An Electric Vehicle Battery Temperature Global Trajectory Optimization Method, System, Device and Medium

Abstract

The present invention discloses an electric vehicle battery temperature global trajectory optimization method, system, device and medium, including: collecting historical operating condition data and temperature trajectories, constructing an initial operating condition prediction model by using RBF and LSTM networks, and obtaining a trained model through Bayesian optimization training; then, using transfer learning for real-time iterative prediction and optimization to obtain an optimized operating condition prediction model, based on this model, performing real-time operating condition prediction to obtain prediction data; constructing a NARX energy consumption prediction model and training it with the prediction data and temperature control targets to achieve real-time energy consumption prediction. Finally, with the goal of minimizing the weighted value of energy consumption and temperature deviation, a fitness function is constructed, and optimization is performed in the population through a genetic algorithm to obtain the optimal global temperature trajectory. The technical solution of the present invention can use short-distance driving data combined with historical operating condition data to predict the global operating condition in real time and optimize the global thermal management temperature trajectory.

Description

Technical Field

[0001] The present invention belongs to the technical field of electric vehicle thermal management, and particularly relates to a method, system, device and medium for optimizing the global temperature trajectory of an electric vehicle battery. Background Art

[0002] The temperature optimization of an electric vehicle battery is the key to improving battery performance and extending battery life. During the charging and discharging process of the battery, heat energy is generated, resulting in energy consumption loss of the battery and affecting the battery performance. Therefore, studying the mechanism and characteristics of the battery, exploring the relationship between the remaining battery capacity, temperature factors and energy consumption, and establishing a driving range energy consumption model are crucial for optimizing the battery temperature of an electric vehicle. However, there are mainly two drawbacks in the current global optimization methods. One is that it can only be used for global optimization under offline and known global working conditions and cannot be applied online in real time. The other is that the commonly used dynamic programming algorithm (DP) is found in practical applications that, since it solves for all possible decision variables at each moment and cannot constrain and optimize global quantities, the optimization result will deviate from the actual value partially, having certain limitations. Summary of the Invention

[0003] The purpose of the present invention is to provide a method, system, device and medium for optimizing the global temperature trajectory of an electric vehicle battery to solve the problems existing in the above-mentioned prior art.

[0004] To achieve the above purpose, the present invention provides a method for optimizing the global temperature trajectory of an electric vehicle battery, including:

[0005] Obtaining historical working condition data and global temperature trajectory of the electric vehicle;

[0006] Combining a radial basis function neural network and a long short-term memory network to construct an initial working condition prediction model, training and optimizing the initial working condition prediction model based on the historical working condition data by using a Bayesian optimization algorithm to obtain a trained working condition prediction model; performing real-time iterative prediction and optimization on the trained working condition prediction model based on a transfer learning algorithm to obtain an optimized working condition prediction model;

[0007] Performing real-time working condition prediction based on the optimized working condition prediction model to obtain working condition prediction data;

[0008] Constructing an initial heat management system energy consumption prediction model based on a non-linear autoregressive neural network;

[0009] Training the initial heat management system energy consumption prediction model based on the working condition prediction data and a preset temperature control target, and performing real-time energy consumption prediction of the heat management system based on the trained heat management system energy consumption prediction model to obtain heat management system energy consumption prediction data;

[0010] Construct a fitness function with the goal of minimizing the weighted value of the energy consumption prediction data of the thermal management system and the temperature offset value of the neutral temperature; construct a population with the global temperature trajectory as an individual, and based on the fitness values corresponding to each of the individuals, perform iterative optimization within the population in combination with the genetic optimization algorithm, and output the optimal global temperature trajectory after the iteration is completed.

[0011] Optionally, before constructing the initial condition prediction model by combining the radial basis function neural network and the long short-term memory network, it further includes:

[0012] Perform clustering analysis on the historical condition data, and use the clustering analysis result as the model structure parameter;

[0013] Based on the model structure parameter, combine the radial basis function neural network and the long short-term memory network to construct an initial condition prediction model.

[0014] Optionally, training and optimizing the initial condition prediction model based on the historical condition data in combination with the Bayesian optimization algorithm specifically includes:

[0015] Take the minimum validation loss of the initial condition prediction model under the hyperparameters as the objective function;

[0016] Based on the Bayesian optimization algorithm and the objective function, perform hyperparameter optimization on the initial condition prediction model to obtain the best hyperparameters, and train the condition prediction model corresponding to the best hyperparameters based on the historical condition data to obtain the trained condition prediction model.

