Battery pack dominant frequency prediction method and device based on neural network, and electronic equipment

Abstract

The invention discloses a battery pack dominant frequency prediction method and device based on a neural network and electronic equipment, and relates to the technical field of battery design, and the method comprises the steps: obtaining historical battery pack project data of a vehicle, and constructing a dominant frequency prediction original data set based on the historical battery pack project data; constructing a target neural network prediction model based on the dominant frequency prediction original data set; obtaining a plurality of design features of the current battery pack, wherein the design features are macroscopic variable data of the current battery pack in an early total arrangement design stage; and inputting the plurality of design features into the target neural network prediction model to predict and output the battery pack dominant frequency of the current battery pack. Therefore, according to the method, battery pack dominant frequency prediction can be performed based on the macroscopic variables in the early design stage, the cost of later design change and scheme optimization is reduced, the engineering development efficiency is remarkably improved, the development cost is reduced, meanwhile, the test cost is reduced, and the problem of test hysteresis in related technologies is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of battery pack design, and particularly relates to a battery pack main frequency prediction method based on a neural network, an electronic device, a computer readable storage medium and a battery pack main frequency prediction device based on a neural network. BACKGROUND

[0002] The main frequency of a battery pack, also commonly known as the first-order natural frequency or the first-order modal frequency, is an important parameter of the structural dynamics characteristics of the battery pack, and reflects the lowest frequency at which the battery pack resonates when subjected to external excitation. The main frequency has an important influence on the structural safety, comfort and fatigue life of the battery pack.

[0003] In the related art, the method for obtaining the main frequency of the battery pack mainly relies on two means, namely, test and measurement based on trial assembly and modal simulation analysis based on finite elements. However, the test cost of this method is high, and there is a certain time lag. SUMMARY

[0004] The present application aims to at least solve one of the technical problems in the related art. To this end, the first purpose of the present application is to propose a battery pack main frequency prediction method based on a neural network, which predicts the main frequency of the battery pack based on macro variables in the early design stage, reduces the cost of later design changes and scheme optimization, significantly improves the engineering development efficiency and reduces the development cost, and at the same time, the main frequency prediction is completed based on a target neural network prediction model, which reduces the test cost and solves the test lag problem in the related art.

[0005] The second purpose of the present application is to propose an electronic device.

[0006] The third purpose of the present application is to propose a computer readable storage medium.

[0007] The fourth purpose of the present application is to propose a battery pack main frequency prediction device based on a neural network.

[0008] To achieve the above objectives, the first aspect of this application proposes a method for predicting the main frequency of a battery pack based on a neural network. The method includes: acquiring historical battery pack project data of a vehicle; constructing a primary frequency prediction dataset based on the historical battery pack project data, wherein the primary frequency prediction dataset includes dependent variable data and independent variable data, the dependent variable data being battery pack main frequency data, and the independent variable data being macroscopic variable data of the battery pack during the early overall layout design stage; constructing a target neural network prediction model based on the primary frequency prediction dataset; acquiring multiple design features of the current battery pack, wherein the design features are macroscopic variable data of the current battery pack during the early overall layout design stage; and inputting the multiple design features into the target neural network prediction model to predict and output the main frequency of the current battery pack.

[0009] According to the neural network-based battery pack frequency prediction method of this application embodiment, firstly, historical battery pack project data of the vehicle is acquired. Based on this historical data, a primary frequency prediction dataset is constructed. This dataset includes dependent variable data and independent variable data. The dependent variable is the battery pack frequency data, and the independent variable is the macroscopic variable data of the battery pack during the early overall layout design stage. A target neural network prediction model is then constructed based on this dataset. During the battery pack design stage, multiple design features of the current battery pack are acquired and input into the target neural network prediction model to predict the battery pack frequency. Therefore, this method can predict the battery pack frequency based on macroscopic variables from the early design stage, reducing the cost of later design changes and scheme optimization, significantly improving engineering development efficiency and reducing development costs. Furthermore, by completing the frequency prediction based on the target neural network prediction model, testing costs are reduced, and the testing lag problem existing in related technologies is solved.

[0010] In addition, the battery pack main frequency prediction method according to the above embodiments of this application may also have the following additional technical features: According to one embodiment of this application, constructing a target neural network prediction model based on a primary frequency prediction raw data set includes: extracting features from the primary frequency prediction raw data set based on the kernel principal component analysis function and preset conditions to obtain a sample data set; preprocessing the sample data set to obtain training data samples; determining the input and output parameters of a preset neural network architecture and the range of model hyperparameter values ​​to construct a preset neural network model; training the preset neural network model based on the training data samples, and automatically optimizing the parameter variables corresponding to the range of model hyperparameter values ​​based on a target optimization algorithm during the training process to obtain the target neural network prediction model.

[0011] According to one embodiment of this application, feature extraction is performed on the original dataset of main frequency prediction based on the kernel principal component analysis function and preset conditions to obtain a sample dataset, including: feature extraction on the original dataset of main frequency prediction based on the kernel principal component analysis function to obtain a candidate independent variable feature set; feature selection on the candidate independent variable feature set based on preset conditions to obtain a target independent variable feature set; and constructing a sample dataset based on the target independent variable feature set and the corresponding dependent variable feature, wherein the dependent variable feature is the battery pack main frequency.

[0012] According to one embodiment of this application, the preset conditions include that the influence coefficient of the independent variable features in the target independent variable feature set on the battery pack stiffness is greater than a preset coefficient, and that there is no fixed physical mechanism relationship between the multiple independent variable features in the target independent variable feature set.

[0013] According to one embodiment of this application, data preprocessing is performed on a sample dataset to obtain training data samples, including: determining multiple independent variable feature matrices and corresponding target variable features based on the sample dataset; standardizing each independent variable feature matrix to scale each independent variable feature matrix to the target interval; and constructing training data samples based on the standardized multiple independent variable feature matrices and corresponding target variable features.

