Wireless charging control method and system based on vehicle controller

Abstract

The present application provides a wireless charging control method and system based on a vehicle controller, the method comprising: preprocessing initial data generated by vehicle operation to obtain a structured state information matrix; determining a state parameter feature vector according to a preset vector change method and the structured state information matrix; performing correlation analysis on the state parameter feature vector to obtain a correlation strength matrix and an importance weight vector; performing feature fusion on the state parameter feature vector according to the correlation strength matrix and the importance weight vector to obtain a comprehensive state representation vector; performing deep neural network model training on the comprehensive state representation vector to obtain a charging strategy optimization model; performing charging prediction according to the charging strategy optimization model to obtain a charging optimization plan, and the charging plan is used to prompt the vehicle to charge.

Description

Technical Field

[0001] The present application relates to the field of wireless charging technology, and in particular to a wireless charging control method and system based on a vehicle controller. Background Art

[0002] Traditional electric vehicles can usually only display the current remaining battery percentage, lacking effective predictions of power consumption in future trips and personalized charging guidance for specific driving conditions. This limitation may cause drivers to run out of power during their trips, which in turn affects travel plans or causes inconvenience. In addition, simple power display cannot meet users' needs for precise power management because it fails to fully consider factors such as actual road conditions and driving habits. Summary of the invention

[0003] The present application provides a wireless charging control method and system based on a vehicle controller, which are used to accurately manage the power of a vehicle.

[0004] In a first aspect, an embodiment of the present application provides a wireless charging control method based on a vehicle controller, the method comprising:

[0005] Preprocess the initial data generated by vehicle operation to obtain a structured state information matrix;

[0006] Determining a state parameter characteristic vector according to a preset vector change method and the structured state information matrix;

[0007] Performing correlation analysis on the state parameter characteristic vector to obtain a correlation strength matrix and an importance weight vector;

[0008] Performing feature fusion on the state parameter feature vector according to the association strength matrix and the importance weight vector to obtain a comprehensive state representation vector;

[0009] Performing deep neural network model training on the comprehensive state representation vector to obtain a charging strategy optimization model;

[0010] A charging prediction is performed according to the charging strategy optimization model to obtain a charging optimization plan, and the charging plan is used to prompt the vehicle to charge.

[0011] In a second aspect, an embodiment of the present application provides a wireless charging control system based on a vehicle controller, the device comprising:

[0012] A data processing module is used to pre-process the initial data generated by the vehicle operation to obtain a structured state information matrix;

[0013] A vector determination module, used to determine a state parameter characteristic vector according to a preset vector change method and the structured state information matrix;

[0014] A correlation analysis module, used to perform correlation analysis on the state parameter feature vector to obtain a correlation strength matrix and an importance weight vector;

[0015] A feature fusion module, used for performing feature fusion on the state parameter feature vector according to the association strength matrix and the importance weight vector to obtain a comprehensive state representation vector;

[0016] A model training module, used to perform deep neural network model training on the comprehensive state representation vector to obtain a charging strategy optimization model;

[0017] The result output module is used to perform charging prediction according to the charging strategy optimization model to obtain a charging optimization plan, and the charging plan is used to prompt the vehicle to charge.

[0018] In a third aspect, an embodiment of the present application provides an electronic device, the electronic device comprising a memory and a processor;

[0019] The memory is used to store computer programs;

[0020] The processor is used to execute the computer program and implement the wireless charging control method based on the vehicle controller as described in any one of the embodiments of the present application when executing the computer program.

[0021] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the processor implements a wireless charging control method based on a vehicle controller as described in any one of the embodiments of the present application.

[0022] The embodiment of the present application provides a wireless charging control method based on a vehicle controller, which is used for a vehicle controller. The method includes: preprocessing the initial data generated by the operation of the vehicle to obtain a structured state information matrix; determining the state parameter feature vector according to a preset vector change method and the structured state information matrix; performing correlation analysis on the state parameter feature vector to obtain a correlation strength matrix and an importance weight vector; performing feature fusion on the state parameter feature vector according to the correlation strength matrix and the importance weight vector to obtain a comprehensive state representation vector; performing deep neural network model training on the comprehensive state representation vector to obtain a charging strategy optimization model; performing charging prediction according to the charging strategy optimization model to obtain a charging optimization plan, and the charging plan is used to prompt the vehicle to charge. Through the above method, the initial data generated by the vehicle operation is preprocessed and a structured state information matrix is ​​constructed. The state parameter feature vector is extracted based on the preset vector change method, which reduces the redundancy of the data. The state parameter feature vector is subjected to correlation analysis, and the obtained correlation intensity matrix and importance weight vector can accurately reflect the mutual influence relationship between the parameters. The comprehensive state representation vector obtained by feature fusion is used to train the deep neural network, which makes full use of the advantages of deep learning in feature extraction and pattern recognition, so that the charging strategy optimization model has strong generalization ability and adaptability. Charging prediction is performed based on the model and an optimization plan is generated. The charging strategy can be dynamically adjusted according to the real-time status of the vehicle, realizing accurate power management of the vehicle, which has strong practical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying any creative work.

[0024] Figure 1 A schematic flow chart of a wireless charging control method based on a vehicle controller provided in an embodiment of the present application;

[0025] Figure 2 A schematic block diagram of a wireless charging control system based on a vehicle controller provided in an embodiment of the present application. DETAILED DESCRIPTION

[0026] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0028] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0029] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0030] See also Figure 1 , Figure 1 FIG. 1 is a schematic flow chart of a wireless charging control method based on a vehicle controller provided in an embodiment of the present application. Figure 1 As shown, the specific steps of the wireless charging control method based on the vehicle controller include: S101-S106.