[0017] Optionally, performing real-time iterative prediction and optimization on the trained condition prediction model based on the transfer learning algorithm specifically includes:

[0018] Freeze the trained condition prediction model based on the preset number of frozen layers, add a new output layer to the condition prediction model after the model freezing is completed, and perform training iteration on the condition prediction model after adding the new output layer based on the real-time condition data, and obtain the optimized condition prediction model after the iteration is completed.

[0019] Optionally, the iterative optimization within the population in combination with the genetic optimization algorithm based on the fitness values corresponding to each of the individuals specifically includes:

[0020] Perform iterative update on each of the individuals based on the genetic optimization algorithm. In each iterative update process, calculate the fitness values corresponding to each of the individuals, and perform iterative update on the population based on the fitness values corresponding to each of the individuals until the iteration times reach the preset value and then output the optimal individual to obtain the optimal global temperature trajectory.

[0021] An electric vehicle battery temperature global trajectory optimization system includes:

[0022] A data acquisition module, configured to obtain historical driving condition data and global temperature trajectory of an electric vehicle;

[0023] A driving condition prediction module, configured to construct an initial driving condition prediction model by combining a radial basis function neural network and a long short-term memory network, and train and optimize the initial driving condition prediction model based on the historical driving condition data by using a Bayesian optimization algorithm to obtain a trained driving condition prediction model; perform real-time iterative prediction and optimization on the trained driving condition prediction model based on a transfer learning algorithm to obtain an optimized driving condition prediction model; perform real-time driving condition prediction based on the optimized driving condition prediction model to obtain driving condition prediction data;

[0024] A thermal management system energy consumption prediction module, configured to construct an initial thermal management system energy consumption prediction model according to a non-linear autoregressive neural network; train the initial thermal management system energy consumption prediction model based on the driving condition prediction data and a preset temperature control target, and perform real-time energy consumption prediction of the thermal management system based on the trained thermal management system energy consumption prediction model to obtain thermal management system energy consumption prediction data;

[0025] A global optimization module, configured to construct a fitness function with the goal of minimizing the weighted value of the thermal management system energy consumption prediction data and the global temperature offset value; construct a population with the global temperature trajectory as an individual, and perform iterative optimization within the population based on the fitness values corresponding to each individual by using a genetic optimization algorithm, and output the optimal global temperature trajectory after the iteration is completed.

[0026] An electronic device, including a memory and a processor, where the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the described method for globally optimizing the temperature trajectory of an electric vehicle battery.

[0027] A computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, the described method for globally optimizing the temperature trajectory of an electric vehicle battery is implemented.

[0028] The technical effects of the present invention are as follows:

[0029] The present invention provides an effective method for global driving condition prediction and global optimization of battery temperature trajectory. By combining clustering analysis, RBF and LSTM neural networks, and transfer learning technology, high-precision, high-real-time and strong-adaptability driving condition prediction is realized. A thermal management system energy consumption model is established by using a NARX neural network, and global optimization of the predicted global driving conditions is realized by using a genetic algorithm. The present invention can use short-distance driving data combined with historical driving condition data to predict global driving conditions and optimize the global thermal management temperature trajectory. Description of the Drawings

[0030] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required in the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0031] The accompanying drawings that form a part of this application are used to provide a further understanding of this application. The schematic embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation to this application. In the accompanying drawings:

[0032] Figure 1 It is the implementation flowchart in the embodiments of the present invention. Specific implementation manners

[0033] Now, various exemplary implementation manners of the present invention will be described in detail. This detailed description should not be considered as a limitation to the present invention, but rather as a more detailed description of certain aspects, characteristics, and implementation schemes of the present invention.

[0034] It should be understood that the terms described in the present invention are only for describing specific implementation manners and are not used to limit the present invention. Additionally, for the numerical ranges in the present invention, it should be understood that each intermediate value between the upper and lower limits of the range is also specifically disclosed. Each intermediate value within any stated value or stated range, as well as each smaller range between any other stated value or intermediate value within the stated range, is also included in the present invention. The upper and lower limits of these smaller ranges can be independently included or excluded from the range.