[0014] According to one embodiment of this application, the training data sample includes multiple independent variable feature matrices and corresponding target variable features. The input and output parameters of a preset neural network architecture and the range of model hyperparameter values ​​are determined to construct a preset neural network model. This includes: determining the number of input layer neurons in the preset neural network architecture based on the number of features in the independent variable feature matrices; determining the number of output layer neurons in the preset neural network architecture based on the number of features in the target variable features; determining the number of neurons in the first hidden layer of the preset neural network architecture based on the number of input layer neurons and the number of output layer neurons; and determining the preset neural network model based on the number of input layer neurons, the number of output layer neurons, the number of neurons in the first hidden layer, and the range of model hyperparameter values. The range of model hyperparameter values ​​includes one or more of the following: the range of the number of hidden layers, the range of the dropout rate, the range of the optimizer selection, the range of the learning rate, and the range of the batch size.

[0015] According to one embodiment of this application, the target optimization algorithm is a genetic algorithm. During the training process, the target optimization algorithm is used to automatically optimize the parameter variables corresponding to the range of hyperparameter values ​​of the model to obtain the target neural network prediction model. The algorithm includes: designing a genetic algorithm and constructing a fitness function of the genetic algorithm based on mean squared error, coefficient of determination, and mean absolute percentage error; determining the size of the genetic algorithm and determining the initial population based on the fitness function; performing population iteration based on the genetic algorithm, starting from the initial population, and obtaining the iteration generation and fitness value corresponding to each iteration; determining the best individual based on the iteration generation and fitness value corresponding to each iteration; and constructing the target neural network prediction model based on the hyperparameters of the best individual.

[0016] To achieve the above objectives, a second aspect of this application provides an electronic device, including a memory, a processor, and a battery pack frequency prediction program stored in the memory and executable on the processor. When the processor executes the battery pack frequency prediction program, it implements the above-described battery pack frequency prediction method.

[0017] The electronic device according to the embodiments of this application implements the above-mentioned battery pack frequency prediction method when the processor executes the battery pack frequency prediction program. Based on the above-mentioned battery pack frequency prediction method, the cost of later design changes and scheme optimization is reduced, the engineering development efficiency is significantly improved, the development cost is reduced, and the testing cost is reduced, thus solving the testing lag problem existing in related technologies.

[0018] To achieve the above objectives, a third aspect of this application provides a computer-readable storage medium storing a battery pack frequency prediction program thereon, which, when executed by a processor, implements the above-described battery pack frequency prediction method.

[0019] According to the embodiments of this application, a computer-readable storage medium stores a battery pack frequency prediction program thereon. When the program is executed by a processor, the battery pack frequency prediction method described above is implemented. Based on the battery pack frequency prediction method described above, the cost of later design changes and scheme optimization is reduced, the engineering development efficiency is significantly improved, the development cost is reduced, and the testing cost is also reduced, thus solving the problem of testing lag in related technologies.

[0020] To achieve the above objectives, a fourth aspect of this application proposes a battery pack main frequency prediction device based on a neural network. The device includes: a first acquisition module, configured to acquire historical battery pack project data of a vehicle and construct a main frequency prediction raw data set based on the historical battery pack project data, wherein the main frequency prediction raw data set includes dependent variable data and independent variable data, the dependent variable data being battery pack main frequency data and the independent variable data being macroscopic variable data of the battery pack in the early stage of overall layout design; a construction module, configured to construct a target neural network prediction model based on the main frequency prediction raw data set; a second acquisition module, configured to acquire multiple design features of the current battery pack, wherein the design features are macroscopic variable data of the current battery pack in the early stage of overall layout design; and a prediction module, configured to input the multiple design features into the target neural network prediction model to predict and output the battery pack main frequency of the current battery pack.

[0021] According to the battery pack frequency prediction device of this application embodiment, a first acquisition module acquires historical battery pack project data of the vehicle, and constructs a primary frequency prediction raw data set based on the historical battery pack project data. The primary frequency prediction raw data set includes dependent variable data and independent variable data. The dependent variable data is the battery pack frequency data, and the independent variable data is the macroscopic variable data of the battery pack in the early stage of overall layout design. A construction module constructs a target neural network prediction model based on the primary frequency prediction raw data set. A second acquisition module acquires multiple design features of the current battery pack, which are macroscopic variable data of the current battery pack in the early stage of overall layout design. A prediction module inputs the multiple design features into the target neural network prediction model to predict and output the battery pack frequency of the current battery pack. Therefore, this device predicts the battery pack frequency based on macroscopic variables from the early design stage, reducing the cost of later design changes and scheme optimization, significantly improving engineering development efficiency and reducing development costs. Simultaneously, completing the frequency prediction based on the target neural network prediction model reduces testing costs and solves the testing lag problem existing in related technologies.

[0022] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0023] Figure 1 This is a flowchart of a battery pack main frequency prediction method based on a neural network according to an embodiment of this application; Figure 2 This is a schematic diagram of the structure of a battery pack according to an embodiment of this application; Figure 3 This is a schematic diagram of the mean square error according to an embodiment of this application; Figure 4 This is a schematic diagram of a neural network architecture according to an embodiment of this application; Figure 5 This is a flowchart of a battery pack frequency prediction method according to an embodiment of this application; Figure 6 This is a block diagram of an electronic device according to an embodiment of this application; Figure 7 This is a connection diagram of a neural network-based battery pack frequency prediction device according to an embodiment of this application. Detailed Implementation

[0024] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0025] The following description, with reference to the accompanying drawings, outlines a battery pack frequency prediction method, prediction device, electronic device, and computer-readable storage medium.

[0026] As the "heart" of new energy vehicles, the power battery is the primary power source and a key component of electric vehicles. Battery pack design not only needs to consider energy storage capacity but also structural strength, collision safety, fatigue reliability, ventilation and heat dissipation, insulation and waterproofing, and electromagnetic compatibility. The battery pack's dominant frequency, also known as the first-order natural frequency or first-order modal frequency, is an important parameter of the battery pack's structural dynamics. It reflects the lowest frequency at which the battery pack's structure resonates under external excitation, and has a significant impact on the battery pack's structural safety, comfort, and fatigue life.