[0031] S101. Preprocessing initial data generated by vehicle operation to obtain a structured state information matrix.

[0032] For example, a large amount of data is generated during the driving of electric vehicles, such as speed, battery power, temperature, etc. These raw data often contain noise or incomplete information, and direct use will affect the accuracy of subsequent analysis. Therefore, these data need to be preprocessed, including denoising and sorting. Arrange all the related data after denoising in a certain format to form a table or matrix form, so that you can easily see the relationship between different parameters. This sorted data set is called a "structured state information matrix", which is like establishing a detailed "health file" for each vehicle, providing a solid foundation for further in-depth analysis.

[0033] In some embodiments, preprocessing includes: performing outlier cleaning and noise removal on the initial data to obtain cleaned data; performing data standardization on the cleaned data to obtain standardized data; performing time alignment on the standardized data to obtain aligned data; performing matrix reconstruction on the aligned data according to the time series to obtain a structured state information matrix.

[0034] S102: Determine a state parameter characteristic vector according to a preset vector change method and a structured state information matrix.

[0035] Exemplarily, an improved Skip-gram algorithm is used to vectorize the preprocessed state information. First, the context window size of the state parameter is defined as N (where N is determined based on experimental optimization). For the state parameter at each time point, the state parameters at the N time points before and after are used as its context. Then, a state parameter co-occurrence matrix is ​​constructed to record the pairs of state parameters that co-occur in the context window. Based on the co-occurrence matrix, the Skip-gram algorithm model is trained to map each state parameter to a D-dimensional vector space (D is the vector dimension, determined by experimental optimization). During the training process, constraints based on physical laws are introduced to ensure that state parameters with strong correlation (such as battery temperature and charging efficiency) are close in the vector space. Each state parameter is represented as a D-dimensional feature vector, which contains the intrinsic correlation between the state parameters.

[0036] S103, performing correlation analysis on the state parameter characteristic vector to obtain a correlation strength matrix and an importance weight vector.

[0037] Exemplarily, based on the state parameter feature vector, the correlation strength between different state parameters is calculated. The improved cosine similarity algorithm is used to calculate the similarity between vectors, and the similarity threshold is set in combination with the actual physical meaning to screen out strongly correlated state parameter pairs. At the same time, a charging time prediction model based on XGBoost is constructed, and the SHAP (SHapley Additive ex Planations) value is used to evaluate the importance contribution of each state parameter to the charging time prediction. Specifically, for each state parameter, its SHAP value in different samples is calculated to obtain the average contribution of the parameter to the model prediction results. The SHAP value is normalized to obtain the correlation strength matrix and importance weight vector of the state parameter.

[0038] S104. Perform feature fusion on the state parameter feature vector according to the correlation strength matrix and the importance weight vector to obtain a comprehensive state representation vector.

[0039] Exemplarily, based on the importance weight vector, the state parameter feature vectors are weighted and fused. Each state parameter feature vector is multiplied by its corresponding importance weight to obtain a weighted feature vector. Considering the correlation between the state parameters, a feature fusion network based on the attention mechanism is designed, which can adaptively adjust the interaction strength between different feature vectors. Specifically, for each pair of related state parameters, the interaction weight is designed according to their correlation strength to achieve effective fusion of feature information. All weighted and interacted feature vectors are aggregated to generate a comprehensive state representation vector at the current moment.

[0040] S105. Perform deep neural network model training on the comprehensive state representation vector to obtain a charging strategy optimization model.

[0041] Exemplarily, based on the comprehensive state representation vector, a deep neural network model is constructed to optimize the charging strategy. The model adopts a multi-layer perceptron structure. The input layer receives the vehicle state representation vector, the hidden layer uses the LeakyReLU activation function for nonlinear transformation, and the output layer predicts the optimal charging power and the estimated charging completion time. During the model training process, a multi-objective loss function is designed, considering the two goals of charging time optimization and battery life protection. The loss function includes a charging time prediction error term, a battery temperature control term, and a charging efficiency term. The Adam optimizer is used to optimize the model parameters, and an early stopping strategy is introduced to prevent overfitting. In addition, by adjusting the weight coefficient of each loss term, a balance between charging efficiency and battery protection is achieved.

[0042] S106: Perform charging prediction according to the charging strategy optimization model to obtain a charging optimization plan, which is used to prompt the vehicle to charge.

[0043] For example, after the charging strategy optimization model is trained, real-time power prediction results are output during vehicle driving, and charging reminders and power management strategies are given in a timely manner according to different driving conditions.

[0044] The embodiment of the present application provides a wireless charging control method based on a vehicle controller, which is used for a vehicle controller. The method includes: preprocessing the initial data generated by the operation of the vehicle to obtain a structured state information matrix; determining the state parameter feature vector according to a preset vector change method and the structured state information matrix; performing correlation analysis on the state parameter feature vector to obtain a correlation strength matrix and an importance weight vector; performing feature fusion on the state parameter feature vector according to the correlation strength matrix and the importance weight vector to obtain a comprehensive state representation vector; performing deep neural network model training on the comprehensive state representation vector to obtain a charging strategy optimization model; performing charging prediction according to the charging strategy optimization model to obtain a charging optimization plan, and the charging plan is used to prompt the vehicle to charge. Through the above method, the initial data generated by the vehicle operation is preprocessed and a structured state information matrix is ​​constructed. The state parameter feature vector is extracted based on the preset vector change method, which reduces the redundancy of the data. The state parameter feature vector is subjected to correlation analysis, and the obtained correlation intensity matrix and importance weight vector can accurately reflect the mutual influence relationship between the parameters. The comprehensive state representation vector obtained by feature fusion is used to train the deep neural network, which makes full use of the advantages of deep learning in feature extraction and pattern recognition, so that the charging strategy optimization model has strong generalization ability and adaptability. Charging prediction is performed based on the model and an optimization plan is generated. The charging strategy can be dynamically adjusted according to the real-time status of the vehicle, realizing accurate power management of the vehicle, which has strong practical value.

