Prediction method for wireless power transmission electromagnetic environment based on PSO-RBF neural network

Abstract

The invention discloses a wireless power transmission electromagnetic environment prediction method based on a PSO-RBF neural network, and the method comprises the steps: arranging a plurality of sensors in a target region of a wireless power transmission system, collecting space coordinate data (xi, yi, zi) and electromagnetic field intensity data Bi, and carrying out the preprocessing of the collected data, including data cleaning and normalization; constructing a radial basis function neural network model for predicting the electromagnetic field intensity by taking the space coordinates as input and the electromagnetic field intensity as output; the method comprises the following steps: constructing a PSO-RBF neural network wireless power transmission electromagnetic environment prediction model, optimizing RBF neural network parameters by using a PSO algorithm, training the PSO-optimized RBF neural network, and performing multiple iterations and cross validation to obtain a trained RBF neural network model to predict the electromagnetic field intensity; in the to-be-measured area, the electromagnetic field intensity of the corresponding position is predicted according to the space coordinates of the to-be-measured position. According to the invention, the electromagnetic environment of the wireless power transmission system can be monitored and evaluated in real time, and potential electromagnetic environment risks can be found in time.

Description

Technical Field

[0001] The present invention relates to wireless power transmission technology, and in particular to a method for predicting the electromagnetic environment of wireless power transmission based on particle swarm optimization (PSO) and radial basis function (RBF) neural network. Background Art

[0002] With the rapid development of wireless power transfer (WPT) technology in many fields, its application in scenarios such as wireless charging of electric vehicles, charging of mobile communication devices, and power supply of industrial automation equipment has become more and more extensive. WPT technology transmits energy in space through electromagnetic fields, overcoming the limitations of traditional cable transmission and solving electrical safety problems such as electric shock, fire and explosion. However, the electromagnetic field generated during wireless power transmission will have a certain impact on the surrounding environment, especially potential risks to human health, such as effects on the nervous system and changes in electrochemical balance in organisms. Existing wireless power transmission technology has made certain progress in improving transmission efficiency and controllability, but there are still technical bottlenecks in the safety assessment and control of the electromagnetic environment.

[0003] At present, the mainstream electromagnetic environment prediction methods are mainly based on electromagnetic wave propagation models. These methods rely on electromagnetic wave propagation theory and predict the intensity and distribution of the electromagnetic field by analyzing the propagation path, reflection, refraction, attenuation and other characteristics of electromagnetic waves in the medium. Commonly used models include ray tracing method, finite element method (FEM), etc. These methods can provide accurate electromagnetic field distribution information in complex geometric structures and material environments, and are suitable for simulation calculations in static or quasi-static environments.

[0004] However, methods based on electromagnetic wave propagation models have problems such as high computational complexity, poor adaptability and strong data dependence. These methods usually involve complex mathematical calculations, especially when the environment geometry is complex, the amount of calculation is very large, requiring a lot of computing resources and time, and it is difficult to meet real-time requirements. Summary of the invention

[0005] The purpose of the present invention is to propose a method for predicting the electromagnetic environment of wireless power transmission based on a PSO-RBF neural network to solve the problems of high computational complexity, poor real-time performance, weak adaptability and strong data dependence in the electromagnetic environment prediction of the prior art.

[0006] The technical solution to achieve the purpose of the present invention is: a method for predicting the electromagnetic environment of wireless power transmission based on a PSO-RBF neural network, comprising the following steps:

[0007] Step 1: Data collection and preprocessing

[0008] Multiple sensors are arranged in the target area of ​​the wireless power transmission system to collect spatial coordinate data (x i ,y i ,z i ) and electromagnetic field strength data B i , and preprocess the collected data, including data cleaning and normalization;

[0009] Step 2: Construction of RBF neural network model

[0010] With spatial coordinates as input and electromagnetic field intensity as output, a radial basis function neural network model for electromagnetic field intensity prediction is constructed;

[0011] Step 3: Construct a PSO-RBF neural network wireless power transmission electromagnetic environment prediction model

[0012] Construct a PSO-RBF neural network radio power transmission electromagnetic environment prediction model, use the PSO algorithm to optimize the RBF neural network parameters, train the RBF neural network after PSO optimization, and obtain a trained RBF neural network model through multiple iterations and cross-validation to predict the electromagnetic field strength;

[0013] Step 4: Electromagnetic field prediction

[0014] In the area to be measured, the electromagnetic field strength at the corresponding position is predicted according to the spatial coordinates of the position to be measured.