[0027] In related technologies, the following two methods are used to obtain the main frequency of the battery pack, as detailed below: 1. Test based on prototype assembly: Test is conducted by physical sample to obtain the dynamic response of the battery pack. This test method has the advantages of being intuitive and accurate, but it has the following significant drawbacks: (1) It depends on product prototype, which requires the manufacture of physical samples, resulting in high costs; (2) Time lag: The test needs to be carried out in the later stage of product development and cannot provide predictions in the early stage of the project; (3) High resource consumption: A large number of sample tests consume a lot of resources and time.

[0028] 2. Finite Element-Based Modal Simulation Analysis: Modal simulation analysis utilizes a finite element model to analyze the dynamic characteristics of the battery pack, calculating its natural frequencies and mode shapes. This avoids the need for physical prototype fabrication and the debugging and installation of experimental equipment, saving time and costs. While this method offers convenience, flexibility, and scalability, the modeling process still involves significant time consumption due to the precise application of parameters such as mesh generation, material properties, and boundary conditions. Furthermore, the large deformation nonlinearity and material nonlinearity inherent in battery pack simulation make the modeling and calculation processes time-consuming. The accuracy of the simulation results depends heavily on the model's precision and the adequacy of computational resources, and is also heavily reliant on detailed 3D structural design input. Due to the complexity of battery pack design, simulation analysis often requires multiple iterations, each consuming substantial time and computational resources. This makes it impossible to provide the dominant frequency as a reference for the overall layout design during the early stages of the project.

[0029] To address at least one of the aforementioned technical problems, this application proposes a method for predicting the main frequency of a battery pack during the early design phase. This method enables main frequency prediction, thereby reducing the cost of later design changes and scheme optimization, significantly improving engineering development efficiency, and lowering development costs, thus possessing significant practical engineering value. The battery pack main frequency prediction method of this application will be described in detail below with reference to the accompanying drawings.

[0030] Figure 1 This is a flowchart of a neural network-based battery pack frequency prediction method according to an embodiment of this application.

[0031] like Figure 1 As shown, the battery pack main frequency prediction method based on neural networks in this application embodiment may include: S1. Obtain historical battery pack project data for the vehicle, and construct a primary frequency prediction raw data set based on the historical battery pack project data. The primary frequency prediction raw data set includes dependent variable data and independent variable data. The dependent variable data is the battery pack primary frequency data, and the independent variable data is the macroscopic variable data of the battery pack in the early stage of overall layout design. In other words, based on the battery pack project database, historical data is organized to establish a raw dataset for predicting the main frequency of new energy vehicle battery packs. The variable data extracted and organized from this raw dataset only includes macroscopic variable parameters from the early overall layout design stage. The dependent variable data is the prediction target corresponding to each sample data, i.e., the first-order main frequency. Independent variables include, but are not limited to: material properties, such as elastic modulus and Poisson's ratio; macroscopic parameters of structural design, such as length, width, height, and wall thickness; connection methods, such as the connection method between internal modules and the outer shell, and the installation and fixing method and number of fixing points between the outer shell and the vehicle; the number of crossbeams, longitudinal beams, and fixing points that affect the overall stiffness; and the total weight of the battery pack.

[0032] The historical battery pack project data can include two types of vehicle models: pure electric vehicles and hybrid electric vehicles. The vehicle class includes A, B, and C classes, which covers both sedans and SUVs (Sport Utility Vehicles).

[0033] S2, Construct a target neural network prediction model based on the original dataset of the main frequency prediction; In other words, a neural network model is trained using the original dataset of the main frequency prediction to obtain a target neural network prediction model that can be used for battery pack main frequency prediction. The input of this target neural network prediction model is the independent variable features in the original dataset of the main frequency prediction, and the output is the battery pack main frequency, i.e., the dependent variable features.

[0034] S3, obtain multiple design features of the current battery pack, where the design features are macroscopic variable data of the current battery pack in the early stage of overall layout design; In other words, during the battery pack design phase, after obtaining the actual macroscopic variables of the new battery pack in the early stage of overall layout design, multiple design features of the current battery pack are acquired, such as the material properties of the battery pack, macroscopic parameters of structural design, connection methods, and total weight of the battery pack.

[0035] S4 inputs multiple design features into the target neural network prediction model to predict the battery pack frequency of the current battery pack.

[0036] In other words, the new battery pack parameter combination, i.e., multiple design features, is used as input conditions and input into the target neural network prediction model. The target neural network prediction model predicts the battery pack's main frequency and outputs the main frequency value of one section of the current battery pack. The obtained battery pack main frequency can be used for model design of the current battery pack, thereby reducing the cost of later design changes and scheme optimization, significantly improving engineering development efficiency and reducing development costs. At the same time, the main frequency prediction based on the target neural network prediction model reduces testing costs and solves the testing lag problem existing in related technologies.

[0037] In some embodiments of this application, constructing a target neural network prediction model based on the original dataset of primary frequency prediction includes: extracting features from the original dataset of primary frequency prediction based on the kernel principal component analysis function and preset conditions to obtain a sample dataset; preprocessing the sample dataset to obtain training data samples; determining the input and output parameters of the preset neural network architecture and the range of model hyperparameter values ​​to construct a preset neural network model; training the preset neural network model based on the training data samples, and automatically optimizing the parameter variables corresponding to the range of model hyperparameter values ​​based on the target optimization algorithm during the training process to obtain the target neural network prediction model.

[0038] In other words, based on principal component analysis and expert judgment, the dimensionality of the independent variables is reduced. The feature components X = {x1, x2, … xn} of the independent variables to be used in training and the first-order principal frequency Y of the dependent variable are extracted from the original dataset, i.e., the original dataset of the principal frequency prediction. The dataset required for training the artificial neural network with N samples is constructed to obtain the sample data set.

[0039] Then, the sample dataset is preprocessed, including reading sample data from files, dividing it into training, validation and test sets, and standardizing the data to obtain training data samples.