[0045] In order to more clearly introduce the technical solution of the present application, the technical solution of the present application will be introduced through specific embodiments below. It should be noted that the specific embodiments are used to expand the technical solution of the present application, but are not intended to limit the present application.

[0046] In some embodiments, determining the state parameter characteristic vector according to a preset vector change method and a structured state information matrix includes: S1021-S1026.

[0047] S1021. Divide the state parameters in the structured state information matrix into context windows to obtain a context data set of the state parameters, wherein the size of the context window is a preset value N.

[0048] For example, assuming that the preset window size N=5, for a state parameter at a certain time t (such as battery temperature), its context window contains all state parameters from t-5 to t+5. For example, at time t, the battery temperature is 45°C, and its context window contains the battery temperature, power value, power consumption rate, battery voltage, ambient temperature and other parameter values ​​at these 10 time points. Such a sliding window will move over the entire time series, collect all possible context combinations, and form a context data set. This method can capture the changing patterns of state parameters over time and the relationship between parameters.

[0049] S1022. Construct a co-occurrence matrix according to the context data set, wherein each element in the co-occurrence matrix represents the number of times corresponding two state parameters co-occur in the same context window.

[0050] For example, if the battery temperature and the power consumption rate appear 100 times in the same window, then the corresponding element value in the co-occurrence matrix is ​​100. Specifically, assuming there are n state parameters, an n×n co-occurrence matrix will be formed. Each element (i, j) in the matrix represents the number of times parameter i and parameter j appear together in all windows. This matrix is ​​symmetric because the number of times parameters i and j appear together is the same as the number of times parameters j and i appear together.

[0051] Constructing eigenvectors of the co-occurrence matrix: This step converts the co-occurrence relationship into a vector representation. By decomposing the co-occurrence matrix, the eigenvector of each parameter can be obtained. These eigenvectors reflect the distribution relationship of the parameters in the multidimensional space. For example, the temperature parameter may be represented as a multidimensional vector [0.8, 0.5, 0.3], and each dimension of this vector represents the strength of association with other parameters such as pressure, flow, humidity, etc.

[0052] S1023. Construct eigenvectors of the co-occurrence matrix to obtain initial eigenvectors.

[0053] Exemplarily, a matrix decomposition method such as singular value decomposition (SVD) is used to represent each state parameter as a high-dimensional vector. For example, the battery temperature can be represented as a 100-dimensional vector, which contains the association information between the parameter and other parameters. The dimension of the initial feature vector is usually high to retain as much information as possible.

[0054] S1024. Constrain the initial eigenvector according to preset physical law constraints to obtain a constrained eigenvector.

[0055] For example, when the battery temperature rises, the power consumption rate will increase. In the feature vector space, the vectors of these two parameters should be positively correlated. By adding such constraints, the initial feature vector can be corrected to make it more consistent with the actual law. This process can be achieved through optimization algorithms, while maintaining the original mathematical properties and satisfying physical constraints.

[0056] The constraint relationship between battery temperature and power value: When the battery temperature increases, the available power will decrease accordingly, and the two are negatively correlated. Add a distance constraint in the vector space: d(v temperature, v power) ≤ θ1, where θ1 is the empirical threshold, and introduce a loss function term: L temperature-power = max(0, d(v temperature, v power)-θ1).

[0057] The constraint relationship between battery temperature and ambient temperature: The increase in ambient temperature will lead to an increase in battery temperature, and the two are positively correlated. Add a vector angle constraint: cos(vambient temperature, vbattery temperature) ≥ α1, where α1 is the minimum angle threshold. Introduce the loss function term: Lenvironment-battery = max(0, α1-cos(vambient temperature, vbattery temperature)).

[0058] The constraint relationship between the power value and the power consumption speed: the faster the power consumption speed, the faster the remaining power decreases, showing a negative correlation. Add a vector direction constraint: v power · v consumption speed ≤ - β1, where β1 is the negative correlation coefficient threshold. Introduce the loss function term: L power - consumption = max (0, v power · v consumption speed + β1).

[0059] The constraint relationship between battery voltage and charge value: The battery voltage and charge value are positively correlated. The higher the charge, the higher the voltage. Add a vector similarity constraint: sim(v voltage, v charge) ≥ γ1, where γ1 is the minimum similarity threshold. Introduce a loss function term: L voltage - charge = max(0, γ1-sim(v voltage, v charge)).

[0060] The constraint relationship between battery temperature and power consumption rate: Too high temperature will lead to faster power consumption, showing a positive correlation. Add a vector distance constraint: d(v temperature, v consumption rate) ≤ θ2, where θ2 is the empirical threshold, and introduce a loss function term: L temperature-consumption = max(0, d(v temperature, v consumption rate)-θ2).

[0061] S1025. Reduce the dimension of the constrained feature vector to obtain a feature vector after dimension reduction.