[0015] Further, step 1, data collection and preprocessing, the specific method is:

[0016] 1.1. Arrangement of measuring points:

[0017] In the working area of ​​the wireless power transmission system, multiple measuring points are arranged at predetermined spatial intervals, and sensors are installed at each measuring point for real-time monitoring of electromagnetic field intensity data and spatial coordinate data;

[0018] 1.2. Data collection:

[0019] Real-time record of the electromagnetic field strength value B at each measuring point i i and its corresponding spatial coordinates (x i ,y i ,z i ), forming a unified representation of spatial coordinate data and electromagnetic field intensity:

[0020] d i =(x i1 ,y i1 ,z i1 ,B i1 ,x i2 ,yi2 ,z i2 ,B i2 ,...,x im ,y im ,z im ,B im );

[0021] 1.3. Data cleaning and normalization:

[0022] a. Use the triple standard deviation method to eliminate outliers, clean the data collected by sensor i, remove outliers and noise, and ensure the validity and integrity of the data;

[0023] i. Calculate the mean of the electromagnetic field strength data collected by all sensors and standard deviation σ;

[0024] ii. Remove more than or less than data points to ensure data quality;

[0025] b. Normalize the electromagnetic field strength data after cleaning to the interval [0,1]. The normalization formula is as follows:

[0026]

[0027] Among them, d ij ' is the normalized electromagnetic field strength data, the subscript represents the electromagnetic field strength data collected by the sensor in the i-th row and j-th column in the sensor array, d j,min is the minimum value of the electromagnetic field strength in the jth column after data cleaning, d j,max is the maximum value of the electromagnetic field intensity in the jth column after data cleaning;

[0028] c. Get the normalized data matrix D

[0029]

[0030] Wherein, m represents the total number of sensors in a row of the sensor array, and n represents the total number of sensors in a column of the sensor array.

[0031] Furthermore, for the application of wireless charging scenarios for electric vehicles, the sensor layout positions are: ground sensor array, sensor array above the charging plate, and sensor array around the charging device, among which:

[0032] a. Ground sensor array

[0033] The sensor array is in the shape of a rectangular array, and the layout is as follows:

[0034] The sensors are arranged on the ground in the form of a rectangular grid in the parking area, with the long and short sides of the grid parallel to the long and wide sides of the parking space, respectively, to ensure that the sensors are evenly distributed throughout the parking area, achieving full coverage and high-resolution monitoring of the electromagnetic field;

[0035] The distance between each sensor is 50-100 cm. The specific distance depends on the size of the charging area and the accuracy required for monitoring. The smaller the distance, the higher the monitoring accuracy, but the number of sensors required and the cost will also increase.

[0036] b. Sensor array above the charging plate

[0037] The sensor array is in the shape of a rectangular array, and the layout is as follows:

[0038] The sensor array is suspended above the charging plate in a rectangular array to improve the spatial coverage of the electromagnetic field. In the rectangular array, the spacing between sensors is 20-30 cm. The long and short sides of the sensor array are parallel to the long and wide sides of the charging plate, respectively. Each sensor is arranged parallel to the edge of the charging plate to form a neat rectangular grid.

[0039] c. Sensor array around charging equipment

[0040] The sensor array is in the shape of a ring array, and the layout is as follows:

[0041] A circular sensor array is arranged around the charging device to form a concentric circle or elliptical structure to monitor the electromagnetic field strength around the charging device. The radius of the circular array is determined according to the size of the charging device and the sensor detection range. The radius is 30-50 cm. 2-3 circles of sensors are arranged in the ring, with the inner circle close to the device and the outer circle further extended to ensure all-round monitoring around the device.