[0035] Without departing from the scope or spirit of the present invention, various improvements and changes can be made to the specific implementation manners of the description of the present invention, which are obvious to those skilled in the art. Other implementation manners obtained from the description of the present invention are obvious to those skilled in the art. The description and embodiments of this application are only exemplary.

[0036] Regarding the terms "comprising", "including", "having", "containing", etc. used herein, they are all open-ended terms, meaning including but not limited to.

[0037] It should be noted that, without conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The following will refer to the accompanying drawings and combine with the embodiments to detail this application.

[0038] Embodiment 1

[0039] Such as Figure 1As shown in the figure, in this embodiment, a global trajectory optimization method for the battery temperature of an electric vehicle is provided, including: obtaining the historical operating condition data and the global temperature trajectory of the electric vehicle; constructing an initial operating condition prediction model by combining a radial basis function neural network and a long short-term memory network, and training and optimizing the initial operating condition prediction model based on the historical operating condition data by using a Bayesian optimization algorithm to obtain a trained operating condition prediction model; performing real-time iterative prediction and optimization on the trained operating condition prediction model based on a transfer learning algorithm to obtain an optimized operating condition prediction model; performing real-time operating condition prediction based on the optimized operating condition prediction model to obtain operating condition prediction data; constructing an initial energy consumption prediction model for the thermal management system based on a non-linear autoregressive neural network; training the initial energy consumption prediction model for the thermal management system based on the operating condition prediction data and a preset temperature control target, and performing real-time energy consumption prediction on the thermal management system based on the trained energy consumption prediction model for the thermal management system to obtain energy consumption prediction data for the thermal management system; constructing a fitness function with the goal of minimizing the weighted value of the energy consumption prediction data of the thermal management system and the global temperature offset value; constructing a population with the global temperature trajectory as an individual, and performing iterative optimization within the population based on the fitness values corresponding to each individual by using a genetic optimization algorithm, and outputting the optimal global temperature trajectory after the iteration is completed.

[0040] 1. Big data operating condition eigenvalue clustering analysis: Use the k-means algorithm to perform eigenvalue clustering on a large amount of historical operating condition data, and analyze the center and width parameters of the hidden layer neurons, so as to determine the structure of the adaptive neural network model.

[0041] 2. Neural network establishment: Combine a radial basis function (RBF) neural network and a long short-term memory (LSTM) network to establish a hybrid neural network model. This model can learn the time series characteristics in historical data, and adaptively modify the model structure according to the input data for iterative optimization to improve the prediction accuracy.

[0042] 3. Neural network model training and prediction: Use a large amount of historical data to perform hyperparameter optimization and training on the established hybrid neural network model through Bayesian optimization, and obtain the prediction of the global operating condition according to the short-distance operating condition eigenvalue through cross-validation.

[0043] 4. Real-time iterative prediction and model optimization: Through transfer learning technology, update the prediction results in real time, and use the new operating condition data to optimize the neural network to adapt to the new operating condition changes.

[0044] 5. Energy consumption model establishment for the thermal management system: Use the vehicle thermal management data to establish and train a non-linear autoregressive neural network (NARX) with external input, so that it can be used as an energy consumption model that can predict the energy consumption of the thermal management system in real time according to the input vehicle speed and temperature control target.

[0045] 6. Global optimization using genetic algorithm: Establish a fitness function based on the energy consumption model, write genetic algorithm operators, and use the genetic algorithm to optimize the battery temperature control target trajectory under global working conditions.

[0046] The method of this embodiment can provide high-precision global working condition prediction, with good real-time performance and adaptability. Through iterative optimization and model update, the prediction accuracy can be continuously improved. The energy consumption model and genetic algorithm based on neural network modeling can accurately and quickly complete global optimization.

[0047] Specific implementation manner of this embodiment:

[0048] Step 1: Cluster analysis of big data working condition eigenvalue;

[0049] Collect historical working condition data and extract eigenvalues. Use the k-means algorithm to perform cluster analysis on the data to facilitate the construction of the subsequent RBF adaptive neural network model.

[0050] 1. Data reading and preprocessing: First, perform data reading to create a data set D = {(x 1 , y 1 ), (x 2 , y 2 ), …, (x n , y n )}, where x i is the feature vector and y i is the target variable.

[0051] 2. Standardize features: Standardize the eigenvalues to facilitate subsequent classification training and division. The specific method is as follows:

[0052] Let , where μ is the eigenvalue mean and σ is the standard deviation.

