A method for optimizing transmission performance of a wireless power transmission system for an electric vehicle

Abstract

The application is suitable for the field of wireless power transmission technology, and provides a transmission performance optimization method for a wireless power transmission system of an electric vehicle, comprising the following steps: step 1, quantifying the transmission efficiency uncertainty of the electric vehicle WPT system; step 2, improving the multi-objective optimization algorithm according to the result in step 1; step 3, optimizing the transmission performance of the electric vehicle WPT system through the improved algorithm in step 2. The improved multi-objective hawk optimization algorithm is used to optimize the design of the WPT system compensation circuit and the magnetic energy coil group structure, so that the charging efficiency and robustness of the electric vehicle WPT system can be improved simultaneously; the multiple network ensemble learning method is used to improve the solving accuracy and calculation efficiency; the original optimization algorithm is improved, and the search accuracy and efficiency of the optimal structure parameter group of the WPT system are effectively improved.

Description

Technical Field

[0001] This invention belongs to the field of wireless power transmission technology, and particularly relates to a method for optimizing the transmission performance of a wireless power transmission system for electric vehicles. Background Technology

[0002] With the accelerating development of new energy electric vehicles, wireless power transfer (WPT) technology has become a focus of attention for many research institutions and automotive companies worldwide. Compared to traditional wired charging, WPT technology has many advantages, including waterproofing, dustproofing, operational safety, and no mechanical wear, making its application prospects quite broad. However, as the application of WPT technology in electric vehicles becomes more widespread, challenges are gradually emerging. A typical problem is that the transmission efficiency of WPT systems is quite sensitive to external interference factors, which to some extent limits the further development of this technology.

[0003] In the above problems, uncertainty quantification methods are of great significance because they can obtain relevant statistical characteristic parameters of the transmission efficiency of WPT systems, thereby providing guidance for the structural optimization design of WPT systems.

[0004] To address the uncertainty in the transmission efficiency of electric vehicle WPT systems, directly using the traditional Monte Carlo method to quantify the uncertainty of transmission efficiency leads to low computational efficiency.

[0005] Current research on WPT system transmission performance optimization typically employs single-objective optimization algorithms to optimize the structural parameters of the WPT system. However, since the optimization objective of a single-objective optimization algorithm is a single object, it cannot truly improve the transmission performance of the WPT system. Therefore, multi-objective optimization algorithms are needed. Summary of the Invention

[0006] The purpose of this invention is to provide a method for optimizing the transmission performance of a wireless power transfer system (WPT) for electric vehicles. This method aims to address the problem of suboptimal transmission performance in WPT systems under the influence of uncertainties. Directly using the traditional Monte Carlo method to quantify the uncertainty of transmission efficiency leads to low computational efficiency. Current research on WPT system transmission performance optimization typically employs single-objective optimization algorithms to optimize the structural parameters of the WPT system. However, since the optimization objective of a single-objective algorithm is limited to a single target, it cannot truly improve the transmission performance of the WPT system. Therefore, multi-objective optimization algorithms are needed.

[0007] The present invention is implemented as follows: a method for optimizing the transmission performance of a wireless power transmission system for electric vehicles, the method comprising the following steps:

[0008] Step 1: Quantification of transmission efficiency uncertainty in electric vehicle WPT system:

[0009] Considering the potential scenarios during wireless power transmission in electric vehicles, this paper establishes a quantitative model for the transmission efficiency uncertainty of the WPT system based on a deep learning network. This model takes the tilt angle between the transmitting and receiving coils, the vertical distance between them, the horizontal offset between their centers, the equivalent resistance of the transmitting circuit, the equivalent resistance of the receiving circuit, and the load resistance as random and uncertain input variables. The model obtains the mean, variance, and probability density distribution of the transmission efficiency within the fluctuation range. By comparing the results with those obtained using the classic Monte Carlo method, the accuracy and efficiency of deep learning in quantifying the uncertainty of wireless power transmission efficiency in electric vehicles are verified.

[0010] Step 2, Improvement of the multi-objective optimization algorithm:

[0011] An improvement was made to the multi-objective Eagle Optimization (AO) algorithm. In the population initialization stage, a good population allocation mechanism was formed by combining Tent chaotic mapping. In the global exploration stage, the exploration mode was improved by combining adaptive inertial weights. This improved the global search capability in the early stage of the algorithm iteration, making it less likely to get stuck in local optima. At the same time, it also improved the local search capability in the later stage of the algorithm iteration, thereby improving the search accuracy of the global optimum.