[0040] Define the input and output of the neural network architecture to construct a predefined neural network model. Define the range of values ​​for the model's hyperparameter variables. The number of neurons in the input layer equals the number of features determined in the sample data set, and the number of neurons in the output layer equals the number of predicted target variables Y, which is 1. Other parameters to be optimized and their value ranges include: the number of hidden layers in the neural network. The number of neurons in the first hidden layer Drop rate Optimizer Learning rate Batch size An automatic optimization algorithm is constructed to automatically optimize the parameters in the model that have a range of values ​​during the training process of a preset neural network model based on training data samples. That is, the parameter variables corresponding to the range of values ​​of the model hyperparameters are automatically optimized to obtain the target neural network prediction model. This avoids a large amount of manual architecture design and trial and error process and improves the efficiency of model building.

[0041] In some embodiments of this application, feature extraction is performed on the original dataset of main frequency prediction based on the kernel principal component analysis function and preset conditions to obtain a sample dataset, including: feature extraction on the original dataset of main frequency prediction based on the kernel principal component analysis function to obtain a candidate independent variable feature set; feature selection on the candidate independent variable feature set based on preset conditions to obtain a target independent variable feature set; and constructing a sample dataset based on the target independent variable feature set and the corresponding dependent variable feature, wherein the dependent variable feature is the battery pack main frequency.

[0042] Specifically, the kernel function in kernel principal component analysis can be a nonlinear type such as a Gaussian kernel function or a polynomial kernel function. The purpose of the preset conditions is to select variables that have a significant impact on the battery pack stiffness, and to discard variables with fixed physical mechanism relationships. For example, since mass equals the product of volume and material density, mass and structural parameters can be selected as independent variables while material density is discarded. Thus, by sequentially using the kernel principal component analysis function and preset conditions, feature extraction and filtering are performed on the original dataset for main frequency prediction to construct a sample dataset, thereby reducing data redundancy while ensuring the effectiveness of the sample dataset.

[0043] For example, in one embodiment of this application, the principal components of the independent variables are the first nine principal factors, i.e., X = {x1, x2, … x9}, where x1-x9 represent the weight of the battery pack, respectively. ,length ,width ,thickness The number of front and rear lifting points 1, the number of side lifting points 2, the number of middle lifting points 3, the number of crossbeams 4, and the number of longitudinal beams 5 are shown in the diagram. Figure 2 As shown in Table 1, the dependent variable Y represents the first-order dominant frequency, the sample size N=200, and an example dataset is shown in Table 1.

[0044] Table 1

[0045] In some embodiments of this application, the preset conditions include that the influence coefficient of the independent variable features in the target independent variable feature set on the battery pack stiffness is greater than a preset coefficient, and that there is no fixed physical mechanism relationship between the multiple independent variable features in the target independent variable feature set.

[0046] In other words, the preset conditions are set based on expert experience and knowledge to select variables that have a significant impact on the stiffness of the battery pack, as well as to select variables that have a fixed physical mechanism relationship with each other. For example, since mass is equal to the product of volume and material density, mass and structural parameters can be selected as independent variables while material density is discarded, thereby filtering the data.

[0047] In some embodiments of this application, data preprocessing is performed on the sample data set to obtain training data samples, including: determining multiple independent variable feature matrices and corresponding target variable features based on the sample data set; performing standardization processing on each independent variable feature matrix to scale each independent variable feature matrix to the target interval; and constructing training data samples based on the standardized multiple independent variable feature matrices and corresponding target variable features.

[0048] In other words, data preprocessing is performed on the sample dataset, including reading sample data from files, dividing it into training, validation, and test sets, and standardizing the data.

[0049] Specifically, sample data can be stored in regular files such as .xlsx, .csv, and .tx. The Pandas library in Python is used to read the data, obtain the sample data set, and transform it into a feature matrix X (i.e., the independent variable feature matrix) and a target variable Y (i.e., the target variable feature). The ratio of the training set, validation set, and test set is chosen to be 7:2:1. Furthermore, 80% of the sample data contained in the training set is further refined into training set and validation set.

[0050] The data standardization method uses the following formula (1) to standardize the feature matrix, thereby scaling all features to the [0,1] interval.

[0051] (1) In this embodiment, the data structures of the feature matrix X and the target variable Y are (200, 9) and (200, 1), respectively. Then, training data samples are constructed based on the standardized feature matrices of multiple independent variables and the corresponding target variable features to ensure the model training effect.

[0052] In some embodiments of this application, the training data samples include multiple independent variable feature matrices and corresponding target variable features. The input and output parameters of a preset neural network architecture and the range of model hyperparameter values ​​are determined to construct a preset neural network model. This includes: determining the number of input layer neurons in the preset neural network architecture based on the number of features in the independent variable feature matrices; determining the number of output layer neurons in the preset neural network architecture based on the number of features in the target variable features; determining the number of neurons in the first hidden layer of the preset neural network architecture based on the number of input layer neurons and the number of output layer neurons; and determining the preset neural network model based on the number of input layer neurons, the number of output layer neurons, the number of neurons in the first hidden layer, and the range of model hyperparameter values. The range of model hyperparameter values ​​includes one or more of the following: the range of the number of hidden layers, the range of the dropout rate, the range of the optimizer selection, the range of the learning rate, and the range of the batch size.

[0053] Specifically, the input and output of the neural network architecture are defined as follows: the number of neurons in the input layer equals the number of features determined in the feature matrix of independent variables, and the number of neurons in the output layer equals the number of predicted target variables Y, which is 1. Other parameters to be optimized and their value ranges include: the number of hidden layers in the neural network. The number of neurons in the first hidden layer Drop rate Optimizer Learning rate Batch size The number of neurons in the first hidden layer is calculated according to the formula. Perform an estimate and round it down, where, is the number of hidden layer neurons, input is the number of input layer neurons, output is the number of output layer neurons, and c is a constant between [1, 10]. A pre-defined neural network model is constructed based on the number of input layer neurons, the number of output layer neurons, the number of neurons in the first hidden layer, and the range of model hyperparameter values.