[0053] 3. Divide the training set and the test set:

[0054] Set the randomness and divide the data set into a training set and a test set, namely X test , X train , Y test , Y train .

[0055] 4. K-means clustering initialization;

[0056] 1) Use the K-means++ initialization method. First, randomly select a point from the data set as the clustering center c 1 .

[0057] 2) Then, for each data point x, calculate its distance from the clustering center D(x) = min j ||x - c j|| 2 。

[0058] 3) Select new cluster centers according to the ratio of the square of the distance D(x). Specifically: taking the square of the distance of each point as the probability, select the new center:

[0059] 4) Repeat steps 2) and 3) until k cluster centers are selected.

[0060] 5. Cluster analysis;

[0061] 1) Assign each data point x i to the nearest cluster center c j ;

[0062] 2) Update the cluster center c j , calculate the average value of all points in each cluster: where, N j is the number of samples assigned to cluster j;

[0063] 3) Repeat the above 1) and 2) until the cluster centers no longer change or reach the maximum number of iterations.

[0064] Step 2: Combine the clustering analysis method in Step 1 to create an RBF class, realize the adaptive adjustment of the neural network structure, automatically adjust its parameters (such as the center position, width, and weight) according to the changes in the input data, so as to better approximate the real data. And combine the RBF and LSTM models through keras, and use the Adam optimizer for gradient descent of the model.

[0065] First, create an RBF class to realize the adaptive adjustment of the neural network structure:

[0066] 1. Define the basic parameters of the RBF layer;

[0067] Output dimension: d;

[0068] Center: Set as c i represents the center of each output, i = 1, 2,..., d;

[0069] Beta value: Set as β, controlling the width of the function, i = 1, 2,..., d.

[0070] 2. Initialize;

[0071] In the initialization stage, the center and beta value are initialized:

[0072] Initialization of the center: c i ~ U(0, 1), for i = 1, 2,..., d Here U(0, 1) means randomly generated from the uniform distribution;

[0073] Initialization of Beta value: β i = 1.0 for i = 1, 2, …, d.

[0074] 3. Forward propagation calculation;

[0075] Given the input x ∈ R n (where n is the dimension of the input), the output layer calculates the output value through the following steps:

[0076] 1) Expand the center: Each center is expanded in the new dimension;

[0077] 2) Calculate the distance: Calculate the distance between the input sample and each center: H = C - x;

[0078] 3) Calculate the output: The calculation formula for the output is: Here, ||c i - x|| 2 represents the square of the Euclidean distance between c i and the input x.

[0079] After that, combine the RBF and LSTM models:

[0080] 1. Preprocess the input;

[0081] Given the input time series data: x = {x 1 , x 2 , … x T}.

[0082] 2. Create an LSTM layer;

[0083] The input of the LSTM is the output of the RBF, and the sequence representation is obtained:

[0084] Assume the RBF output is y RBF , and the output after input to the LSTM layer is h(t) = LSTM(y RBF (t), h(t - 1), c(t - 1)), where h(t) is the hidden state of the LSTM at time step t, and c(t - 1) is the memory cell state of the LSTM.

[0085] The core formula of the LSTM is:

[0086] 3. Output layer;

[0087] The output layer is usually a fully connected layer, and the output prediction y pred = W out ·h(T) + b out Here, W outis the output layer weight, b out is the bias.

[0088] 4. Loss function;

[0089] Use the mean squared error (MSE) as the loss function:

[0090] 5. Adam optimizer;

[0091] Use the Adam optimizer to update the network weight parameter θ, and the steps are as follows:

[0092] 1) Calculate the gradient:

[0093] 2) Update the first-order momentum and second-order momentum:

[0094] 3) Correct the bias:

[0095] 4) Update the parameter:

[0096] Step 3: Use a large amount of historical data to optimize and train the hyperparameters of the established hybrid neural network model through Bayesian optimization, and obtain the prediction of the global working condition based on the short-distance working condition characteristic values through cross-validation.

[0097] 1. Define hyperparameters: Let the model be f(x; h), where x is the input and h is the hyperparameter, and optimize the performance of the input model by adjusting the hyperparameter h.

