CN114626301B - Neural network model construction method for analyzing influence of surrounding metal environment of EC-WPT system - Google Patents

Abstract

The present invention provides a method for constructing a neural network model for analyzing the influence of the metal environment around an EC‑WPT system. The neural network algorithm is used to approximately solve the functional relationship between different characteristics of the metal environment and the system energy efficiency in a nonlinear fitting manner. According to the weight and bias matrix obtained by training, the energy efficiency model of the EC‑WPT system under the metal environment is derived. The model can accurately analyze the influence of different types of metal environments on the energy efficiency characteristics of the system, effectively solving the problems of low precision of the current simulation method, high cost of the experimental method, high difficulty in solving the general formula derivation method and complex derivation process, and provides theoretical guidance for further studying related methods for suppressing the influence of the metal environment on the energy efficiency characteristics of the EC‑WPT system.

Description

A neural network model construction method for analyzing the impact of metal environment around EC-WPT system

Technical Field

The present invention relates to wireless power transmission technology, and in particular to a method for constructing a neural network model for analyzing the influence of a metal environment around an EC-WPT system.

Background Art

Wireless Power Transfer (WPT) technology refers to the technology that uses magnetic field, electric field, microwave and other media to transmit electric energy from power grid or battery to power-consuming equipment in a non-electrical contact manner by comprehensive application of electrical theory, power electronics technology and control theory. It has the advantages of safety, convenience, reliability and flexibility, and can effectively solve the problems of limited equipment flexibility and safety hazards caused by traditional wired power supply methods. Magnetic Coupling Wireless Power Transfer (MC-WPT) and Electric-field Coupling Wireless Power Transfer (EC-WPT) are the two WPT systems that have attracted the most attention from domestic and foreign researchers. Among them, MC-WPT system is more mature and widely used. Since electric field is similar to magnetic field in many characteristics, and the two show duality in basic theory, EC-WPT system with electric field as transmission medium has also attracted great attention from domestic and foreign scholars. Compared with the MC-WPT system, the coupling mechanism of the EC-WPT system is light in weight, changeable in shape and low in cost; the electromagnetic interference (EMI) generated by the coupling mechanism is low; the eddy current loss caused on the metal conductors between or around the coupling mechanism is very small; and it can also transmit energy across the metal barrier. Due to the above characteristics of EC-WPT technology, this technology has become one of the current research hotspots in the WPT field. Domestic and foreign scholars have conducted research on many aspects of EC-WPT technology, including system modeling, coupling mechanism, high-frequency power converter, resonant network, control method and parallel transmission of power signals. After development in recent years, EC-WPT technology has been able to meet the needs of many WPT applications in terms of transmission distance and transmission power, and has gradually been applied to consumer electronics, electric vehicles, industrial manufacturing and underwater equipment.

For the above common EC-WPT system practical application scenarios, there are generally metal objects in the surrounding environment. For example: the support frame of most mobile phones will use metal materials to ensure the longitudinal and lateral pressure resistance, bending resistance and heat dissipation capacity of the fuselage; the shells of most common cars and ships are also metal materials; the drill pipes in most oil drilling wells will also use high-grade steel to ensure that the drill pipes can withstand huge internal and external pressures, twisting, bending and vibration; for underwater unmanned submersibles (Unmanned Underwater Vehicle, UUV), although its shell is made of deep-sea buoyancy material (providing positive buoyancy for UUV), it still contains a large number of metal parts and metal mechanical structures, and the transmitter base station of its wireless charging system is on the seabed, and generally uses high-strength, high-pressure, low-temperature and corrosion-resistant alloy steel plate materials; the common stereo garage is a large mechanical structure, and its longitudinal beams, connecting blocks, bottom brackets and vertical plates are all made of metal bodies. It can be seen that for the EC-WPT system, it is very common for the system to be in a metal environment, and the structure of the metal environment usually changes with different application scenarios.