[0012] Step 3: Optimize the transmission performance of the electric vehicle WPT system:

[0013] Based on the uncertainty quantification results of the wireless power transmission efficiency of electric vehicles, this invention further adopts an improved multi-objective Tianying optimization algorithm to optimize the structure of the electric vehicle WPT system, and finally achieves a significant improvement in the transmission performance of the electric vehicle WPT system.

[0014] The present invention provides a method for optimizing the transmission performance of a wireless power transmission system for electric vehicles, which has the following beneficial effects:

[0015] 1. This invention mainly establishes an uncertainty quantification framework based on deep learning networks to obtain relevant statistical characteristic parameters such as the mean, variance, and probability density distribution that characterize the uncertainty of wireless power transmission efficiency in automobiles. Taking the mean and variance of transmission efficiency as optimization objectives, the improved multi-objective Tianying optimization algorithm is used to optimize the design of the compensation circuit and magnetic coil group structure of the WPT system, so as to simultaneously improve the charging efficiency and robustness of the electric vehicle WPT system.

[0016] 2. This invention proposes to use deep learning neural networks as a framework for quantifying the uncertainty of transmission efficiency in WPT systems. By introducing dropout and batch normalization modules into the network layers and utilizing the integration of multiple networks, the solution accuracy and computational efficiency are effectively improved.

[0017] 3. This invention proposes an improvement to the original Skyhawk optimization algorithm, which effectively improves the algorithm's search accuracy and efficiency for the optimal structural parameter set of the WPT system. Attached Figure Description

[0018] Figure 1 A flowchart illustrating the main calculation example of a method for optimizing the transmission performance of a wireless power transmission system for electric vehicles, provided in an embodiment of the present invention.

[0019] Figure 2 A schematic diagram of an SS topology compensation circuit for a wireless power transmission system for electric vehicles provided in an embodiment of the present invention;

[0020] Figure 3 A spatial misalignment view of the coil group exhibiting multi-directional horizontal offset during actual operation of an electric vehicle wireless power transmission system provided in this embodiment of the invention;

[0021] Figure 4 A spatial misalignment view of a coil group exhibiting both horizontal and angular offsets during actual operation of an electric vehicle wireless power transmission system, provided as an embodiment of the present invention.

[0022] Figure 5 A DNN structure diagram combining a dropout module is provided in an embodiment of the present invention;

[0023] Figure 6 A comparison chart of multiple MC simulation test results based on the wireless power transmission efficiency of electric vehicles is provided for an embodiment of the present invention;

[0024] Figure 7 A comparison chart of multiple DNN uncertainty quantization results based on wireless power transmission efficiency of electric vehicles is provided in an embodiment of the present invention.

[0025] Figure 8 A macroscopic schematic diagram of a wireless power transmission system for electric vehicles provided in an embodiment of the present invention;

[0026] Figure 9 A comparison diagram of the probability density distribution function of wireless power transfer efficiency of electric vehicles calculated by DNN and MC is provided in an embodiment of the present invention;

[0027] Figure 10 Comparison of algorithm simulation results for a method for optimizing the transmission performance of a wireless power transmission system for electric vehicles provided in an embodiment of the present invention;

[0028] Figure 11 This invention provides a method for optimizing the transmission performance of a wireless power transmission system for electric vehicles—a Pareto optimal solution of an improved multi-objective Skyhawk optimization algorithm.

[0029] Figure 12 This is a comparison diagram of the probability density distribution of wireless power transmission efficiency of electric vehicles before and after optimization, provided as an embodiment of the present invention. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0031] The specific implementation of the present invention will be described in detail below with reference to specific embodiments.