[0062] The dimensionality reduction process usually uses algorithms such as t-SNE or PCA to reduce high-dimensional feature vectors to lower dimensions (such as from 100 dimensions to 20 dimensions). This not only reduces the amount of calculation, but also removes noise and retains the most important feature information. For example, a 100-dimensional vector representing battery temperature can be reduced to 20 dimensions while maintaining its main correlation with other parameters.

[0063] S1026. Perform cluster analysis based on the feature vector after dimensionality reduction to obtain a state parameter feature vector.

[0064] For example, cluster analysis usually uses algorithms such as K-means to group similar state parameters. For example, it may be found that temperature-related parameters such as battery temperature, ambient temperature, and radiator temperature are clustered into the same group, while electrical parameters such as voltage and current are clustered into another group. This clustering result can help understand the intrinsic relationship between parameters and obtain the state parameter feature vector to provide a basis for subsequent charging control strategies.

[0065] In some embodiments, constructing a feature vector of the co-occurrence matrix to obtain an initial feature vector includes: S231-S237.

[0066] S231. Divide each element in the co-occurrence matrix by the sum of the elements in the row to obtain a normalized co-occurrence matrix.

[0067] Exemplarily, the co-occurrence matrix normalization normalizes each row of the co-occurrence matrix so that the sum of the elements of each row is 1. For example, assuming there is a 3×3 co-occurrence matrix, where the first row is [4, 2, 2], it becomes [0.5, 0.25, 0.25] after normalization. The normalized matrix can reflect the relative correlation strength between state parameters rather than the absolute number of occurrences.

[0068] S232. Normalize the co-occurrence matrix and construct an input layer node matrix according to the product of the number of elements of the normalized co-occurrence matrix and the preset vector dimension.

[0069] For example, the size of the co-occurrence matrix is ​​M×M (M is the number of state parameters), the preset vector dimension is D, and the input layer node matrix dimension is M×(M×D). The specific construction method is to expand each element of the normalized co-occurrence matrix into a D-dimensional vector to form a larger matrix.

[0070] S233. Construct an output layer target matrix according to the dimension of the co-occurrence matrix.

[0071] Exemplarily, the output layer target matrix is ​​a supervisory signal in the neural network training process, which defines the ideal output learned by the model. The output layer target matrix can directly use the normalized co-occurrence matrix, or it can be appropriately transformed according to specific needs. For example, the co-occurrence values ​​can be logarithmically transformed to reduce numerical differences, or some weight factors can be introduced to emphasize certain specific co-occurrence relationships. Each element of the output layer target matrix represents the expected correlation strength between two state parameters. This correlation strength can be a direct co-occurrence probability or a value transformed by a specific function.

[0072] S234, multiply the input layer node matrix by the preset weight matrix, and calculate the cross entropy loss with the output layer target matrix, and obtain the node weight matrix through stochastic gradient descent iteration.

[0073] Exemplarily, the input layer node matrix is ​​multiplied by the preset weight matrix to obtain the predicted output. The initial value of the weight matrix is ​​usually randomly initialized, such as using a uniform distribution or a normal distribution to generate small random numbers. Then, the predicted output is compared with the target matrix to calculate the cross entropy loss. The cross entropy loss function can effectively measure the difference between the predicted distribution and the target distribution. The stochastic gradient descent algorithm is used for optimization to minimize the loss function by continuously adjusting the value of the weight matrix. In each iteration, the gradient of the loss function to the weight is calculated, and the weight is updated in the opposite direction of the gradient. This process continues until the preset number of iterations is reached or the loss value converges below a certain threshold.

[0074] S235. Multiply the input layer node matrix by the node weight matrix to obtain a feature mapping matrix.

[0075] S236. Decompose the eigenmapping matrix into the product of a left singular matrix, a singular value diagonal matrix, and a right singular matrix to obtain an eigendecomposition matrix.

[0076] For example, the feature map matrix contains the representation of the state parameters in the learned feature space. The feature map matrix is ​​decomposed into the product of three matrices using singular value decomposition (SVD): the left singular matrix U, the singular value diagonal matrix Σ, and the right singular matrix V T . The elements in the singular value diagonal matrix Σ are arranged from large to small, representing the importance of different feature directions. This decomposition can capture the main change patterns of the data and map complex high-dimensional data to a more explanatory space. The column vectors of the left singular matrix U form a set of orthogonal bases, each of which represents a main change direction of the data.

[0077] S237. According to the preset vector dimension, the left singular matrix in the eigendecomposition matrix is ​​used to construct the initial eigenvector of the state parameter.

[0078] Exemplarily, the first D columns are selected from the left singular matrix U as the initial eigenvectors of the state parameters. These D columns correspond to the largest D singular values ​​and contain the most important change information in the data. For example, if the preset dimension D=5, each state parameter will get a 5-dimensional eigenvector. These eigenvectors are a low-dimensional approximation of the original high-dimensional space, and they retain the most important correlations between state parameters. Each dimension of the eigenvector can be understood as a potential feature or attribute of the data, and the numerical value in the vector indicates the strength of the state parameter on these features.

[0079] In some embodiments, a correlation analysis is performed on the state parameter feature vector to obtain a correlation strength matrix and an importance weight vector, including: S1031-S1036.

[0080] S1031. Calculate the state quotient of the dot product of each state parameter feature vector and the vector modulus, and calculate the product of the state quotients of every two state parameter feature vectors to obtain a state parameter similarity value.