[0098] 2. Objective function: The input in hyperparameter optimization is the hybrid neural network model obtained in Step 2. Define the objective function L(h) as the validation loss of the model f(x; h) under the hyperparameter h: L(h) = E[L val (h)] where L val (h) is the validation loss obtained through cross-validation when the hyperparameter h is given.

[0099] 3. Bayesian optimization: Bayesian optimization uses a surrogate model to model the objective function and selects hyperparameters based on this model.

[0100] 1) Surrogate model: Use the Gaussian process GP to model the objective function L(h) ∼ GP(m(h), k(h, h')) where m(h) is the mean function and k(h, h') is the covariance function.

[0101] 2) Selection criterion: In Bayesian optimization, the selection criterion is the expected improvement (EI): where L *is the currently known best validation loss, and p(z|h) is the distribution predicted by the Gaussian process.

[0102] 4. Hyperparameter update: Find the next hyperparameter h by maximizing the selection criterion EI(h) next = argmax h EI(h).

[0103] 5. Update and iteration: After selecting the new hyperparameter, evaluate the model performance and update the surrogate model. Iterate this process and evaluate the update until the performance metric no longer improves or reaches the maximum number of iterations.

[0104] 6. Obtain the hybrid neural network model with the best hyperparameters according to the results of Bayesian optimization, and train the best model.

[0105] 7. Use the test set for testing, complete cross-validation, and obtain the hybrid neural network model that can predict the global working condition through short-term eigenvalue.

[0106] Step 4: Utilize transfer learning, perform real-time iterative prediction, update the prediction results, and simultaneously optimize the neural network using new working condition data. Collect new working condition data in real time. Utilize transfer learning technology to integrate new data into the existing model for real-time prediction. Adjust and optimize the neural network according to the prediction results and actual working conditions.

[0107] 1. Freeze the model layers, freeze all layers except the last few layers for model update, which is essentially training similar models with the same logic but different details:

[0108] For each layer i = (1, 2,..., N - 1), set trainable = 0, where N is the total number of model layers, and the last layer l N is the trainable layer.

[0109] 2. Add a new output layer to adapt to and learn new data;

[0110] y = f(x; W, b) = W · x + b where y is the output, W is the new weight parameter, b is the bias term, and x is the input data.

[0111] 3. Model compilation, which requires a smaller learning rate to quickly compile similar models:

[0112] Loss = L(y ture , y pred ) = huber_loss(y ture , f(x; W, b)), optimizer = Adam(α = 1 × 10 -5 )

[0113] 4. Model training, continue to train the model using the new dataset:

[0114]

[0115] 5. Model prediction, use the test set for prediction to test the model, and use R 2 for evaluation:

[0116] where, is the mean value of y true .

[0117] Step 5: Use the vehicle thermal management data of historical working conditions to establish and train a non - linear autoregressive neural network model with external input (NARX), so that it can predict the energy consumption of the global thermal management system in real - time through the global working conditions predicted in Steps 3 and 4 and the given temperature control target.

[0118] 1. Adjust the vehicle thermal management data of historical working conditions to conform to the form of the NARX model:

[0119] The input data X 1 is the global temperature trajectory controlled by the thermal management system, and the input data X 2 is the global working condition, i.e., the vehicle speed. The output y is the energy consumption of the thermal management system under the corresponding temperature control trajectory. Concatenate X 1 and X 2 in the second dimension (time T) to form the new input data X.

[0120] 2. Establish the NARX model:

[0121] Define the model f(X; W, b), where W is the weight of the model and b is the bias;

[0122] Define the hidden layer h = ReLU(XW 1 +b 1 ), define the output layer y pred = hW 2 +b 2 , where W 1 and W 2 are the weight matrices of the corresponding layers, and b 1 and b 2 are the biases.

[0123] 3. Compile the model: The loss function is the mean square error:

[0124] 4. Train the model, train the model, and update the weights W 1 , W 2 and the biases b 1 , b 2:

[0125] W, b ← train_model(X, y) performs 100 rounds of training with a batch size of B = 10.

[0126] Step 6: Based on the energy consumption model established in Step 5, construct a fitness function, specifically the weighted value of the global energy consumption and the offset value of the average temperature in a draw. Write a genetic algorithm to optimize the minimum value of the fitness function. Due to the input variable requirements of the genetic algorithm, the generated initial temperature trajectory and the obtained optimal temperature rule are both a number of scatter points, which will be used as the global phased temperature control target of the thermal management system.