[0042] Further, step 2, the construction of the RBF neural network model, the specific method is:

[0043] 2.1. Constructing RBF neural network model structure

[0044] The structure includes input layer, hidden layer and output layer;

[0045] 2.2. Design of hidden layer structure

[0046] The hidden layer consists of multiple radial basis function neurons, each neuron uses a Gaussian function as the activation function, and its expression is:

[0047]

[0048] Among them, c i is the center position of the ith neuron, σ is the expansion parameter, r = ||vc i|| is the Euclidean distance between the input vector and the neuron center, and v is the spatial coordinate data collected by sensor i;

[0049] 2.3. Model output calculation

[0050] Predicted values ​​of electromagnetic field strength The weighted sum of all neurons is calculated as:

[0051]

[0052] Among them, w i is the output layer weight parameter, N is the number of RBF neurons, and (x, y, z) is the spatial coordinates of the electromagnetic field strength that needs to be predicted.

[0053] Further, in step 3, a PSO-RBF neural network wireless power transmission electromagnetic environment prediction model is constructed. The specific method is as follows:

[0054] 3.1.RBF neural network model training

[0055] The RBF neural network model training process includes: determining the center c of the RBF neuron i and width σ i , calculate the weight w of the output layer i To minimize the prediction error, as follows:

[0056] a. Determine the center c i

[0057] RBF neuron center c i Determination: Using the K-means clustering algorithm, the spatial coordinate data (x, y, z) collected by all sensors are divided into M clusters, and the center of each cluster is used as the center c of an RBF neuron i , the algorithm is as follows:

[0058] Randomly select M spatial coordinate data points as the initial center;

[0059] Calculate the Euclidean distance from each spatial coordinate data point to all centers, and assign the corresponding spatial coordinate data point to the cluster where the nearest center is located according to the size of the Euclidean distance;

[0060] Recalculate the center of each cluster, that is, update where n i is the number of spatial coordinate data points in the i-th cluster, l is the spatial coordinate data selected as the initial center;

[0061] Repeat the above process until all centers no longer change significantly or the maximum number of iterations is reached;

[0062] b. RBF neuron width σ i Calculation

[0063] Calculate the width σ of each RBF neuron i , σ i Set to the average distance between two adjacent centers, or calculated according to formula (13):

[0064]

[0065] Among them, ||c i -c j || represents the Euclidean distance between the i-th and j-th centers;

[0066] c. Calculate the output layer weight w i

[0067] Calculate the spatial coordinate data of all sensor inputs through the response of RBF neurons to form a matrix Φ, whose elements φ ij Represents the output of the i-th spatial coordinate data point through the j-th RBF neuron:

[0068]

[0069] Among them, v i is the spatial coordinate of the i-th input;

[0070] The least squares method is used to calculate the weight w i :Solve the weight w by minimizing the objective function error E i , the error is defined as:

[0071] E=‖Y-ΦW‖ 2 (15)

[0072] Among them, Y is the actual output target vector, that is, the electromagnetic field strength, Φ is the matrix composed of the response values ​​of all spatial coordinate data, and W is the weight vector to be solved. The least squares method is used to solve: W = (Φ T Φ) -1 Φ T Y;

[0073] 3.2. Optimizing the parameters of RBF neural network using particle swarm optimization algorithm

[0074] a. Initialize particle population: Initialize a number of particles in the parameter space, each particle represents a set of parameters of the RBF neural network (c i ,σ i ,w i ), and set the initial speed and position at the same time;

[0075] b. Define the fitness function: Use mean square error (MSE) as the fitness function, and the goal is to minimize MSE:

[0076]

[0077] c. Particle velocity and position update: According to the PSO algorithm rules, iteratively update the particle velocity v i (t) and position x i (t):

[0078] v i (t+1)=w·v i (t)+c 1 ·r 1 ·(p best -x i (t))+c 2 ·r 2 ·(g best -x i (t))(11)

[0079] x i (t+1) = x i (t) + v i (t+1) (12)

[0080] Among them, the parameters in formula (11), i.e. the speed update formula, are explained as follows:

[0081] v i (t+1): The velocity of particle i at time t+1. The velocity determines the moving direction and step size of the particle in the search space.

[0082] w·v i (t): Inertia term, which indicates the influence of the particle's velocity at the previous moment on the current velocity. w is the inertia weight, which is used to adjust the particle's dependence on history. A larger w value is conducive to global search, while a smaller w value is conducive to local search.

[0083] c 1 ·r 1 ·(p best -x i (t)): cognitive term, indicating that the particle has its own historical optimal position p best to update the speed. c1 is the learning factor, which is used to adjust the degree to which the particle learns from its own historical optimal position. r1 is a random number between [0,1], which increases the randomness of the search.