Although the EC-WPT system will not generate large eddy current losses on the metal bodies in the surrounding environment, the metal conductors in the surrounding environment will form cross-coupling capacitance with the system coupling mechanism plates. The generation of cross-coupling capacitance will mainly affect the equivalent capacitance _CS of the system coupling mechanism. In different application scenarios, the size, material, and distance of metal objects in the environment around the EC-WPT system from the system coupling mechanism are different, that is, the degree of influence on the equivalent capacitance _CS of the coupling mechanism is different. Changes in the equivalent capacitance _CS of the coupling mechanism of the EC-WPT system will affect the energy efficiency characteristics of the system and even cause the system resonant frequency to drift. At present, domestic and foreign research on the impact of metal conductors on WPT systems focuses on the field of foreign object detection (FOD) in MC-WPT systems. For the EC-WPT system, although the system can transfer energy across metal barriers and will not generate large eddy currents, there is still a lack of relevant research methods and theoretical results at home and abroad on the influence of the metal environment around the system on the energy efficiency characteristics of the system. In most application scenarios of the EC-WPT system, it is difficult to avoid the situation where the system is in a metal environment. Therefore, it is necessary to conduct in-depth research on the impact of metal environment on the energy efficiency characteristics of EC-WPT system, and lay a theoretical foundation for further research on key technologies of EC-WPT system under metal environment and related methods to inhibit the impact of metal environment on the energy efficiency characteristics of EC-WPT system.

Summary of the invention

Based on the above needs, the purpose of the present invention is to propose a method for constructing a neural network model for analyzing the influence of the metal environment around the EC-WPT system. The method uses a neural network algorithm to approximate the functional relationship between different characteristics of the metal environment and the system energy efficiency by nonlinear fitting. According to the weight and bias matrix obtained by training, the energy efficiency model of the EC-WPT system under the metal environment is derived. Through this model, the influence of different types of metal environments on the energy efficiency characteristics of the system can be accurately analyzed, which effectively solves the problems of low precision of the current simulation method, high cost of the experimental method, high difficulty in solving the general formula derivation method and complex derivation process, and provides theoretical guidance for further research on related methods to suppress the influence of the metal environment on the energy efficiency characteristics of the EC-WPT system.

In order to achieve the above object, the specific technical solutions adopted by the present invention are as follows:

A method for constructing a neural network model for analyzing the impact of the metal environment around an EC-WPT system includes the following steps:

S1: Determine the circuit structure of the EC-WPT system and the surrounding metal environment variables;

S2: The output power of the current EC-WPT system circuit structure under the influence of different surrounding metal environments is obtained through experiments and used as a sample data set;

S3: Processing the sample data set, encoding multiple surrounding metal environment variables to form an input feature vector, and taking the output power as the output variable;

S4: construct a neural network model and train it using the input feature vector and output variable processed in step S3;

S5: Optimize the hyperparameters of the neural network model through grid search method;

S6: Based on the weight and bias matrix of the neural network, the optimal energy efficiency model is derived as the neural network model for analyzing the impact of the metal environment around the EC-WPT system.

Optionally, the surrounding metal environment variables determined in step S1 include distance and/or area and/or material and/or position.

Optionally, step S2 obtains the output power of the current EC-WPT system circuit structure under the influence of different surrounding metal environments through physical experiments.

Optionally, in step S3, one-hot coding is used to encode the variables representing the positions in the sample data set.

Optionally, the number of neurons in the input layer of the neural network model constructed in step S4 is 5, the number of neurons in the output layer is 1, and the activation function adopts the Sigmoid function.

Optionally, when the hyperparameters of the neural network model are searched through grid search in step S5, the search range of the number of hidden layers is (2, 3, 4), the search range of the number of hidden layer neurons is (16, 32, 64, 128), and the search range of the batch size is (4, 8, 16, 32).

Optionally, when the circuit structure of the EC-WPT system is bilateral LC compensation, and the surrounding metal environment variables are distance, area, and position, the optimal energy efficiency model obtained in step S6 has 2 hidden layers, 128 hidden layer neurons, and a batch size of 4;

Input vector of hidden layer 1 represents the input vector of the neural network model, d _me represents the distance of the metal environment, a _me represents the area of the metal environment, One-hot encoding vector representing the location of the metal environment, and They represent the connection weight matrix and bias of the input layer and hidden layer 1 of the neural network model respectively. The output vector of hidden layer 1 is represents a real number, and the superscript represents the data dimension;

Input vector of hidden layer 2 and the output vector The calculation formula is: in, and Respectively represent the connection weight matrix and bias of hidden layer 1 and hidden layer 2 of the neural network model;

The system output power obtained at the output layer Where l _me represents the location of the metal environment, and They represent the connection weight matrix and bias between the hidden layer 2 and the output layer of the neural network model respectively.