[0054] For example, in one embodiment, the number of hidden layers ranges from 1 to 3, the number of neurons in the first hidden layer ranges from 4 to 16, the dropout rate ranges from 0.1 to 0.5, the optimizer is selected from one of Adam (Adaptive Moment Estimation), RMSprop (Root Mean Square Propagation), and SGD (Stochastic Gradient Descent), the learning rate ranges from 0.0001 to 0.01, and the batch size ranges from 3 to 20.

[0055] In some embodiments of this application, the target optimization algorithm is a genetic algorithm. During training, the target optimization algorithm is used to automatically optimize the parameter variables corresponding to the range of hyperparameter values ​​of the model to obtain the target neural network prediction model. This includes: designing a genetic algorithm and constructing a fitness function of the genetic algorithm based on mean squared error, coefficient of determination, and mean absolute percentage error; determining the size of the genetic algorithm and determining the initial population based on the fitness function; performing population iteration based on the genetic algorithm, starting from the initial population, and obtaining the iteration generation and fitness value corresponding to each iteration; determining the best individual based on the iteration generation and fitness value corresponding to each iteration; and constructing the target neural network prediction model based on the hyperparameters of the best individual.

[0056] Specifically, firstly, an automatic optimization algorithm is constructed to automatically optimize parameter variables that have a range of values: The objective optimization algorithm employs a genetic algorithm. In a genetic algorithm, an individual represents the configuration of a neural network, including the number of layers, the number of neurons in the first hidden layer, the dropout rate, the optimizer, the learning rate, and the batch size. The genetic algorithm population is a collection of multiple individuals.

[0057] The fitness function of the genetic algorithm used in this embodiment is to assign weights to the mean squared error, coefficient of determination, and mean absolute percentage error. It is constructed by linear accumulation, as shown in the following formula.

[0058]

[0059] In the fitness function, the number of neurons in the hidden layers of the neural network, excluding the first layer, is determined by halving the number of neurons in the first layer and then rounding down. For example, if the number of neurons in the first hidden layer is Neu, then the number of neurons in the second hidden layer is [Neu / 2], and so on. The dropout rate is set using the same strategy.

[0060] Furthermore, the epoch value required for training in the fitness function is a fixed value. For example, in this embodiment, the weight coefficients are respectively set to... The epoch value is 1000.

[0061] Then, set the initial population size and the number of generations for the genetic algorithm iterations, and create the initial individuals and population using a random initialization method: For example, in this embodiment, the initial population size is P, and the number of generations is set to G. The values ​​of the number of hidden layers, the number of neurons in the first hidden layer, the dropout rate, the optimizer, the learning rate, and the batch size for each individual in the initial population are all randomly selected within a set range. Taking an initial population size P of 10 and a generation G of 100 as an example in this embodiment, an example of initializing the individual values ​​of 10 individuals in the population is shown in Table 2.

[0062] Table 2

[0063] Then, the genetic generations are iterated repeatedly to find the optimal individual: The current population undergoes selection, crossover, and mutation operations to generate a new generation. The fitness values ​​of the new generation are calculated and recorded in the corresponding individual attributes. Statistical information of the new generation is calculated and recorded, such as maximum fitness, minimum fitness, average fitness, and corresponding mean squared error, coefficient of determination, and mean absolute percentage error. The loop terminates when a set number of generations G is reached or the fitness function value is less than a predetermined value (i.e., the optimal individual meeting the accuracy requirements has been found).

[0064] The selection of individuals in the current population can be performed using tournament selection methods, such as the ternary tournament selection method. Crossover can be performed using a two-point crossover method, randomly selecting two points from the gene sequences of two parent individuals and then exchanging the gene segments between these two points. Mutation can be performed using a custom mutation function, randomly changing various parameters of the individual. Integer parameters (such as the number of layers, number of neurons, and batch size) are randomly generated with new values ​​within their respective ranges; floating-point parameters (such as dropout rate and learning rate) are also randomly generated with new values ​​within their respective ranges. Mutation helps prevent the population from getting trapped in local optima, thereby increasing the probability of finding the global optimum. For example, the crossover probability and mutation probability can be set to 0.5 and 0.2 respectively, with the loop termination condition being G=100 or the fitness value being less than or equal to 2.0.

[0065] Then, based on the statistical information of the population recorded above, the best individual is output: The optimal individual is defined as the individual corresponding to a predetermined fitness value if the loop terminates when the fitness reaches a predetermined value. If the loop terminates when the predetermined value of the algebra G is reached, the individual corresponding to the minimum fitness value recorded throughout the entire iteration process is output. The mean squared error, coefficient of determination, and mean absolute percentage error corresponding to the optimal individual are also output simultaneously.

[0066] Assume that the optimal individual achieves its minimum fitness value at the completion of the 90th iteration. The mean squared error (MSE) of the iterative process is recorded as follows: Figure 3 As shown. The neural network parameters corresponding to the optimal individual are: 2 hidden layers, 16 neurons in the first hidden layer, dropout rate of 0.1, optimizer Adam, learning rate of 0.001, and batch size of 5. The neural network model constructed by the optimal individual is as follows. Figure 3 As shown, it is used as the target neural network prediction model.

[0067] The neural network model is saved, evaluated, and used. Specifically, the neural network model corresponding to the best individual is saved, and its fitness, MSE (Mean Squared Error), R^2 (Coefficient of Determination), and MPAE (Mean Absolute Percentage Error) values ​​are calculated based on the test set to evaluate model performance. A new combination of battery pack parameters is used as input conditions and fed into the obtained target neural network prediction model. The resulting predicted value is the first-order dominant frequency value of the battery pack obtained by this method.

[0068] The neural network model corresponding to the best-preserved individual can be summarized by the following expression:

[0069] In the formula, X0 is the input of the 0th layer, corresponding to the independent variable feature component X, Y0 is the output of the 0th layer, wi is the weight matrix of the i-th layer, bi is the bias matrix of the i-th layer, n is the total number of hidden layers and output layers, and f() is the activation function of the i-th layer.