[0081] For example, suppose there are two eigenvectors of state parameters a=[1, 2, 3] and b=[2, 3, 4]. First, calculate their dot product (a·b), and then calculate their moduli (|a| and |b|) respectively. The state quotient is equal to the dot product divided by the product of the moduli, that is, (a·b) / (|a|·|b|). The closer this value is to 1, the more similar the two vectors are. The state quotient is calculated for all eigenvectors in pairs, and the two state quotients are multiplied to obtain the state parameter similarity value. This calculation method takes into account the direction and size of the vector and can effectively measure the correlation between state parameters.

[0082] S1032. Construct a similarity matrix according to the state parameter similarity values, and perform eigenvalue decomposition on the similarity matrix to obtain an eigenvalue sequence and a corresponding eigenvector sequence.

[0083] Exemplarily, all the state parameter similarity values ​​obtained in the previous step are organized into a matrix form. For example, for n state parameters, an n×n symmetric matrix is ​​obtained. Each element (i, j) in the symmetric matrix represents the similarity value between the i-th and j-th state parameters. Then, the symmetric matrix is ​​subjected to eigenvalue decomposition to obtain a set of eigenvalues ​​and corresponding eigenvectors. The eigenvalues ​​reflect the importance of the main change direction of the data, while the eigenvectors represent the specific forms of these change directions.

[0084] S1033. Sort the eigenvalue sequence to obtain a sorted eigenvalue sequence, and screen the eigenvector sequence according to a preset threshold and the sorted eigenvalue sequence to obtain a main eigenvector set.

[0085] Exemplarily, the eigenvalue sequence is sorted from large to small, and the eigenvector corresponding to the larger eigenvalue contains more information. Set a preset threshold (such as retaining 90% of the information), and stop when the proportion of the cumulative eigenvalue exceeds this threshold, and retain the corresponding eigenvector. For example, if there are 10 eigenvectors, only the first 5 most important eigenvectors may be retained. This can reduce the data dimension while retaining the main information.

[0086] S1034. Construct an XGBoost model for the state parameters according to the main feature vector set to obtain a charging time prediction model, wherein the tree depth, number of leaf nodes and learning rate of the XGBoost model are preset parameters.

[0087] For example, the filtered feature vector is used as input to build an XGBoost model to predict charging time. Model parameters include: maximum tree depth (such as 6 layers), maximum number of leaf nodes per tree (such as 64), learning rate (such as 0.1), etc. XGBoost fits the residuals in the data by continuously building new decision trees. Each tree is optimized based on the previous tree to form a powerful integrated model.

[0088] S1035. Perform SHAP value calculation on each feature of the charging time prediction model to obtain a contribution value of the state parameter to the prediction result, wherein the SHAP value calculation is based on the Shapley value method in game theory.

[0089] Exemplarily, the SHAP (SHapleyAdditiveexPlanations) value is used to explain the contribution of each feature to the model prediction results. For example, for predicting charging time, the SHAP value of each feature such as battery temperature and remaining power can be calculated. This value represents the average change in the model prediction results when the feature is introduced. The calculation of the SHAP value takes into account the interaction between features and can more accurately reflect the importance of the features.

[0090] S1036. Normalize the contribution values ​​to obtain a correlation strength matrix and an importance weight vector.

[0091] In some embodiments, a deep neural network model is trained on the comprehensive state representation vector to obtain a charging strategy optimization model, including: S1041-S1047.

[0092] S1041. Hierarchically map the comprehensive state representation vector according to the structure of 256 nodes in the first hidden layer, 128 nodes in the second hidden layer, and 64 nodes in the third hidden layer to obtain a multilayer perceptron network.

[0093] For example, assuming that the dimension of the input comprehensive state representation vector is 50, the network structure is: input layer (50) → first hidden layer (256) → second hidden layer (128) → third hidden layer (64) → output layer. A full connection is used between each layer, for example, the connection weight matrix dimension from the input layer to the first hidden layer is 50×256. This decreasing node design can gradually extract the abstract representation of the feature, which is similar to the layer-by-layer compression and purification process of information.

[0094] S1042, multiplying the output value of each hidden layer node in the multilayer perceptron network by the slope parameter 0.2 and performing a comparison operation with the input value to obtain an activated hidden layer node sequence.

[0095] For example, the output of each hidden layer is nonlinearly transformed using the LeakyReLU activation function with a slope parameter of 0.2. For example, for input x, when x>0, the output is x, and when x≤0, the output is 0.2x. This activation function can avoid the dead neuron problem compared to the traditional ReLU because a small gradient is still retained on the negative semi-axis. This can maintain the expressive power of the network while improving the stability of training.

[0096] S1043. Subtract the mean from each batch of data in the activated hidden layer node sequence and divide by the standard deviation, then multiply by the scaling factor 1.5 and add an offset 0.5 to obtain a standardized feature sequence.

[0097] For example, the output sequence of a hidden layer node sequence after activation in a batch is [1, 2, 3, 4, 5]. First, the mean 3 and standard deviation 1.58 are calculated, and after standardization, [-1.26, -0.63, 0, 0.63, 1.26] are obtained. Then, it is multiplied by the scaling factor 1.5 to obtain [-1.89, -0.95, 0, 0.95, 1.89], and the offset 0.5 is added to obtain [-1.39, -0.45, 0.5, 1.45, 2.39]. This processing can accelerate the convergence of network training.

[0098] S1044. Multiply the mean square error between the predicted value and the actual value of the charging time in the standardized feature sequence by a coefficient of 0.5, multiply the difference between the predicted value and the target value of the battery temperature by a coefficient of 0.3, and multiply the difference between the predicted value and the ideal value of the charging efficiency by a coefficient of 0.2, and add them together to obtain a comprehensive loss function value.