Optionally, the optimal energy efficiency model obtained in step S6 is loaded onto a host computer, a Raspberry Pi, a PC terminal or a smart terminal to form an EC-WPT system surrounding metal environment impact analysis system with human-computer interaction function, and by inputting the restricted range of the metal environment characteristic parameter values and the metal environment parameters to be evaluated, the optimal energy efficiency model is called to perform online real-time calculation and analysis of the actual output power of the system.

Optionally, the output power data calculated by the optimal energy efficiency model obtained in step S6 is used to draw an influence curve between different values of the surrounding metal environment variables and the system output power.

The effects of the present invention are:

The present invention utilizes a neural network algorithm to approximately solve the functional relationship between different characteristics of the metal environment and the system energy efficiency by means of nonlinear fitting. According to the weight and bias matrix obtained through training, the energy efficiency model of the EC-WPT system under the metal environment is derived. Through this model, the influence of different types of metal environments on the energy efficiency characteristics of the system can be accurately analyzed, which effectively solves the problems of low precision of the current simulation method, high cost of the experimental method, high difficulty in solving the general formula derivation method and complex derivation process, and provides theoretical guidance for further research on related methods to suppress the influence of the metal environment on the energy efficiency characteristics of the EC-WPT system.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the specific implementation of the present invention or the technical solution in the prior art, the drawings required for use in the specific implementation or the description of the prior art are briefly introduced below.

FIG1 is a flow chart of a method for constructing a neural network model for analyzing the impact of the metal environment around an EC-WPT system provided by the present invention;

FIG2 is a circuit diagram of an EC-WPT system in a specific embodiment of the present invention;

FIG3 is a schematic diagram of a coupling mechanism in a specific embodiment of the present invention;

FIG4 is a schematic diagram of the corresponding relationship of metal environment variables in a specific embodiment of the present invention;

Figure 5 is a physical experimental data curve showing the influence of different characteristics of metal environment on the output power of EC-WPT system;

Figure 6 is a curve showing the change of relative error with the number of training times;

Figure 7 is a data curve showing the influence of metal environment on system energy efficiency characteristics;

FIG8 is a diagram of a neural network energy efficiency model of an EC-WPT system in a metal environment constructed by the present invention;

FIG9 is a schematic diagram of the human-computer interaction interface for analyzing the energy efficiency characteristics of the system using the energy efficiency model of the EC-WPT system in a metal environment.

DETAILED DESCRIPTION

The following embodiments of the technical solution of the present invention are described in detail in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and are therefore only used as examples, and cannot be used to limit the protection scope of the present invention.

It should be noted that, unless otherwise specified, the technical terms or scientific terms used in this application should have the common meanings understood by those skilled in the art to which the present invention belongs.

As shown in FIG1 , this embodiment provides a method for constructing a neural network model for analyzing the influence of the metal environment around the EC-WPT system, including the following steps:

S1: Determine the circuit structure of the EC-WPT system and the surrounding metal environment variables;

In specific implementation, different compensation networks are often used in EC-WPT systems according to different application scenarios, such as LC, LCL, LCLC, LCC, etc., and the LC compensation circuit has the advantages of low circuit order, insensitivity to system parameter changes, and simple circuit topology compared to other compensation circuits. Therefore, this embodiment is based on the bilateral LC compensation EC-WPT system to verify the energy efficiency characteristics of the system under different metal environments mentioned in the present invention. The determined EC-WPT system circuit structure is shown in FIG2, and the system parameter configuration is shown in Table 1:

Table 1 System parameters

Since the circular plate can avoid the tip discharge problem caused by the square plate, this embodiment adopts the circular plate coupling mechanism shown in Figure 3. Of course, it can also be modeled based on plate structures of other shapes. For the surrounding metal environment variables, in actual application scenarios, the metal objects in different application scenarios such as mobile phones, electric vehicles, ships, UUVs and stereo garages are quite different in shape, material and volume. In the experiment, the difficulty and cost of reproducing these actual application scenarios are high. In order to improve the feasibility of the experiment, we simplify the shapes of various metal objects in the actual scene, and abstract the metal objects around the transmitting end and the receiving end of the system coupling mechanism into regular geometric bodies. By changing the variables such as the side length, material and vertical distance between the geometric body and the system coupling mechanism, different types of metal environments around the system are approximately simulated. In order to more comprehensively simulate various types of metal environments around the system in actual scenes, this embodiment sets four variables of distance, area, material and position to describe different types of metal environments. FIG4 is a schematic diagram of the metal environment distance, area and position variables in this example, wherein the distance of the metal environment mainly refers to the vertical distance (d _me ) between the electrode plate P ₆ (same electrode plate P ₅ ) and the coupling mechanism in FIG4 (a); the area mainly refers to the rectangular area (S _me ) directly opposite the electrode plate P ₅ (same electrode plate P ₆ ) and the coupling mechanism in FIG4 (b); and the three values of the position correspond to the three placement conditions of the electrode plates shown in FIG4 (c).