[0070] For example, the model's performance on the validation set is shown in Table 3, with a fitness value of 1.473, MSE=3.5, R2=0.945, and MAPE=1.068. As can be seen, the method provided in this embodiment can achieve first-order main frequency prediction of battery packs, and the prediction accuracy can fully meet the requirements for prediction accuracy in the early stages of engineering development. However, the prediction efficiency is greatly improved, basically reaching the second level of prediction, and it does not rely on detailed 3D geometric data.

[0071] Table 3

[0072] In this embodiment, a new combination of battery pack parameters is used as a condition input into the model, and the predicted value obtained is the first-order main frequency value of the battery pack obtained by this method. Three sets of examples and their prediction results are shown in Table 4.

[0073] Table 4

[0074] As a specific embodiment of this application, such as Figure 5 As shown, the method for predicting the main frequency of a new energy vehicle battery pack based on artificial neural networks proposed in this application includes the following steps: S101, based on the battery pack project database, organizes historical data and establishes a raw data set for predicting the main frequency of new energy vehicle battery packs.

[0075] The dependent variable data is the prediction target corresponding to each sample data, i.e., the first-order dominant frequency. The independent variables include, but are not limited to: material properties, such as elastic modulus and Poisson's ratio; macroscopic parameters of structural design, such as length, width, height, and wall thickness; connection methods, such as the connection method between internal modules and the outer shell, the installation and fixing method between the outer shell and the whole vehicle, and the number of fixing points; the number of crossbeams and longitudinal beams and the number of fixing points that affect the overall stiffness; and the total weight of the battery pack.

[0076] S102, based on principal component analysis and expert judgment, achieves dimensionality reduction of independent variables, extracts and determines the feature components X = {x1, x2, … xn} of independent variables participating in training from the original data set in S101, and the first-order principal frequency Y of dependent variable, and constructs the dataset required for training the artificial neural network containing N samples.

[0077] S103, perform data preprocessing on the sample dataset. This includes reading sample data from files, dividing the dataset into training, validation, and test sets, and standardizing the data.

[0078] S104 defines the input and output of the neural network architecture and defines the range of values ​​for the model hyperparameter variables.

[0079] The number of neurons in the input layer is equal to the number of features determined in S102, and the number of neurons in the output layer is equal to the number of predicted target variables Y, which is 1. Other parameters to be optimized and their value ranges include: the number of hidden layers in the neural network, the number of neurons in the first hidden layer, the dropout rate, the optimizer, the learning rate, and the batch size.

[0080] S105, Construct an automatic optimization algorithm to automatically optimize the parameter variables in S104 that have a range of values.

[0081] S106 sets the initial population size and the number of generations for the genetic algorithm iterations, and uses a random initialization method to create the initial individuals and population.

[0082] S107, Start iteratively going through generations of the genetic algorithm to find the optimal individual.

[0083] S108 outputs the best individual based on the population statistics recorded in S107.

[0084] S109, Save, evaluate and use the neural network model corresponding to the best individual.

[0085] Specifically, the neural network model corresponding to the best individual in S108 is saved; the fitness, MSE, R^2, and MPAE values ​​are calculated based on the test set to evaluate the model performance; the new battery pack parameter combination is used as input conditions and input into the neural network model obtained in S8, and the predicted value is the first-order main frequency value of the battery pack obtained by this method.

[0086] Therefore, this method has the following technical effects: 1. A method for predicting the main frequency of battery packs in the early design stage. This method enables the prediction of the main frequency, thereby reducing the cost of later design changes and scheme optimization, significantly improving engineering development efficiency and reducing development costs, and has important practical engineering value.

[0087] 2. The battery pack main frequency prediction method based on artificial neural networks provided in this application uses training data based on historical test and simulation results of existing vehicle models and battery pack models, eliminating the need for extensive data generation work.

[0088] 3. In the method provided in this application, the input variables are based only on the overall macroscopic layout parameters of the battery pack, without the need for a detailed battery pack design scheme.

[0089] 4. In the method provided in this application, the artificial neural network model is derived from the automatic optimization of the architecture and hyperparameters of the artificial neural network using a genetic optimization algorithm, which avoids the repetitive work of manually conducting a large number of architecture design attempts and hyperparameter adjustment experiments.

[0090] In summary, the battery pack frequency prediction method based on neural networks according to the embodiments of this application firstly acquires historical battery pack project data of the vehicle, and constructs a primary frequency prediction dataset based on this historical data. This dataset includes dependent variable data and independent variable data. The dependent variable data is the battery pack frequency data, and the independent variable data are macroscopic variable data of the battery pack during the early overall layout design stage. A target neural network prediction model is then constructed based on this dataset. Thus, during the battery pack design stage, multiple design features of the current battery pack are acquired, and these features are input into the target neural network prediction model to predict and output the battery pack frequency. Therefore, this method can predict the battery pack frequency based on macroscopic variables from the early design stage, reducing the cost of later design changes and scheme optimization, significantly improving engineering development efficiency and reducing development costs. Simultaneously, completing the frequency prediction based on the target neural network prediction model reduces testing costs and solves the testing lag problem existing in related technologies.

[0091] Corresponding to the above embodiments, this application also proposes an electronic device.

[0092] like Figure 6 As shown, the electronic device 100 of this application embodiment includes a memory 110, a processor 120, and a battery pack main frequency prediction program stored in the memory 110 and executable on the processor 120. When the processor 120 executes the battery pack main frequency prediction program, it implements the above-mentioned neural network-based battery pack main frequency prediction method.

[0093] The electronic device according to the embodiments of this application implements the above-mentioned battery pack frequency prediction method when the processor executes the battery pack frequency prediction program. Based on the above-mentioned neural network-based battery pack frequency prediction method, the cost of later design changes and scheme optimization is reduced, the engineering development efficiency is significantly improved, the development cost is reduced, and the testing cost is reduced, thus solving the testing lag problem existing in related technologies.

[0094] Corresponding to the above embodiments, this application also proposes a computer-readable storage medium.

[0095] The computer-readable storage medium of this application embodiment stores a battery pack main frequency prediction program thereon. When the battery pack main frequency prediction program is executed by a processor, it implements the above-described neural network-based battery pack main frequency prediction method.