[0032] like Figure 1 As shown in the embodiment of the present invention, a method for optimizing the transmission performance of a wireless power transmission system for electric vehicles includes the following steps:

[0033] Step 1: Quantify the uncertainty of the transmission efficiency of the electric vehicle WPT system:

[0034] Considering the potential scenarios during wireless power transmission in electric vehicles, this paper establishes a framework for quantifying the uncertainty of transmission efficiency in WPT systems based on deep learning networks. This framework includes the tilt angle between the transmitting and receiving coils, the vertical distance between them, the horizontal offset between their centers, the equivalent resistance of the transmitting circuit's transmitting loop, the equivalent resistance of the receiving circuit's receiving loop, and the load resistance as random and uncertain input variables. The goal is to obtain relevant statistical parameters such as the mean, variance, and probability density distribution of the transmission efficiency within its fluctuation range. By comparing the results with those obtained using the classical Monte Carlo method, the accuracy and efficiency of deep learning in quantifying the uncertainty of wireless power transmission efficiency in electric vehicles are verified.

[0035] Step 2: Improve the multi-objective optimization algorithm:

[0036] This invention improves the multi-objective Eagle Optimization (AO) algorithm by combining Tent chaotic mapping in the population initialization stage to form a good population allocation mechanism, and by introducing adaptive inertial weights in the global exploration stage to improve the exploration mode. This improves the global search capability in the early stage of the algorithm iteration, making it less likely to get stuck in local optima, and also improves the local search capability in the later stage of the algorithm iteration, thereby improving the search accuracy of the global optimum.

[0037] Step 3: Optimize the transmission performance of the electric vehicle WPT system using the improved algorithm from Step 2:

[0038] Based on the uncertainty quantification results of the wireless power transmission efficiency of electric vehicles, this invention further adopts an improved multi-objective Tianying optimization algorithm to optimize the structure of the electric vehicle WPT system, and finally achieves a significant improvement in the transmission performance of the electric vehicle WPT system.

[0039] Specific procedures:

[0040] S1. Considering the possible situations that may occur in the actual charging process of the electric vehicle WPT system, select several random variables that affect the transmission efficiency and determine the distribution type of each variable;

[0041] S2. Establish a deep learning neural network model;

[0042] S3. Sample data from the distribution range of the above random variables as training samples for the network model, and use the transmission efficiency value corresponding to each group of samples as training labels to train the model.

[0043] S4. After the network model is trained, sample quantitative data from the distribution range of the above variables as test samples for the model, and calculate the mean, variance and probability density distribution function and other statistical characteristic parameters of the wireless power transmission efficiency of the WPT system.

[0044] S5. The Tent chaotic mapping and adaptive inertial weight strategy are combined into the Skyhawk optimization algorithm to obtain the improved Skyhawk optimization algorithm.

[0045] S6. Determine the structural parameters of the WPT system that need to be optimized;

[0046] S7. Based on the improved multi-objective Tianying optimization algorithm, the WPT system structure is optimized with the mean and variance of transmission efficiency within the fluctuation range as the optimization objective, and the optimal structural parameter set is obtained.

[0047] S8. The probability density distribution function of the optimized WPT system transmission efficiency is obtained by using a deep learning network. The effectiveness of this technical solution is verified by comparing it with the probability density distribution function of the transmission efficiency before optimization.

[0048] 1. Quantification of transmission efficiency uncertainty in electric vehicle WPT systems:

[0049] In recent years, uncertainty quantification based on machine learning theory has been vigorously promoted as an emerging technology. In the data preprocessing stage of machine learning, there are two main types of common uncertainty sources:

[0050] 1. The validation set and training set data do not match;

[0051] 2. The data contains internal interference;

[0052] In the scenario of wireless power transmission in electric vehicles, the transmission efficiency value is easily affected by random variables such as spatial dislocations of magnetic coil groups and parameters of compensation circuit components, and has strong uncertainty. Therefore, this invention mainly builds a deep learning network (DNN) to quantify the uncertainty of transmission efficiency for the second type of uncertainty source.

[0053] A DNN itself consists of many hidden layers. The relationship between the input x and the output y of a single node in a hidden layer can be expressed as:

[0054] (1)

[0055] In equation (1), σ() represents the nonlinear transfer function, w is the linear mapping, and b is the bias term. In the DNN model constructed in this invention, the ReLU function is selected as the nonlinear transfer function of the hidden layer nodes. Before training the DNN, the values ​​of w and b are randomly assigned. During network training, the values ​​of w and b are continuously updated through backpropagation until the error between the output value and the training label tends to be infinitesimal. At this point, it can be preliminarily considered that the DNN model has approximately fitted a model of uncertainty quantification of transmission efficiency. The evaluation index function of its training process can be expressed as:

[0056] (2)

[0057] In equation (2), N represents the number of training samples, and y i y* represents the network output value, and y* represents the training label. During the data acquisition phase of this DNN model, the following parameters will be considered: the tilt angle α between the transmitting and receiving coils of the electric vehicle WPT system; the vertical distance d between the transmitting and receiving coils; the horizontal offset Δx and Δy between the centers of the transmitting and receiving coils; the equivalent resistance R1 of the transmitting circuit; the equivalent resistance R2 of the receiving circuit; and the load resistance R. L As input variables for the model, this invention adopts a multi-network integration approach to build a three-part DNN model to quantify the uncertainty of the transmission efficiency of the WPT system. Figure 2 , Figure 3 and Figure 4 The figures show the SS compensation circuit topology used in the study and a spatial view of the coil misalignment.

[0058] 1) Part One: DNN Model

[0059] The input quantities are α, d, Δx, and Δy, and the output quantity is the mutual inductance of the coil M. The first part of the DNN model consists of six ordinary fully connected layers and one batch normalization layer. The number of nodes in the six fully connected layers from beginning to end are 4, 64, 32, 32, 16, and 1, respectively. The number of input features in the batch normalization layer is 32. Introducing the batch normalization layer can, to some extent, prevent overfitting during DNN training and enhance the model's generalization ability. Its mathematical model can be expressed as:

[0060] (3)

[0061] (4)

[0062] (5)

[0063] In equation (3), γ and β are parameter vectors, with default values ​​of 0 and 1 respectively, while ε is used to ensure numerical stability, with a default value of 1e. -5

[0064] 2) Part Two: DNN Model

[0065] The second part, the DNN model, is a single fully connected layer consisting of three neuron nodes, used to process R1, R2, and R... L The data is compressed so that the model can learn and absorb information more effectively.

[0066] 3) Part Three: DNN Model

[0067] The third part of the DNN model consists of five fully connected layers and a dropout layer. The number of nodes in the five fully connected layers from front to back are 4, 64, 32, 16, and 1, respectively. Dropout functions similarly to batch normalization and can also weaken the overfitting effect of DNN to some extent.

[0068] Combined with appendix Figure 5 It can be seen that by introducing the dropout module, some neuron nodes can be randomly discarded, thereby reducing the number of intermediate features and preventing the DNN from being too closely aligned with the training set samples. After incorporating the dropout module, the calculation formula for neuron nodes is updated as follows:

[0069] (6)

[0070] (7)

[0071] In equation (6), r l Let p represent a random number that follows a Bernoulli distribution, and let p represent the corresponding probability. In the DNN model built in this invention, p is set to 0.5.

[0072] After the DNN is trained, by analyzing R1, R2, and R... L Quantitative data is collected again within the distribution range of random variables α, d, Δx, and Δy as test samples for the DNN model, and the uncertainty statistical characteristic parameters of the wireless power transmission efficiency of the WPT system can be calculated. Regarding the selection of the number of training and test samples, the ideal range for the number of samples can be estimated by observing the training effect of the DNN under different sample allocation conditions. Multiple Monte Carlo (MC) simulation experiments have shown that when the number of samples is 10,000 or more, the obtained probability density distribution of the wireless power transmission efficiency of the WPT system is relatively accurate. To save on related computational costs, the MC results of 10,000 samples are used as the standard.

[0073] Combined with appendix Figure 6 and attached Figure 7 The experimental results comparison chart shows that when the number of test samples for DNN is 10,000, and the total number of samples is 50,000, 100,000, 200,000, 290,000, 300,000, and 310,000, it can be concluded that the uncertainty quantization result of DNN with a total number of samples of 300,000 is most similar to the MC result with 10,000 samples. When the total number of samples is less than 300,000, the training effect of DNN model is not ideal, while when the total number of samples is greater than 300,000, DNN model has a large deviation from MC due to overfitting. Therefore, it is roughly estimated that the ideal value of the total number of DNN samples is about 300,000.