[0099] For example, the predicted power supply time of a sample is 120 minutes, the actual value is 100 minutes, the MSE is 400, and the weight is 0.5, which contributes 200; the predicted value of the battery temperature is 30°C, the target value of the battery temperature is 25°C, the difference of 5 has a weight of 0.3, and contributes 1.5; the predicted value of the power consumption rate is 0.9, the ideal value is 0.95, the difference of 0.05 has a weight of 0.2, and contributes 0.01. The calculated comprehensive loss function value is 201.51.

[0100] S1045. Substitute the comprehensive loss function value into the Adam optimizer with a learning rate of 0.001, β1 of 0.9, and β2 of 0.999, and update the gradient by calculating the first-order moment estimate and the second-order moment estimate to obtain the optimized model parameters.

[0101] For example, the current value of a weight parameter of the Adam optimizer is 0.5, and its gradient is -0.1. The hyperparameters (lr=0.001, β1=0.9, β2=0.999) are used to calculate the first-order moment estimate and second-order moment estimate after bias correction to obtain the optimized parameter value. The Adam optimizer combines the advantages of the momentum method and RMSprop and can adaptively adjust the learning rate of each parameter.

[0102] S1046. Substitute the optimized model parameters into the validation data set, calculate the mean square error, mean absolute error and goodness of fit between the predicted value and the true value, and obtain a model performance evaluation indicator sequence.

[0103] For example, for 100 validation samples, the mean square error (MSE), mean absolute error (MAE) and R² value are calculated. If MSE=100, it means that the average deviation of the predicted value is 10 units; MAE=8 means that the average deviation of the predicted value is 8 units; R²=0.95 means that the model can explain 95% of the data variation. These indicators reflect the predictive ability of the model from different angles.

[0104] S1047. Input the model performance evaluation indicator sequence into the grid search algorithm with a step size of 0.1, and obtain the charging strategy optimization model by traversing the learning rate interval [0.0001, 0.01], the momentum parameter interval [0.8, 0.99] and the regularization coefficient interval [0.01, 0.1].

[0105] For example, for a learning rate in the interval [0.0001, 0.01], sampling is performed every 0.1 to obtain 100 candidate values; the momentum parameter and the regularization coefficient are sampled similarly. Model training and verification are performed for all parameter combinations, and the parameter combination with the best performance is selected as the charging strategy optimization model.

[0106] In some embodiments, the model performance evaluation indicator sequence is input into a grid search algorithm with a step size of 0.1, and the charging strategy optimization model is obtained by traversing the learning rate interval [0.0001, 0.01], the momentum parameter interval [0.8, 0.99] and the regularization coefficient interval [0.01, 0.1], including: S471-S475.

[0107] S471. Multiply the mean square error, mean absolute error and goodness of fit in the model performance evaluation index sequence by weight coefficients 0.4, 0.3 and 0.3 respectively, and then sum them up to obtain a comprehensive evaluation score.

[0108] S472. Construct a three-dimensional parameter grid according to the comprehensive evaluation score, and evenly divide the learning rate interval, momentum parameter interval, and regularization coefficient interval according to a step size of 0.1 to obtain a parameter combination matrix.

[0109] S473. Perform cross-validation on each set of parameters in the parameter combination matrix, divide the training data set into 5 subsets, use 4 subsets for training in turn, and use the remaining 1 subset for validation, to obtain a validation result sequence.

[0110] S474. Calculate the average performance index of each group of parameters according to the verification result sequence, and compare the average performance index with a preset threshold value to screen out parameter combinations with performance indexes greater than the threshold value to obtain a candidate parameter set.

[0111] S475. Optimize the parameter combinations in the candidate parameter set using a preset Bayesian function, establish a mapping relationship between the parameters and the model performance through Gaussian process regression, select the parameter combination with the largest expected increment, and obtain a charging strategy optimization model.

[0112] For example, by introducing a weighted comprehensive evaluation system, the three key indicators of mean square error, mean absolute error and goodness of fit are reasonably integrated to ensure the comprehensiveness and accuracy of model evaluation. Secondly, a three-dimensional parameter grid search method is used to systematically explore the optimal combination of learning rate, momentum parameter and regularization coefficient, avoiding the local optimal problem that may be caused by single parameter tuning.

[0113] In the specific implementation process, a 5-fold cross-validation method is used to evaluate each set of parameters. This method can effectively prevent overfitting and improve the generalization ability of the model. By setting the performance index threshold for parameter screening, obviously unsuitable parameter combinations can be quickly eliminated to improve optimization efficiency. By introducing the Bayesian optimization method and using Gaussian process regression to establish the mapping relationship between parameters and model performance, this method has higher search efficiency and better optimization effect than traditional grid search.

[0114] The above technical solution achieves efficient optimization of the charging strategy model parameters through the organic combination of multiple mechanisms such as comprehensive evaluation, grid search, cross-validation, threshold screening and Bayesian optimization. It not only ensures the reliability of the model performance, but also improves the efficiency of parameter tuning, providing strong technical support for the optimization of electric vehicle charging strategies.

[0115] In some embodiments, charging prediction is performed according to a charging strategy optimization model to obtain a charging optimization plan, including: obtaining a predicted charging time and an estimated charging completion time through the charging strategy optimization model; performing weighted averaging of the predicted charging time and the current charging power value to obtain a charging interval; and generating a charging optimization plan according to the charging interval and preset map information.