[0096] According to the embodiments of this application, a computer-readable storage medium stores a battery pack frequency prediction program thereon. When the program is executed by a processor, it implements the above-mentioned neural network-based battery pack frequency prediction method. Based on the above-mentioned battery pack frequency prediction method, the cost of later design changes and scheme optimization is reduced, the engineering development efficiency is significantly improved, the development cost is reduced, and the testing cost is reduced. This solves the problem of testing lag in related technologies.

[0097] Corresponding to the above embodiments, this application also proposes a battery pack main frequency prediction device based on neural networks.

[0098] like Figure 7 As shown, the battery pack main frequency prediction device based on neural network according to this application embodiment includes: a first acquisition module 10, a construction module 20, a second acquisition module 30, and a prediction module 40.

[0099] The first acquisition module 10 is used to acquire historical battery pack project data of the vehicle and construct a primary frequency prediction raw data set based on the historical battery pack project data. The primary frequency prediction raw data set includes dependent variable data and independent variable data. The dependent variable data is the battery pack primary frequency data, and the independent variable data is the macroscopic variable data of the battery pack in the early stage of overall layout design. The construction module 20 is used to construct a target neural network prediction model based on the primary frequency prediction raw data set. The second acquisition module 30 is used to acquire multiple design features of the current battery pack. The design features are the macroscopic variable data of the current battery pack in the early stage of overall layout design. The prediction module 40 is used to input the multiple design features into the target neural network prediction model to predict and output the primary frequency of the current battery pack.

[0100] According to one embodiment of this application, the construction module 20 constructs a target neural network prediction model based on the original dataset of the main frequency prediction. Specifically, it is used to: extract features from the original dataset of the main frequency prediction based on the kernel principal component analysis function and preset conditions to obtain a sample dataset; preprocess the sample dataset to obtain training data samples; determine the input and output parameters of the preset neural network architecture and the range of model hyperparameter values ​​to construct a preset neural network model; train the preset neural network model based on the training data samples, and automatically optimize the parameter variables corresponding to the range of model hyperparameter values ​​based on the target optimization algorithm during the training process to obtain the target neural network prediction model.

[0101] According to one embodiment of this application, the construction module 20 performs feature extraction on the original dataset of main frequency prediction based on the kernel principal component analysis function and preset conditions to obtain a sample dataset. Specifically, it is used to: extract features from the original dataset of main frequency prediction based on the kernel principal component analysis function to obtain a set of candidate independent variable features; perform feature screening on the set of candidate independent variable features based on preset conditions to obtain a set of target independent variable features; and construct a sample dataset based on the set of target independent variable features and the corresponding dependent variable features, wherein the dependent variable feature is the battery pack main frequency.

[0102] According to one embodiment of this application, the preset conditions include that the influence coefficient of the independent variable features in the target independent variable feature set on the battery pack stiffness is greater than a preset coefficient, and that there is no fixed physical mechanism relationship between the multiple independent variable features in the target independent variable feature set.

[0103] According to one embodiment of this application, the construction module 20 performs data preprocessing on the sample data set to obtain training data samples, specifically for: determining multiple independent variable feature matrices and corresponding target variable features based on the sample data set; performing standardization processing on each independent variable feature matrix to scale each independent variable feature matrix to the target interval; and constructing training data samples based on the standardized multiple independent variable feature matrices and corresponding target variable features.

[0104] According to one embodiment of this application, the training data sample includes multiple independent variable feature matrices and corresponding target variable features. The construction module 20 determines the input and output parameters of the preset neural network architecture and the range of model hyperparameter values ​​to construct the preset neural network model. Specifically, it is used to: determine the number of input layer neurons of the preset neural network architecture based on the number of features in the independent variable feature matrices; determine the number of output layer neurons of the preset neural network architecture based on the number of features in the target variable features; determine the number of first hidden layer neurons of the preset neural network architecture based on the number of input layer neurons and the number of output layer neurons; and determine the preset neural network model based on the number of input layer neurons, the number of output layer neurons, the number of first hidden layer neurons, and the range of model hyperparameter values. The range of model hyperparameter values ​​includes one or more of the following: the range of the number of hidden layers, the range of the dropout rate, the range of the optimizer selection, the range of the learning rate, and the range of the batch size.

[0105] According to one embodiment of this application, the target optimization algorithm is a genetic algorithm. During the training process, the construction module 20 automatically optimizes the parameter variables corresponding to the range of hyperparameter values ​​of the model based on the target optimization algorithm to obtain the target neural network prediction model. Specifically, it is used to: design a genetic algorithm and construct the fitness function of the genetic algorithm based on the mean square error, the coefficient of determination, and the mean absolute percentage error; determine the size of the genetic algorithm and determine the initial population based on the fitness function; perform population iteration based on the genetic algorithm, starting from the initial population, and obtain the iteration generation and fitness value corresponding to each iteration; determine the best individual based on the iteration generation and fitness value corresponding to each iteration; and construct the target neural network prediction model based on the hyperparameters of the best individual.

[0106] It should be noted that for details not disclosed in the neural network-based battery pack frequency prediction device of this application embodiments, please refer to the details disclosed in the neural network-based battery pack frequency prediction method of the above embodiments of this application, which will not be repeated here.

[0107] According to the embodiment of this application, a neural network-based battery pack frequency prediction device acquires historical battery pack project data of a vehicle through a first acquisition module. Based on this historical data, a raw data set for frequency prediction is constructed. This raw data set includes dependent variable data and independent variable data. The dependent variable data is the battery pack frequency data, and the independent variable data are macroscopic variable data of the battery pack during the early overall layout design stage. A construction module builds a target neural network prediction model based on this raw data set. A second acquisition module acquires multiple design features of the current battery pack, which are macroscopic variable data of the current battery pack during the early overall layout design stage. The prediction module inputs these design features into the target neural network prediction model to predict and output the battery pack frequency of the current battery pack. Therefore, this device predicts the battery pack frequency based on macroscopic variables from the early design stage, reducing the cost of later design changes and scheme optimization, significantly improving engineering development efficiency and reducing development costs. Simultaneously, by completing the frequency prediction based on the target neural network prediction model, testing costs are reduced, and the testing lag problem existing in related technologies is solved.