[0074] 2. Improved multi-objective optimization algorithm:

[0075] This invention improves the Eagle Optimization (AO) algorithm by combining an adaptive inertia weighting strategy with a Tent chaotic mapping mechanism. It aims to optimize the WPT system using the improved multi-objective Eagle Optimization algorithm. The AO algorithm's optimization process mainly consists of five stages: population initialization, global exploration, local exploration, global mining, and local mining, which will be described in detail below:

[0076] 1) Population initialization

[0077] (8)

[0078] In equation (8), X min Denotes the lower bound of a variable, X max This indicates the upper bound of the variable, and rand is a random number between 0 and 1;

[0079] 2) Global Exploration

[0080] (9)

[0081] In equation (9), X1(t+1) represents the solution generated after (t+1) iterations during the global exploration process, X best (t) represents the optimal solution obtained before the t-th iteration, T represents the maximum number of iterations, and X M (t) represents the average value of the current solution at the t-th iteration, and rand is a random number between 0 and 1;

[0082] 3) Local exploration

[0083] (10)

[0084] (11)

[0085] (12)

[0086] In the above system of equations, X2(t+1) represents the solution generated after (t+1) iterations during the local exploration process, X best (t) represents the optimal solution obtained before the t-th iteration, X R (t) represents the random solution obtained within the population size range at the t-th iteration, and Levy(D) represents the Levy flight distribution function of the eagle. In its mathematical model, s and q are fixed constants, taking values ​​of 0.01 and 1.5 respectively, and μ and v represent the distributions following N(0, σ). 2 Random numbers from a Gaussian distribution of N(0,1) and N(0,1).

[0087] 4) Global mining

[0088] (13)

[0089] In equation (13), X3(t+1) represents the solution generated after (t+1) iterations in the global mining process, X best (t) represents the approximate position of the prey before the t-th iteration, X M (t) represents the average value of the current solution at the t-th iteration, where ξ and δ are both mining adjustment parameters;

[0090] 5) Local mining

[0091] (14)

[0092] (15)

[0093] In the formula, X4(t+1) represents the solution generated after (t+1) iterations in the local mining process, X(t) represents the current solution at the t-th iteration, t and T represent the current iteration number and the maximum iteration number, respectively, g(rand) is a random number in the interval (-0.2, 1), which reflects the various movement trajectories of the eagle when tracking its prey at close range, while the scaling factor 2 (1-(t / T)) of Levy(D) reflects the flight speed of the eagle. As the number of iterations increases, the eagle will gradually tend to stop and eventually track the target.

[0094] Although the AO algorithm has excellent search capabilities and can be used to solve many target optimization problems, it still has some shortcomings in the population initialization stage and the global exploration stage. Therefore, this invention will improve the AO algorithm around the above two parts.

[0095] 1) Population initialization based on chaotic mapping

[0096] In the initial stage of Algorithm Optimization (AO), the random allocation of individual population positions can easily lead to uneven population distribution, directly impacting the search speed and accuracy in later stages. To mitigate this deficiency, Tent chaotic mapping is introduced to create a favorable population allocation mechanism, thereby effectively improving population diversity and idealizing the initial distribution. The mathematical model of chaotic mapping can be expressed as follows:

[0097] (16)

[0098] 2) Adaptive inertia weights

[0099] When the Algorithm Exploration (AO) is in its first exploration phase, its behavior pattern fails to adequately balance global and local search capabilities. Therefore, this invention proposes an adaptive inertia weight to improve the mathematical model of the exploration phase: In the early stages of algorithm iteration, the eagle moves at a faster speed, enhancing global search capabilities and reducing the likelihood of getting trapped in local optima. In the later stages of algorithm iteration, the eagle moves at a slower speed, enhancing local search capabilities and improving the accuracy of global optimization. The new AO behavior pattern, after introducing the adaptive inertia weight, can be expressed as:

[0100] (17)

[0101] 3. Optimization of transmission performance of WPT system in electric vehicles

[0102] This invention utilizes the uncertainty quantification results from deep learning and further employs an improved multi-objective AO algorithm to optimize the magnetic energy coil assembly and compensation circuit structure of the electric vehicle WPT system, thereby improving the transmission performance of the WPT system under uncertain interference. Considering the constraints of the WPT system coil assembly size and arrangement, as well as the compensation circuit structure, the optimized design parameters are set as follows: (e.g., transmitting coil radius r) a The equivalent resistance R1 of the transmitting circuit of the compensation circuit, the equivalent resistance R2 of the receiving circuit of the compensation circuit, and the load resistance R L As the number of algorithm iterations increases, AO will eventually obtain the optimal solutions for the mean and variance of the wireless power transfer efficiency of the WPT system, as well as the corresponding optimal set of system structural parameters. Under optimal design conditions, the transmission performance of the electric vehicle WPT system will achieve significant improvement.