[0127] 1. Determine the genetic individuals: Let P be the population, containing N individuals, and each individual can be represented as a chromosome x i : P = {x 1 , x 2 , …, x N}. The information carried by each individual is the row matrix of global temperature trajectory scatter points, that is, x i = T i ,

[0128] 2. Construct a fitness function. To ensure low thermal management energy consumption while enabling the battery to work more generally at a more suitable temperature, the fitness function should limit both the thermal management energy consumption and the average battery temperature, specifically:

[0129] Among them, is the global average temperature, y i is the output of the energy consumption model established in Step 5, and W 1 and W 2 are the corresponding weights.

[0130] 3. Generate a new population through selection, crossover, and mutation:

[0131] 1) Selection operation:

[0132] The selection operation uses roulette wheel selection, and the probability of selecting x is proportional to its fitness. i

[0133] 2) Crossover operation: For the two selected parent individuals x a and x b , perform a single-point crossover operation to generate two new individuals xc hild1 and x child2 , x child1 = [t 1 , t 2 , …, t c , t' c+1 , …, t' m, x child2 = [t' 1 , t' 2 , …, t' c , t c+1 , …, t m , where c is the selected intersection point.

[0134] 3) Mutation operation: Non-uniform mutation is adopted. A variation amount ∈ and a uniformly distributed random number u are randomly generated, and a mutation probability p is set. Then the mutated individual

[0135] 4. Genetic termination: When the maximum number of iterations is reached or the fitness change is less than the threshold, genetic termination occurs. At this time, the individual with the highest fitness in the population is the best individual, and the global temperature trajectory formed by the global temperature trajectory scatter point row matrix carried by it is the global optimal temperature trajectory under the current working condition.

[0136] This embodiment provides an effective global working condition prediction and global optimization method for battery temperature trajectory. By combining k-means clustering analysis, RBF and LSTM neural networks, and transfer learning technology, high-precision, high-real-time and strong-adaptability working condition prediction is achieved. A power consumption model of the thermal management system is established through the NARX neural network, and the global optimization of the predicted global working conditions is realized by using the genetic algorithm. Overall, it can complete the prediction of global working conditions and the optimization of the global thermal management temperature trajectory by combining short-distance driving data with historical working condition data.

[0137] An electric vehicle battery temperature global trajectory optimization system, comprising:

[0138] A data acquisition module, used to obtain the historical working condition data and global temperature trajectory of the electric vehicle;

[0139] A working condition prediction module, used to construct an initial working condition prediction model by combining a radial basis function neural network and a long short-term memory network, train and optimize the initial working condition prediction model based on the historical working condition data combined with the Bayesian optimization algorithm to obtain a trained working condition prediction model; perform real-time iterative prediction and optimization on the trained working condition prediction model based on the transfer learning algorithm to obtain an optimized working condition prediction model; perform real-time working condition prediction based on the optimized working condition prediction model to obtain working condition prediction data;

[0140] A thermal management system power consumption prediction module, used to construct an initial thermal management system power consumption prediction model according to a non-linear autoregressive neural network; train the initial thermal management system power consumption prediction model based on the working condition prediction data and a preset temperature control target, and perform real-time power consumption prediction of the thermal management system based on the trained thermal management system power consumption prediction model to obtain thermal management system power consumption prediction data;

[0141] A global optimization module is configured to construct a fitness function with the goal of minimizing the weighted value of the energy consumption prediction data of the thermal management system and the flat temperature offset value; construct a population with the global temperature trajectory as an individual, and perform iterative optimization within the population based on the fitness values corresponding to each of the individuals in combination with a genetic optimization algorithm, and output the optimal global temperature trajectory after the iteration is completed.

[0142] An electronic device includes a memory and a processor. The memory is used to store a computer program, and the processor runs the computer program to cause the electronic device to execute the method for optimizing the global trajectory of an electric vehicle battery temperature described above. A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for optimizing the global trajectory of an electric vehicle battery temperature described above is implemented.