S2: The output power of the current EC-WPT system circuit structure under the influence of different surrounding metal environments is obtained through experiments and used as a sample data set. During the specific implementation, the output power of the current EC-WPT system circuit structure under the influence of different surrounding metal environments needs to be obtained through physical experiments.

Figure 5 shows the regular curve of the influence of different types of metal environments around the EC-WPT system on the system output power, which is drawn based on experimental data. From the distance-output power curve in Figure 5, it can be seen that there is an obvious nonlinear relationship between the system output power and the distance from the metal environment. The closer the distance between the metal environment and the system, the greater the decrease in the system output power. When the area of the metal environment is ^0.49m2 , the maximum value of the system output power decrease is about 56.44W (about 17% decrease compared with the original output power), and the stable value is 292.41W. The stable values of the three curves in the figure are 299.29W (area is ^0.09m2 ), 295.84W (area is ^0.25m2 ) and 292.41W (area is ^0.49m2 ). Although there is a difference between the three stable values, the difference is very small and can be ignored compared with the original output power, indicating that there is no obvious correlation between the stable value of the system output power and the area of the metal environment. However, it should be noted that the stable values of the three curves are all lower than the original output power (332W), and there is a large difference between them and the original output power.

As shown in the area-output power curve in Figure 5, there is an approximate linear relationship between the metal environment area and the system output power, and the correlation between the system output power and the metal environment area is also related to the distance of the metal environment. When the distance is equal to 10cm, the metal environment area and the system output power are negatively correlated, and the system output power will decrease as the metal environment area increases, and the maximum decrease reaches 25.75W (about 7.75% lower than the original output power); when the distance is greater than 20cm, the metal environment area and the system output power are positively correlated, and the system output power will increase as the metal environment area increases. When the distance is 50cm and the area is 0.49m2, the system output power reaches 342.25W, which is a small increase compared to the original output power (332W). From the above analysis, we can know that there is a dividing point between 10cm and 20cm. On both sides of the dividing point, there is a difference in the mathematical relationship between the system output power and the metal environment area. When the distance between the metal environment and the system is less than this dividing point, it will have a negative impact on the system output power. When the distance is greater than this dividing point, the presence of the metal environment will have a certain effect on improving the system output power.

It can be seen from the material-output power curve in Figure 5 that there are differences among the four curves, among which the maximum difference is 13.44W, which is about 4% of the original output power. The proportion is small, indicating that there is no very obvious correlation between the system output power and the metal environment material. In practical applications, the selection of metal environment materials can be more flexible.

As shown in the position-output power curve in Figure 5, the three positions of the metal environment will have different degrees of impact on the system output power, and the stable values of the system output power corresponding to different positions are also different. When the metal environment is at the transmitting end, it has the greatest impact on the system output power, and the maximum value of the system output power decrease reaches 56.44W (about 17% decrease compared to the original output power), and the stable value is 292.41W; when the metal environment is at the receiving end, the system output power does not decrease significantly compared to the original output power (332W), and the stable values of the three curves in the figure are 324.00W (transmitter and receiver), 331.41W (receiver) and 292.41W (transmitter), respectively. The difference between the stable values of the transmitting end and the receiving end reaches 31.59W, indicating that the stable value of the system output power has a strong correlation with the position of the metal environment.