[0116] For example, when generating a charging optimization plan, the preset map information is combined, which includes multi-dimensional data such as the distribution of charging stations, traffic conditions, and the use of charging piles. By comprehensively analyzing this information, the system can recommend the best charging route and charging station selection for users, taking into account not only time efficiency but also the convenience of geographical location. This intelligent decision-making solution based on multi-source data can effectively improve the charging experience of electric vehicle users and reduce charging anxiety, which is of great significance to promoting the promotion and application of electric vehicles.

[0117] See also Figure 2 , Figure 2 1 is a schematic block diagram of a vehicle controller-based wireless charging control system provided in an embodiment of the present application, wherein the vehicle controller-based wireless charging control system 200 is used to execute the aforementioned vehicle controller-based wireless charging control method. The vehicle controller-based wireless charging control system 200 can be configured in a server.

[0118] Among them, the server can be an independent server or a server cluster, or it can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN) and big data and artificial intelligence platforms.

[0119] like Figure 2 As shown, the wireless charging control system 200 based on the vehicle controller includes: a data processing module 201, a vector determination module 202, an association analysis module 203, a feature fusion module 204, a model training module 205 and a result output module 206.

[0120] The data processing module 201 is used to pre-process the initial data generated by the vehicle operation to obtain a structured state information matrix.

[0121] The vector determination module 202 is used to determine the state parameter characteristic vector according to a preset vector change method and the structured state information matrix.

[0122] The correlation analysis module 203 is used to perform correlation analysis on the state parameter feature vector to obtain a correlation strength matrix and an importance weight vector.

[0123] The feature fusion module 204 is used to perform feature fusion on the state parameter feature vector according to the association strength matrix and the importance weight vector to obtain a comprehensive state representation vector.

[0124] The model training module 205 is used to perform deep neural network model training on the comprehensive state representation vector to obtain a charging strategy optimization model.

[0125] The result output module 206 is used to perform charging prediction according to the charging strategy optimization model to obtain a charging optimization plan, and the charging plan is used to prompt the vehicle to charge.

[0126] An embodiment of the present application provides an electronic device, which includes a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and implement a wireless charging control method based on a vehicle controller as any one of the embodiments of the present application when executing the computer program.

[0127] An embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the processor implements a wireless charging control method based on a vehicle controller as described in any one of the embodiments of the present application.

[0128] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A wireless charging control method based on a vehicle controller, characterized in that: For a vehicle controller, the method comprises: Preprocess the initial data generated by vehicle operation to obtain a structured state information matrix; Determining a state parameter characteristic vector according to a preset vector change method and the structured state information matrix; Calculate the state quotient of the dot product of each state parameter feature vector and the vector modulus, calculate the product of the state quotient of every two state parameter feature vectors, and obtain the state parameter similarity value; construct a similarity matrix according to the state parameter similarity value, and perform eigenvalue decomposition on the similarity matrix to obtain an eigenvalue sequence and a corresponding eigenvector sequence; sort the eigenvalue sequence to obtain a sorted eigenvalue sequence, and screen the eigenvector sequence according to a preset threshold and the sorted eigenvalue sequence to obtain a main eigenvector set; construct an XGBoost model for the state parameters according to the main eigenvector set to obtain a charging time prediction model, wherein the tree depth, number of leaf nodes and learning rate of the XGBoost model are preset parameters; perform SHAP value calculation on each feature of the charging time prediction model to obtain a contribution value of the state parameter to the prediction result, wherein the SHAP value calculation is based on the Shapley value method in game theory; normalize the contribution value to obtain an association strength matrix and an importance weight vector; Performing feature fusion on the state parameter feature vector according to the association strength matrix and the importance weight vector to obtain a comprehensive state representation vector; Performing deep neural network model training on the comprehensive state representation vector to obtain a charging strategy optimization model; A charging prediction is performed according to the charging strategy optimization model to obtain a charging optimization plan, and the charging optimization plan is used to prompt the vehicle to charge.

2. The wireless charging control method based on a vehicle controller as claimed in claim 1, characterized in that: The step of determining the state parameter characteristic vector according to the preset vector change method and the structured state information matrix includes: Performing context window division on the state parameters in the structured state information matrix to obtain a context data set of the state parameters, wherein the size of the context window is a preset value N; Constructing a co-occurrence matrix according to the context data set, wherein each element in the co-occurrence matrix represents the number of times corresponding two state parameters co-occur in the same context window; Constructing eigenvectors of the co-occurrence matrix to obtain initial eigenvectors; Constraining the initial eigenvector according to preset physical law constraints to obtain a constrained eigenvector; Performing dimension reduction on the constrained feature vector to obtain a feature vector after dimension reduction; Cluster analysis is performed based on the feature vector after dimension reduction to obtain the state parameter feature vector.

3. The wireless charging control method based on a vehicle controller as claimed in claim 2, characterized in that: The step of constructing a feature vector of the co-occurrence matrix to obtain an initial feature vector includes: Dividing each element in the co-occurrence matrix by the sum of the elements in the row to obtain a normalized co-occurrence matrix; The normalized co-occurrence matrix is ​​used to construct an input layer node matrix according to the product of the number of elements of the normalized co-occurrence matrix and the preset vector dimension; Constructing an output layer target matrix according to the dimensions of the co-occurrence matrix; The input layer node matrix is ​​multiplied by a preset weight matrix, and the cross entropy loss is calculated with the output layer target matrix, and the node weight matrix is ​​obtained by stochastic gradient descent iteration; Multiplying the input layer node matrix by the node weight matrix to obtain a feature mapping matrix; Decomposing the eigenmapping matrix into the product of a left singular matrix, a singular value diagonal matrix and a right singular matrix to obtain an eigendecomposition matrix; According to the preset vector dimension, the left singular matrix in the eigendecomposition matrix is ​​used to construct the initial eigenvector of the state parameter.