[0143] As described above, the above is only a preferred specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for optimizing the global trajectory of electric vehicle battery temperature, characterized in that: include: Obtain historical operating data and global temperature trajectory of electric vehicles; Performing cluster analysis on the historical operating condition data, and using the cluster analysis results as model structure parameters; Based on the model structure parameters, an initial operating condition prediction model is constructed by combining a radial basis function neural network and a long short-term memory network, and the objective function is to minimize the verification loss of the initial operating condition prediction model under hyperparameters; based on the Bayesian optimization algorithm and the objective function, the hyperparameters of the initial operating condition prediction model are optimized to obtain the optimal hyperparameters, and the operating condition prediction model corresponding to the optimal hyperparameters is trained based on the historical operating condition data to obtain the trained operating condition prediction model; based on the transfer learning algorithm, the trained operating condition prediction model is iteratively predicted and optimized in real time to obtain the optimized operating condition prediction model; Execute real-time working condition prediction based on the optimized working condition prediction model to obtain working condition prediction data; Construct an initial thermal management system energy consumption prediction model based on nonlinear autoregressive neural network; The initial thermal management system energy consumption prediction model is trained based on the operating condition prediction data and the preset temperature control target, and the real-time energy consumption prediction of the thermal management system is performed based on the trained thermal management system energy consumption prediction model to obtain the thermal management system energy consumption prediction data; The fitness function is constructed with the goal of minimizing the weighted value of the thermal management system energy consumption prediction data and the average temperature offset value; A population is constructed with the global temperature trajectory as an individual, and an iterative search is performed within the population based on the fitness value corresponding to each individual combined with a genetic optimization algorithm, and the optimal global temperature trajectory is output after the iteration is completed.

2. The method for optimizing the global trajectory of battery temperature of an electric vehicle according to claim 1, characterized in that: The real-time iterative prediction and optimization of the trained working condition prediction model based on the transfer learning algorithm specifically includes: The trained working condition prediction model is frozen based on the preset number of freezing layers, and a new output layer is added to the working condition prediction model after the model freezing is completed. The working condition prediction model with the new output layer added is trained and iterated based on real-time working condition data, and an optimized working condition prediction model is obtained after the iteration is completed.

3. The method for optimizing the global trajectory of battery temperature of an electric vehicle according to claim 1, characterized in that: The iterative optimization in the population based on the fitness value corresponding to each individual combined with a genetic optimization algorithm specifically includes: Based on the genetic optimization algorithm, each individual is iteratively updated. In each iterative update process, the fitness value corresponding to each individual is calculated, and the population is iteratively updated based on the fitness value corresponding to each individual until the number of iterations reaches a preset value, and the optimal individual is output to obtain the optimal global temperature trajectory.

4. A global trajectory optimization system for battery temperature of an electric vehicle, used to implement a global trajectory optimization method for battery temperature of an electric vehicle according to any one of claims 1 to 3, characterized in that: include: Data acquisition module, used to obtain historical operating data and global temperature trajectory of electric vehicles; The working condition prediction module is used to construct an initial working condition prediction model by combining a radial basis function neural network and a long short-term memory network, train and optimize the initial working condition prediction model based on the historical working condition data in combination with a Bayesian optimization algorithm to obtain a trained working condition prediction model; perform real-time iterative prediction and optimization on the trained working condition prediction model based on a transfer learning algorithm to obtain an optimized working condition prediction model; perform real-time working condition prediction based on the optimized working condition prediction model to obtain working condition prediction data; The thermal management system energy consumption prediction module is used to construct an initial thermal management system energy consumption prediction model based on a nonlinear autoregressive neural network; the initial thermal management system energy consumption prediction model is trained based on the operating condition prediction data and the preset temperature control target, and the real-time energy consumption prediction of the thermal management system is performed based on the trained thermal management system energy consumption prediction model to obtain the thermal management system energy consumption prediction data; A global optimization module is used to construct a fitness function with the goal of minimizing the weighted value of the thermal management system energy consumption prediction data and the average temperature offset value; A population is constructed with the global temperature trajectory as an individual, and an iterative search is performed within the population based on the fitness value corresponding to each individual combined with a genetic optimization algorithm, and the optimal global temperature trajectory is output after the iteration is completed.

5. An electronic device, characterized in that: It comprises a memory and a processor, the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute a method for optimizing the global trajectory of the battery temperature of an electric vehicle according to any one of claims 1 to 3.

6. A computer-readable storage medium, characterized in that: It stores a computer program, and when the computer program is executed by a processor, it implements a method for optimizing the global trajectory of the battery temperature of an electric vehicle as described in any one of claims 1-3.

Images