[0108] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0109] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0110] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0111] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0112] In this application, unless otherwise expressly specified and limited, the terms "installation," "connection," "joining," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components, unless otherwise expressly limited. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0113] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for predicting the main frequency of a battery pack based on a neural network, characterized in that, The method includes: Obtain historical battery pack project data for the vehicle, and construct a primary frequency prediction raw data set based on the historical battery pack project data. The primary frequency prediction raw data set includes dependent variable data and independent variable data. The dependent variable data is the battery pack primary frequency data, and the independent variable data is the macroscopic variable data of the battery pack in the early stage of overall layout design. A target neural network prediction model is constructed based on the original dataset of the main frequency prediction. Obtain multiple design features of the current battery pack, wherein the design features are macroscopic variable data of the current battery pack in the early stage of overall layout design; The design features are input into the target neural network prediction model to predict and output the battery pack frequency of the current battery pack.

2. The battery pack main frequency prediction method based on neural networks according to claim 1, characterized in that, Constructing a target neural network prediction model based on the original dataset for the main frequency prediction includes: Based on the kernel principal component analysis function and preset conditions, feature extraction is performed on the original dataset of the main frequency prediction to obtain the sample dataset; The sample dataset is preprocessed to obtain training data samples; Determine the input and output parameters of the preset neural network architecture and the range of values ​​for the model hyperparameters in order to construct the preset neural network model; The preset neural network model is trained based on the training data samples, and during the training process, the parameter variables corresponding to the range of hyperparameter values ​​of the model are automatically optimized based on the target optimization algorithm to obtain the target neural network prediction model.

3. The battery pack main frequency prediction method based on neural networks according to claim 2, characterized in that, Based on the kernel principal component analysis function and preset conditions, feature extraction is performed on the original dataset of the main frequency prediction to obtain a sample dataset, including: Based on the kernel principal component analysis function, feature extraction is performed on the original dataset of the main frequency prediction to obtain a set of candidate independent variable features; Based on the preset conditions, the candidate independent variable feature set is filtered to obtain the target independent variable feature set; The sample data set is constructed based on the target independent variable feature set and the corresponding dependent variable feature, wherein the dependent variable feature is the battery pack main frequency.

4. The battery pack main frequency prediction method according to claim 3, characterized in that, The preset conditions include that the influence coefficient of the independent variable features in the target independent variable feature set on the battery pack stiffness is greater than a preset coefficient, and that there is no fixed physical mechanism relationship between the multiple independent variable features in the target independent variable feature set.

5. The battery pack main frequency prediction method according to claim 2, characterized in that, The sample dataset is preprocessed to obtain training data samples, including: Based on the sample data set, determine multiple independent variable feature matrices and corresponding target variable features; Each of the independent variable feature matrices is standardized to scale each of the independent variable feature matrices to the target interval; The training data sample is constructed based on the standardized feature matrices of the independent variables and the corresponding feature matrices of the target variables.

6. The battery pack main frequency prediction method according to claim 2, characterized in that, The training data samples include multiple independent variable feature matrices and corresponding target variable features. The input and output parameters of the preset neural network architecture and the range of model hyperparameter values ​​are determined to construct the preset neural network model, including: The number of input layer neurons in the preset neural network architecture is determined based on the number of features in the independent variable feature matrix. The number of output layer neurons in the preset neural network architecture is determined based on the number of features of the target variable. The number of neurons in the first hidden layer of the preset neural network architecture is determined based on the number of neurons in the input layer and the number of neurons in the output layer. The preset neural network model is determined based on the number of neurons in the input layer, the number of neurons in the output layer, the number of neurons in the first hidden layer, and the range of model hyperparameter values. The range of model hyperparameter values ​​includes one or more of the following: the range of the number of hidden layers, the range of the dropout rate, the range of the optimizer selection, the range of the learning rate, and the range of the batch size.

7. The battery pack main frequency prediction method according to claim 2, characterized in that, The target optimization algorithm is a genetic algorithm. During training, the algorithm automatically optimizes the parameter variables corresponding to the range of hyperparameter values ​​of the model to obtain the target neural network prediction model, including: Design a genetic algorithm and construct the fitness function of the genetic algorithm based on mean squared error, coefficient of determination, and mean absolute percentage error; The size of the genetic algorithm and the initial population are determined based on the fitness function. Based on the genetic algorithm, the population is iterated starting from the initial population, and the iteration number and fitness value corresponding to each iteration are obtained: The optimal individual is determined based on the iteration number and fitness value corresponding to each iteration. The target neural network prediction model is constructed based on the hyperparameters of the optimal individual.

8. An electronic device, characterized in that, The method includes a memory, a processor, and a battery pack frequency prediction program stored in the memory and executable on the processor. When the processor executes the battery pack frequency prediction program, it implements the neural network-based battery pack frequency prediction method according to any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, It stores a battery pack main frequency prediction program, which, when executed by the processor, implements the neural network-based battery pack main frequency prediction method according to any one of claims 1-7.

10. A battery pack main frequency prediction device based on neural networks, characterized in that, The device includes: The first acquisition module is used to acquire historical battery pack project data of the vehicle and construct a primary frequency prediction raw data set based on the historical battery pack project data. The primary frequency prediction raw data set includes dependent variable data and independent variable data. The dependent variable data is the battery pack primary frequency data, and the independent variable data is the macroscopic variable data of the battery pack in the early stage of overall layout design. The construction module is used to construct a target neural network prediction model based on the original dataset of the main frequency prediction. The second acquisition module is used to acquire multiple design features of the current battery pack, wherein the design features are macroscopic variable data of the current battery pack in the early stage of overall layout design. The prediction module is used to input multiple design features into the target neural network prediction model to predict and output the battery pack frequency of the current battery pack.

Images