4. The wireless charging control method based on a vehicle controller as claimed in claim 1, characterized in that: The performing of deep neural network model training on the comprehensive state representation vector to obtain a charging strategy optimization model includes: The comprehensive state representation vector is hierarchically mapped according to the structure of 256 nodes in the first hidden layer, 128 nodes in the second hidden layer, and 64 nodes in the third hidden layer to obtain a multi-layer perceptron network; The output value of each hidden layer node in the multilayer perceptron network is multiplied by the slope parameter 0.2 and compared with the input value to obtain an activated hidden layer node sequence; Subtract the mean from each batch of data in the activated hidden layer node sequence, divide by the standard deviation, multiply by a scaling factor of 1.5, and add an offset of 0.5 to obtain a standardized feature sequence; The mean square error between the predicted value of charging time and the actual value in the standardized feature sequence is multiplied by a coefficient of 0.5, the difference between the predicted value of battery temperature and the target value is multiplied by a coefficient of 0.3, and the difference between the predicted value of charging efficiency and the ideal value is multiplied by a coefficient of 0.2, and then added to obtain a comprehensive loss function value; Substitute the comprehensive loss function value into the Adam optimizer with a learning rate of 0.001, β1 of 0.9, and β2 of 0.999, and update the gradient by calculating the first-order moment estimate and the second-order moment estimate to obtain the optimized model parameters; Substituting the optimized model parameters into the validation data set, calculating the mean square error, mean absolute error and goodness of fit between the predicted value and the true value, and obtaining a model performance evaluation index sequence; The model performance evaluation index sequence is input into a grid search algorithm with a step size of 0.1, and the charging strategy optimization model is obtained by traversing the learning rate interval [0.0001, 0.01], the momentum parameter interval [0.8, 0.99] and the regularization coefficient interval [0.01, 0.1].

5. The wireless charging control method based on a vehicle controller as claimed in claim 4, characterized in that: The model performance evaluation index sequence is input into a grid search algorithm with a step size of 0.1, and the charging strategy optimization model is obtained by traversing the learning rate interval [0.0001, 0.01], the momentum parameter interval [0.8, 0.99] and the regularization coefficient interval [0.01, 0.1], including: The mean square error, mean absolute error and goodness of fit in the model performance evaluation index sequence are multiplied by weight coefficients 0.4, 0.3 and 0.3 respectively, and then summed to obtain a comprehensive evaluation score; Constructing a three-dimensional parameter grid according to the comprehensive evaluation score, and evenly dividing the learning rate interval, momentum parameter interval, and regularization coefficient interval according to a step size of 0.1 to obtain a parameter combination matrix; Cross-validation is performed on each set of parameters in the parameter combination matrix, the training data set is divided into 5 subsets, 4 subsets are used for training in turn, and the remaining 1 subset is used for validation, to obtain a validation result sequence; Calculate the average performance index of each group of parameters according to the verification result sequence, and compare the average performance index with a preset threshold value to screen out parameter combinations with performance indexes greater than the threshold value to obtain a candidate parameter set; The parameter combinations in the candidate parameter set are optimized using a preset Bayesian function, a mapping relationship between parameters and model performance is established through Gaussian process regression, and the parameter combination with the largest expected increment is selected to obtain the charging strategy optimization model.

6. The wireless charging control method based on a vehicle controller as claimed in claim 1, characterized in that: The charging prediction is performed according to the charging strategy optimization model to obtain a charging optimization plan, including: The predicted charging time and the estimated charging completion time are obtained through the charging strategy optimization model; Performing weighted averaging of the predicted charging time and the current charging power value to obtain a charging interval; The charging optimization plan is generated according to the charging interval and preset map information.

7. A wireless charging control system based on a vehicle controller, characterized in that: The wireless charging control system based on the vehicle controller includes: A data processing module is used to pre-process the initial data generated by the vehicle operation to obtain a structured state information matrix; A vector determination module, used to determine a state parameter characteristic vector according to a preset vector change method and the structured state information matrix; An association analysis module is used to calculate the state quotient of the dot product of each state parameter feature vector and the vector modulus, calculate the product of the state quotient of every two state parameter feature vectors, and obtain the state parameter similarity value; construct a similarity matrix according to the state parameter similarity value, and perform eigenvalue decomposition on the similarity matrix to obtain an eigenvalue sequence and a corresponding eigenvector sequence; sort the eigenvalue sequence to obtain a sorted eigenvalue sequence, and screen the eigenvector sequence according to a preset threshold and the sorted eigenvalue sequence to obtain a main eigenvector set; construct an XGBoost model for the state parameters according to the main eigenvector set to obtain a charging time prediction model, wherein the tree depth, number of leaf nodes and learning rate of the XGBoost model are preset parameters; perform SHAP value calculation on each feature of the charging time prediction model to obtain a contribution value of the state parameter to the prediction result, wherein the SHAP value calculation is based on the Shapley value method in game theory; normalize the contribution value to obtain an association strength matrix and an importance weight vector; A feature fusion module, used for performing feature fusion on the state parameter feature vector according to the association strength matrix and the importance weight vector to obtain a comprehensive state representation vector; A model training module, used to perform deep neural network model training on the comprehensive state representation vector to obtain a charging strategy optimization model; The result output module is used to perform charging prediction according to the charging strategy optimization model to obtain a charging optimization plan, and the charging optimization plan is used to prompt the vehicle to charge.

Images