S3: Processing the sample data set, encoding multiple surrounding metal environment variables to form an input feature vector, and taking the output power as the output variable;

For this example, since there is an obvious nonlinear relationship between the metal environment distance, area and position and the EC-WPT system output power, the mathematical relationship between the system output power and the three characteristics of the metal environment distance, area and position can be described by an implicit function as P _out =f(d _me , a _me , l _me );

Among them, P _out represents the output power of the system; d _me represents the distance of the metal environment; a _me represents the area of the metal environment; and l _me represents the location of the metal environment. Since the location of the sample in the data set is a categorical feature (the value of the feature is a string), it cannot be directly used as the input of the neural network model, so it needs to be converted into a numerical variable. In machine learning, one-hot encoding is a common method for processing categorical features, which can digitize the categorical features of samples. For each unique value in the original categorical column, one-hot encoding creates a new feature and fills these virtual variables with 0 and 1. After one-hot encoding, the location feature of the metal environment will change from a one-dimensional feature to a three-dimensional feature as shown in Table 2.

Table 2 One-hot encoding table of position features

S4: construct a neural network model and train it using the input feature vector and output variable processed in step S3;

According to the results of data preprocessing, the number of input layer neurons and output layer neurons of the current neural network model can be determined as 5 (the input features are the three-dimensional one-hot encoding vectors of the area, distance and position of the metal environment) and 1 (the output is the output power of the system). Therefore, it can be concluded that the hyperparameters to be determined in the current neural network model are learning rate, activation function, batch size, number of neurons and number of hidden layers.

S5: Optimize the hyperparameters of the neural network model through grid search method;

In this embodiment, the activation function adopts the Sigmoid function. According to the prior art, this example performs a grid search according to the hyperparameter search range shown in Table 3, and the search results are shown in Table 4.

Table 3 Hyperparameter grid search value range

Table 4 Relative error of the neural network model corresponding to the grid search hyperparameter group

Table 4 shows the relative error values of the neural network model corresponding to each set of hyperparameters in the grid search after the same number of iterations, where HLN represents the number of hidden layers, BS represents the batch size, NN represents the number of neurons in the hidden layer, and RE (Relative Error) represents the relative error between the model prediction value and the true value. As can be seen from Table 4, when the number of hidden layers is 2, the batch size is 4, and the number of neurons in the hidden layer is 128, the performance of the neural network model is the best, and the relative error between the model prediction value and the true value (the relative error between the model prediction value and the corresponding experimental data) is 1.04%, indicating that the choice of the number of hidden layers of the neural network is not the more the better. When the number of hidden layers increases, the complexity of the model is also increasing, and overfitting is prone to occur, resulting in an increase in the error of the model.

S6: Based on the weight and bias matrix of the neural network, the optimal energy efficiency model is derived as the neural network model for analyzing the impact of the metal environment around the EC-WPT system.

When the circuit structure of the EC-WPT system is bilateral LC compensation, and the surrounding metal environment variables are selected as distance, area and position, the number of hidden layers of the optimal energy efficiency model obtained in step S6 is 2, the number of hidden layer neurons is 128, and the batch size is 4; Figure 6 shows the curve of the relative error between the predicted value of the neural network model trained with the optimal hyperparameter group and the corresponding experimental data as the number of iterations changes. As shown in Figure 6, the accuracy of the model after training reaches 98.96%, indicating that the model can accurately estimate the output power of the EC-WPT system under different types of metal environments.

In order to further verify the effectiveness and accuracy of the neural network energy efficiency model of the EC-WPT system under the metal environment, Figure 7 plots the regular curve of the influence of the metal environment distance, area and position on the output power of the EC-WPT system drawn according to the prediction data of the neural network energy efficiency model. As shown in Figure 7, the trend of the regular curve drawn by the system neural network energy efficiency model is consistent with the trend of the experimental regular curve and matches the experimental regular curve with a high accuracy in the curve stability value, curve slope and change threshold point, indicating that the system neural network energy efficiency model can fit the nonlinear functional relationship between different characteristics of the metal environment and the system output power with a high accuracy.

Table 5 shows the relative errors between the predicted values of the system neural network energy efficiency model and the experimental measured values in 8 randomly selected groups of different types of metal environments. It can be seen from Table 5 that for the output power of the EC-WPT system in different types of metal environments, the maximum relative error between the output value of the neural network model and the experimental measured value is 0.98584%, and the minimum relative error is 0.13851%, and the error is less than 1%. The above conclusions show that the system's neural network energy efficiency model can learn the nonlinear functional relationship between the distance, area and position characteristics of the metal environment and the system output power, and evaluate the actual output power of the EC-WPT system in the metal environment with high accuracy, and has the ability to accurately analyze the influence of different types of metal environments on the output power of the EC-WPT system.