[0103] This invention establishes a simulation model of an electric vehicle WPT system, in which the compensation circuit adopts an SS topology and the magnetic energy coil group is a circular coil. A detailed structural diagram is shown below. Figure 8 As shown. Assume the vertical gap between the transmitting and receiving coils is 0.2m, both the transmitting and receiving coils have 10 turns, and the cross-sectional area of ​​each turn is 3e. -4 m 2 The resonant frequency of the WPT system is 2πe 5 rad / s, load resistance R L The value is 10Ω. Next, this invention will conduct simulation experiments to address the uncertainty quantification problem of transmission efficiency of this WPT system model and the optimization design of system structure.

[0104] Considering that during the actual charging process of the WPT system, coil misalignment caused by the driver's lack of skill and random fluctuations in the compensation circuit parameters can significantly impact the transmission efficiency of the WPT system, a deep learning-based uncertainty quantification framework is used to quantify the uncertainties, using coil misalignment parameters α, d, Δx, Δy and compensation circuit parameters R1, R2, R... L The distribution intervals of each variable, which serves as the input variable to the model, are shown in Table 1.

[0105] Table 1. Uncertain variables considered in WPT application examples

[0106]

[0107] Based on the above simulation model of the electric vehicle WPT system, this invention uses both the classical Monte Carlo (MC) model and the established DNN model to quantify the uncertainty of wireless power transmission efficiency, and further compares the statistical characteristic parameters of transmission efficiency calculated by the two methods. Let the sampling number of MC be 10,000, and the training and validation sample numbers of DNN be 290,000 and 10,000 respectively. The comparison results can be found in Table 2 and... Figure 9 As shown,

[0108] Table 2. Comparison of uncertainty quantification results between MC and DNN

[0109]

[0110] Based on the simulation comparison above, it can be seen that the DNN model proposed in this invention has the same uncertainty quantification accuracy as the MC approximation, while improving the solution rate by approximately 200% and significantly reducing the computational cost. In the subsequent WPT system structure optimization design, the statistical characteristic parameters of transmission efficiency will be used as the optimization objective, and the improved multi-objective AO algorithm will be used to improve the transmission performance of the WPT system. Let the mean of the optimization design parameters be R. L R1, R2, r a The design ranges for each parameter are [R] L / 2,3*R L / 2], [R1 / 2, 3*R1 / 2], [R2 / 2, 3*R2 / 2], [r a / 2,3*r a [2] Then, an improved multi-objective AO algorithm is used to find the optimal parameter combination within the design interval. Its goal is to improve the mean of transmission efficiency within the fluctuation range and reduce the variance of transmission efficiency within the fluctuation range, so as to enable the WPT system to carry out wireless power transmission efficiently and stably under the influence of random uncertain factors.

[0111] To verify the effectiveness of the improved AO in tracking the optimal solution, a further comparison was conducted between the improved AO, the traditional AO, and the classic bi-objective optimization algorithm NSGA-II. The objective value was set as the difference between the variance and mean of the transmission efficiency, and the population size for all three algorithms was 20, with a maximum number of iterations of 14. The comparison results of each algorithm in tracking the optimal solution are as follows: Figure 10 As shown;

[0112] pass Figure 10It can be seen that the NSGA-II has a low tracking speed and is prone to getting trapped in local optima, which reduces the accuracy of finding the global optimum. Although using traditional AO can effectively improve the tracking speed, it can also get trapped in local optima and still cannot accurately find the global optimum. In contrast, the improved AO has a superior tracking speed and accuracy, and is therefore more suitable for solving the structural optimization design problem of electric vehicle WPT systems.

[0113] Figure 11 The figure shows the Pareto optimal solution set of the improved multi-objective AO algorithm, which can intuitively show the transmission efficiency statistical moment characteristics corresponding to different design parameter sets of the WPT system.

