methods, steps, and logic block diagrams in the embodiments of the present application. The general processor can be a microprocessor, or the processor 601 can also be any conventional processor or the like.

[0210] The electronic device 600 is configured to execute the technical solutions provided by the foregoing model training method or battery SOC estimation method embodiments, and the implementation principles and technical effects are similar to those of the foregoing method embodiments, which will not be described here again.

[0211] The embodiments of the present application further provide a computer readable storage medium, and the computer readable storage medium stores computer execution instructions. When the computer execution instructions are executed, the technical solutions of the foregoing model training method or battery SOC estimation method are implemented.

[0212] The computer readable storage medium described above can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic storage, a flash memory, a magnetic disk or an optical disk. The computer readable storage medium can be any available medium that can be accessed by a general or special purpose computer.

[0213] An exemplary readable storage medium is coupled to the processor, so that the processor can read information from the readable storage medium and write information to the readable storage medium. Of course, the readable storage medium can also be an integral part of the processor. The processor and the readable storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the readable storage medium can also exist as discrete components in the control device of the model training device or the battery SOC estimation device.

[0214] The embodiment of the present application further provides a computer program product comprising a computer program which, when executed, is configured to implement the technical solutions of the aforementioned model training method or the battery SOC estimation method.

[0215] In the above embodiments, those skilled in the art can understand that the implementation of the above-mentioned method embodiments can be realized by software, hardware, firmware or any combination thereof, in whole or in part. When implemented by software, it can be realized in whole or in part in the form of a computer program product. The computer program product comprises one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the flow or function according to the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another, for example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through wired (for example, coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (for example, infrared, wireless network, microwave, etc.) mode. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. containing one or more available media sets. The available media can be magnetic media (for example, floppy disk, hard disk, magnetic tape), optical media (for example, DVD), or semiconductor media (for example, Solid State Disk (SSD)) and the like.

[0216] In the above embodiments, the description of each embodiment has its own focus, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments. The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described, however, as long as the combination of these technical features does not exist contradictory, it should be considered as the scope of the present application.

[0217] Other embodiments of the application will be apparent to those skilled in the art from consideration of the specification and practice of the application disclosed herein. The specification and examples given are exemplary only and the true scope and spirit of the application is indicated by the claims which follow. The specification and examples are to be regarded as illustrative only, the true scope and spirit of the application being indicated by the appended claims.

[0218] It is to be understood that the application is not limited to the precise construction already described above and shown in the drawings, and that various modifications and changes can be made by those skilled in the art without departing from the scope of the application. The scope of the application should only be limited by the claims appended hereto.

Claims

1. A model training method, characterized in that, include: Build deep learning models; The hyperparameters of the deep learning model are optimized based on the enhanced golden jackal optimization algorithm to determine the optimal hyperparameters of the deep learning model. The deep learning model is trained and tested based on a pre-built training set and a test set to obtain a battery SOC estimation model; wherein, the battery SOC estimation model is used to obtain battery SOC estimation results based on battery data of the target battery.

2. The method according to claim 1, characterized in that, The hyperparameters of the deep learning model are optimized based on the enhanced golden jackal optimization algorithm to determine the optimal hyperparameters of the deep learning model, including: Initialize the population; wherein the population comprises multiple individuals, the position of the individuals is used to characterize the combination of hyperparameters, and the initial individuals are uniformly distributed in the search space; If the iteration stopping condition is not met, the iteration is performed to update the position of individuals in the population based on their fitness; wherein the iteration stopping condition includes at least one of the current iteration number being equal to the maximum iteration number or the position of the individual converging; When the iteration stopping condition is met, the position of the individual with the highest fitness in the current population is output as the optimal hyperparameter of the deep learning model.

3. The method according to claim 2, characterized in that, The position of individuals in the population is updated based on their fitness, including: Based on the fitness of individuals in the population, male and female golden jackals are identified; wherein the male golden jackal is the individual with the highest fitness, and the female golden jackal is the individual with the second highest fitness. The escape energy of prey in the population is updated; wherein, the prey are all individuals in the population other than the male and female golden jackals; If the escape energy is greater than a preset threshold, the golden jackal is in the exploration phase, and the position of individuals in the population is updated based on the exploration strategy; If the escape energy is less than a preset threshold, the golden jackal is in the encirclement phase, and the position of individuals in the population is updated based on the encirclement strategy.

4. The method according to claim 3, characterized in that, The escape energy satisfies the following formula: Where E represents the escape energy. This represents the initial value of an individual's escape energy. Let represent the energy decrease function, r represent a random coefficient in the range [0, 1], C1 represent a constant coefficient with a value of 1.5, t represent the current iteration number, and T represent the maximum iteration number.

5. The method according to claim 3, characterized in that, The location of individuals in the population is updated based on an exploration strategy, including: Based on a prey exploration model where the male golden jackal leads and the female golden jackal follows, the first relative position and the second relative position of the individual relative to the male golden jackal and the female golden jackal are updated respectively. The updated position of the individual is obtained based on the average position of the first relative position and the second relative position.

6. The method according to claim 3, characterized in that, The location of individuals in the population is updated based on the encirclement strategy, including: Based on the prey encirclement model of male and female golden jackals hunting together, the third and fourth relative positions of the individual with respect to the male and female golden jackals are updated respectively. The updated position of the individual is obtained based on the adaptive weighted position of the third relative position and the fourth relative position.

7. The method according to any one of claims 2-6, characterized in that, The method further includes: The deep learning model is trained based on the location of the individual to determine the fitness of the individual; wherein the fitness is the prediction accuracy, F1 score, loss value, or root mean square error of the deep learning model.

8. A method for estimating battery SOC, characterized in that, include: Obtain battery data for the target battery; The battery data is input into the battery SOC estimation model to obtain the battery SOC estimation result output by the battery SOC estimation model; wherein, the battery SOC estimation model is trained based on the model training method of any one of claims 1-7.

9. An electronic device, characterized in that, Includes a processor and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed, are used to implement the method as described in any one of claims 1 to 8.

Images