Table 5 Comparison between ANN model output and measured values for energy efficiency analysis of EC-WPT system under metal environment

Through the above steps, we can get the structure diagram of the neural network model for energy efficiency analysis of EC-WPT system under metal environment shown in Figure 8. As can be seen from Figure 8, the input vector of hidden layer 1 is represents the input vector of the neural network model, d _me represents the distance of the metal environment, a _me represents the area of the metal environment, One-hot encoding vector representing the location of the metal environment, and They represent the connection weight matrix and bias of the input layer and hidden layer 1 of the neural network model respectively. The output vector of hidden layer 1 is

Input vector of hidden layer 2 and the output vector The calculation formula is: in, and Respectively represent the connection weight matrix and bias of hidden layer 1 and hidden layer 2 of the neural network model;

The system output power obtained at the output layer Where l _me represents the location of the metal environment, and They represent the connection weight matrix and bias between the hidden layer 2 and the output layer of the neural network model respectively.

Since the neural network model has been trained, the parameter values of all the connection weight matrices W and bias b of the network have been determined, thereby obtaining the optimal energy efficiency model.

After obtaining the optimal energy efficiency model, the present invention can also load the optimal energy efficiency model obtained in step S6 onto a host computer, a Raspberry Pi, a PC terminal or a smart terminal to form an EC-WPT system surrounding metal environment impact analysis system with human-computer interaction function. By inputting the limit range of the metal environment characteristic parameter values and the metal environment parameters to be evaluated, the optimal energy efficiency model is called to perform online real-time calculation and analysis of the actual output power of the system, as shown in Figure 9.

By inputting the restricted range of the metal environment characteristic parameter values in the current application scenario and the metal environment parameters that need to be evaluated, the energy efficiency model of the EC-WPT system under the metal environment is called to perform online real-time calculation of the actual output power of the system. Based on the output power data, the influence curve between different characteristics of the metal environment and the system output power can be drawn. For the regular curve between the metal environment distance and the system output power and the regular curve between the metal environment position and the system output power, the horizontal and vertical coordinate data required for drawing can be calculated according to the following formula:

d _me =[d ₁ ,d ₁ +Δ,...,d ₂ ]

P _out =[f(d ₁ ,a _me ,l _me ),f(d ₁ +Δ,a _me ,l _me ),...,f(d ₂ ,a _me ,l _me )]

std _me ∈[d ₁ ,d ₂ ]

a _me ∈[a ₁ ,a ₂ ]

l _me ∈[(1,0,0) ^T ,(0,1,0) ^T ,(0,0,1) ^T ]

For the regular curve between the metal environment area and the system output power, the horizontal and vertical coordinate data required for drawing can be calculated according to the following formula:

a _me =[a ₁ ,a ₁ +Δ,...,a ₂ ]

P _out =[f(d _me ,a ₁ ,l _me ),f(d _me ,a ₁ +Δ,l _me ),...,f(d _me ,a ₂ ,l _me )]

std _me ∈[d ₁ ,d ₂ ]

a _me ∈[a ₁ ,a ₂ ]

l _me ∈[(1,0,0) ^T ,(0,1,0) ^T ,(0,0,1) ^T ]

Among them, d _me represents the data set of the metal environment distance; P _out represents the data set of the system output power; Δ represents the distance division accuracy. The smaller Δ is, the higher the accuracy of the regular curve is, and the more comprehensive it can reflect the influence of the metal environment distance on the system output power; a _me represents the area of the metal environment; l _me represents the position of the metal environment; in the constraint conditions, the d _me interval represents the value range of the distance between the metal environment and the system in the actual scene, d ₁ represents the minimum value of the distance between the metal environment and the system in the actual scene, and d ₂ represents the maximum value; the a _me interval represents the value range of the metal environment area in the actual scene, a ₁ represents the minimum value of the metal environment area in the actual scene, and a ₂ represents the maximum value; the l _me interval represents the value range of the metal environment position in the actual scene, (1,0,0) ^T represents that the metal environment is located at the transmitting end, (0,1,0) ^T represents that the metal environment is located at the receiving end, and (0,0,1) ^T represents that the metal environment is located at both the transmitting end and the receiving end.