[0114] Based on the simulation results of the Pareto optimal solution, the optimal design parameter set (R) can be obtained. L R1, R2, r a )opt, calculated using a deep learning model (R L R1, R2, r a The transmission efficiency statistical characteristics of the )opt parameter group are compared with those before optimization. It can be seen that the mean and variance of the transmission efficiency have achieved ideal changes after optimization by the improved multi-objective AO algorithm. Table 3 shows the WPT system design parameter group and corresponding transmission efficiency statistical characteristics before and after optimization. Figure 12 The image shows a comparison of the probability distribution of WPT system transmission efficiency before and after optimization, derived from a deep learning model:

[0115] Table 3. Comparison of WPT system parameters and transmission efficiency before and after optimization

[0116]

[0117] The simulation examples above demonstrate that the deep learning model proposed in this invention achieves uncertainty quantification accuracy comparable to that of the Monte Carlo Tree (MC) model, while also possessing higher computational efficiency. Furthermore, by optimizing the WPT system structure based on the uncertainty quantification results and incorporating an improved multi-objective AO algorithm, the mean efficiency of automotive wireless power transfer increases by 25.7%, the variance decreases by 65.4%, and the probability density distribution achieves a significantly more ideal result. Therefore, when unavoidable uncertainties arise during electric vehicle charging, the WPT system optimized by the proposed scheme exhibits higher charging efficiency and stronger robustness, better meeting practical engineering requirements.

[0118] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for optimizing the transmission performance of a wireless power transmission system for electric vehicles, characterized in that, The method for optimizing the transmission performance of the wireless power transmission system for electric vehicles includes the following steps: Step 1: Quantify the uncertainty of the transmission efficiency of the electric vehicle WPT system; Step 2: Based on the results in Step 1, improve the multi-objective optimization algorithm; Step 3: Optimize the transmission performance of the electric vehicle WPT system using the improved algorithm from Step 2; The quantification of the transmission efficiency uncertainty of the electric vehicle WPT system includes: S1. Considering the possible situations that may occur in the actual charging process of the electric vehicle WPT system, select several random variables that affect the transmission efficiency and determine the distribution type of each variable; S2. Establish a deep learning neural network model; The improvement of the multi-objective optimization algorithm includes the following steps: S3. Sample data from the distribution range of the random variables in S1 as training samples for the network model. Use the transmission efficiency value corresponding to each group of samples as training labels to train the model. S4. Sample quantitative data from the distribution range of the variables as test samples for the model, and calculate the mean, variance, and statistical characteristic parameters of the probability density distribution function of the wireless power transmission efficiency of the WPT system. S5. The Tent chaotic mapping and adaptive inertia weight strategy are integrated into the multi-objective Skyhawk optimization algorithm to obtain an improved multi-objective Skyhawk optimization algorithm. The optimization of the transmission performance of the electric vehicle WPT system includes the following steps: S6. Determine the structural parameters of the WPT system that need to be optimized; S7. Based on the improved multi-objective Tianying optimization algorithm, the WPT system structure is optimized with the mean and variance of transmission efficiency within the fluctuation range as the optimization objective; S8. The probability density distribution function of the optimized WPT system transmission efficiency is obtained by using a deep learning network.

2. The method for optimizing the transmission performance of a wireless power transmission system for electric vehicles according to claim 1, characterized in that, The relationship between the input x and output y of a single node in the hidden layer of the deep learning neural network model described in S2 is as follows: （1） In equation (1), σ() represents the nonlinear transfer function, w is the linear mapping, and b is the bias term; When the error between the output value and the training label approaches infinity, the evaluation metric function for the training process is: （2） N in formula (2) represents the number of training samples, y i represents the network output value, and y* represents the training label. During the data acquisition phase of this deep learning neural network model, the following parameters were considered: the tilt angle α between the transmitting and receiving coils of the electric vehicle WPT system; the vertical distance d between the transmitting and receiving coils; the horizontal offset Δx and Δy between the centers of the transmitting and receiving coils; the equivalent resistance R1 of the transmitting circuit; the equivalent resistance R2 of the receiving circuit; and the load resistance R. L As input variables for the model, multiple network integration methods are used to build a first part DNN model, a second part DNN model, and a third part DNN model to realize the uncertainty quantification of the transmission efficiency of the WPT system.