In summary, the present invention proposes a method for constructing a neural network model for analyzing the impact of the metal environment around the EC-WPT system. The constructed neural network energy efficiency model can learn the nonlinear functional relationship between the distance, area and position characteristics of the metal environment and the system output power, and evaluate the actual output power of the EC-WPT system in the metal environment with high accuracy, and has the ability to accurately analyze the impact of different types of metal environments on the output power of the EC-WPT system.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some or all of the technical features can be replaced by equivalents. These modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention, and such changes should be included in the scope of the claims and description of the present invention.

Claims (7)

1. A neural network model construction method for analyzing influence of surrounding metal environment of an EC-WPT system is characterized by comprising the following steps:

s1: determining the circuit structure of the EC-WPT system and the environmental variables of surrounding metals;

s2: obtaining output power of a current EC-WPT system circuit structure under the influence of different surrounding metal environments through experiments, and taking the output power as a sample data set;

s3: processing the sample data set, coding a plurality of surrounding metal environment variables to form an input characteristic vector, and taking output power as an output variable;

S4: constructing a neural network model, and training by utilizing the input characteristic vector and the output variable processed in the step S3;

s5: optimizing the super parameters of the neural network model by a grid searching method;

S6: deducing an optimal energy efficiency model according to the weight and the bias matrix of the neural network to be used as a neural network model for analyzing the influence of the surrounding metal environment of the EC-WPT system;

when the circuit structure of the EC-WPT system is bilateral LC compensation and the distance, the area and the position of the surrounding metal environment variable are selected, the hidden layer number of the optimal energy efficiency model obtained in the step S6 is 2, the number of hidden layer neurons is 128, and the batch size is 4;

Input vector of hidden layer 1 Representing the input vector of the neural network model, d _me representing the distance of the metal environment, a _me representing the area of the metal environment,One-hot encoding vectors representing the location of the metal environment,AndRespectively representing the connection weight matrix and bias of the neural network model input layer and the hidden layer 1, wherein the output vector of the hidden layer 1 is

Hidden layer 2 input vectorAnd output vectorThe calculation formula is as follows: Wherein, AndRespectively representing a connection weight matrix and bias of the hidden layer 1 and the hidden layer 2 of the neural network model;

Output layer derived system output power Where l _me denotes the location of the metal environment,AndThe connection weight matrix and the bias of the hidden layer 2 and the output layer of the neural network model are respectively represented.

2. The neural network model construction method for analyzing the environmental influence of metal surrounding an EC-WPT system according to claim 1, wherein the method comprises the steps of: and S2, obtaining the output power of the current EC-WPT system circuit structure under the influence of different surrounding metal environments through a physical experiment.

3. The neural network model construction method for analyzing the environmental influence of surrounding metal of an EC-WPT system according to claim 1 or 2, characterized by: and in the step S3, the single thermal code is adopted to encode the variable of the characterization position in the sample data set.

4. The neural network model construction method for analyzing the environmental influence of metal surrounding an EC-WPT system according to claim 3, wherein: and step S4, the number of neurons of an input layer of the neural network model constructed in the step S4 is 5, the number of neurons of an output layer is 1, and the activating function adopts a Sigmoid function.

5. The neural network model construction method for analyzing the environmental influence of metal surrounding an EC-WPT system according to claim 4, characterized by: in step S5, when the super parameters of the neural network model are searched through the grid, the search range of the hidden layer number is (2, 3, 4), the search range of the hidden layer neuron number is (16,32,64,128), and the search range of the batch size is (4, 8,16, 32).

6. The neural network model construction method for analyzing the environmental influence of surrounding metal of an EC-WPT system according to claim 1 or 2, characterized by: and (3) loading the optimal energy efficiency model obtained in the step (S6) onto an upper computer, a raspberry pie, a PC terminal or an intelligent terminal to form an EC-WPT system surrounding metal environment influence analysis system with a man-machine interaction function, and calling the optimal energy efficiency model to perform online real-time calculation and analysis on the actual output power of the system by inputting a limiting range of the value of the metal environment characteristic parameter and the metal environment parameter to be evaluated.

7. The neural network model construction method for analyzing the environmental influence of metal surrounding an EC-WPT system according to claim 6, characterized by: and (3) drawing an influence rule curve between different values of the surrounding metal environment variable and the system output power by utilizing the output power data calculated by the optimal energy efficiency model obtained in the step (S6).