3. The method for optimizing the transmission performance of a wireless power transmission system for electric vehicles according to claim 2, characterized in that, The first part of the DNN model has inputs of α, d, Δx, and Δy, and an output of the coil mutual inductance M. This first part of the DNN model consists of six ordinary fully connected layers and one batch normalization layer. The number of nodes in the six fully connected layers from beginning to end are 4, 64, 32, 32, 16, and 1, respectively. The batch normalization layer has 32 input features. Its mathematical model is as follows: （3） （4） （5） In equation (3), γ and β are parameter vectors, with default values ​​of 0 and 1 respectively, while ε is used to ensure numerical stability, with a default value of 1e. -5 .

4. The method for optimizing the transmission performance of a wireless power transmission system for electric vehicles according to claim 2, characterized in that, The second part of the DNN model is a single fully connected layer consisting of three neuron nodes, used to process R1, R2, and R... L The data is compressed to improve the model's ability to learn and extract information.

5. The method for optimizing the transmission performance of a wireless power transmission system for electric vehicles according to claim 2, characterized in that, The third part of the DNN model consists of five fully connected layers and a dropout layer. The number of nodes in the five fully connected layers from the beginning to the end are 4, 64, 32, 16, and 1, respectively. After incorporating the dropout module, the formula for calculating neuron nodes is updated as follows: （6） （7） In equation (6), r l Let p represent a random number that follows a Bernoulli distribution, and p represent the corresponding probability. The ideal value of p is 0.

5. After the deep learning neural network is trained, by analyzing R1, R2, and R... L By collecting quantitative data again within the distribution range of random variables α, d, Δx, and Δy as test samples for the deep learning neural network model, the uncertainty statistical characteristic parameters of the wireless power transmission efficiency of the WPT system can be calculated.

6. The method for optimizing the transmission performance of a wireless power transmission system for electric vehicles according to claim 1, characterized in that, The optimization process of the improved multi-objective Skyhawk optimization algorithm includes: population initialization stage, global exploration stage, local exploration stage, global mining stage, and local mining stage; The mathematical model for population initialization is as follows: （8） In equation (8), X min Denotes the lower bound of a variable, X max This indicates the upper bound of the variable, and rand is a random number between 0 and 1; The mathematical model for the global exploration is as follows: （9） In equation (9), X1(t+1) represents the solution generated after (t+1) iterations during the global exploration process, X best (t) represents the optimal solution obtained before the t-th iteration, T represents the maximum number of iterations, and X M (t) represents the average value of the current solution at the t-th iteration, and rand is a random number between 0 and 1; The mathematical model for the local exploration is as follows: （10） （11） （12） In equations (10), (11), and (12), X2(t+1) represents the solution generated after (t+1) iterations during the local exploration process, X best (t) represents the optimal solution obtained before the t-th iteration, X R (t) represents the random solution obtained within the population size range at the t-th iteration, and Levy(D) represents the Levy flight distribution function of the eagle. In its mathematical model, s and q are fixed constants, taking values ​​of 0.01 and 1.5 respectively, and μ and v represent the distributions following N(0, σ). 2 Random numbers from a Gaussian distribution of N(0,1) and N(0,1). The mathematical model for global mining is as follows: （13） In equation (13), X3(t+1) represents the solution generated after (t+1) iterations in the global mining process, X best (t) represents the approximate position of the prey before the t-th iteration, X M (t) represents the average value of the current solution at the t-th iteration, where ξ and δ are both mining adjustment parameters; The mathematical model for the localized mining is as follows: （14） （15） In the formula, X4(t+1) represents the solution generated after (t+1) iterations in the local mining process, X(t) represents the current solution at the t-th iteration, t and T represent the current iteration number and the maximum iteration number, respectively, g(rand) is a random number in the interval (-0.2, 1), which reflects the various movement trajectories of the eagle when it closely tracks its prey, while the scaling factor 2 (1-(t / T)) of Levy(D) reflects the flight speed of the eagle.

7. The method for optimizing the transmission performance of a wireless power transmission system for electric vehicles according to claim 6, characterized in that, The aforementioned population initialization phase, In the initial stage of AO, the Tent chaotic mapping is introduced. The mathematical model of the chaotic mapping can be expressed as: （16） The aforementioned global exploration phase, When the AO is in the global exploration phase, an adaptive inertia weight is introduced, and the new AO behavior pattern can be represented as: （17）。

Images