[0084] c 2 ·r 2 ·(g best -x i(t)):Social term, which means the particle has the best historical position g in the group best to update the speed. c2 is the learning factor, which is used to adjust the degree to which the particle learns from the historical optimal position of the group. r2 is also a random number between [0,1].

[0085] Formula (12), i.e. the parameters of the position update formula, are explained as follows:

[0086] x i (t+1): The position of particle i at time t+1. The position represents a potential solution for the particle in the search space.

[0087] x i (t): The position of particle i at time t.

[0088] v i (t+1): The velocity of particle i at time t+1. This velocity is calculated by the velocity update formula and determines how the particle moves to the new position.

[0089] d. Convergence judgment: Repeat the update process until the convergence conditions are met and the optimal RBF neural network parameter set is obtained;

[0090] 3.3. Training and verification of RBF neural network model parameters

[0091] In order to verify the generalization ability of the model, the k-fold cross validation method is used, as follows:

[0092] a. Divide the training data set into k subsets, use k-1 subsets to train the RBF neural network model each time, and use the remaining subset for verification, and repeat this process k times;

[0093] b. Calculate the mean square error MSE for each validation: Among them, N i is the number of data points for the i-th validation, B j is the actual value, is the predicted value;

[0094] c. The validation error of the final model is the average of all k validation errors:

[0095] d. If E cv If the value exceeds the set threshold, the algorithm returns to step 3.2 to optimize the parameters and repeat the training and verification process until the expected accuracy is achieved.

[0096] Further, step 4, electromagnetic field prediction and result output, the specific method is:

[0097] Given the spatial coordinates (x, y, z) of the position to be measured, the electromagnetic field strength at the corresponding position can be calculated:

[0098]

[0099] Where N is the number of RBF neurons, w i 、c i and σ i are the parameters after training.

[0100] Furthermore, it also includes:

[0101] According to ICNIRP or IEEE C95.1 standards, the prediction results are classified as follows:

[0102] i. Low risk level: The predicted value is less than 10% of the international standard limit, indicating that the exposure risk is extremely low and no additional protective measures are required;

[0103] ii. Medium risk level: The predicted value is between 10% and 50% of the international standard limit, indicating that there is a certain risk and it is recommended to monitor the exposure time or use protective measures;

[0104] iii. High risk level: The predicted value is close to or exceeds the international standard limit, indicating a high risk and requiring immediate protective measures or reduction of exposure.

[0105] A wireless power transmission electromagnetic environment prediction system based on PSO-RBF neural network implements the wireless power transmission electromagnetic environment prediction method based on PSO-RBF neural network to achieve the wireless power transmission electromagnetic environment prediction based on PSO-RBF neural network.

[0106] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the method for predicting the electromagnetic environment of wireless power transmission based on a PSO-RBF neural network is implemented to realize the prediction of the electromagnetic environment of wireless power transmission based on a PSO-RBF neural network.

[0107] A computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the method for predicting the electromagnetic environment of wireless power transmission based on a PSO-RBF neural network is implemented to realize the prediction of the electromagnetic environment of wireless power transmission based on a PSO-RBF neural network.

[0108] Compared with the prior art, the present invention has the following significant advantages: First, the present invention optimizes the parameters of the radial basis function (RBF) neural network using a particle swarm optimization (PSO) algorithm, thereby significantly improving the accuracy of electromagnetic field intensity prediction. Compared with traditional electromagnetic field prediction methods, this method can more accurately capture the changing characteristics of the electromagnetic field during wireless power transmission, reduce errors, and enhance the reliability of the prediction model.

[0109] Second, the PSO-RBF neural network used in the present invention combines the fast global search capability of the particle swarm optimization algorithm with the nonlinear mapping capability of the RBF neural network, achieving efficient processing of large-scale data. Compared with traditional numerical simulation methods, this method has a faster calculation speed and can meet the needs of real-time electromagnetic environment prediction.

[0110] Third, the method of the present invention can monitor and evaluate the electromagnetic environment of the wireless power transmission system in real time, and timely discover potential electromagnetic environment risks. This method provides important technical support for the application of wireless power transmission technology in complex environments, and helps to improve the safety and stability of the system. It is suitable for electromagnetic environment assessment and optimization control in complex and changeable environments such as wireless charging of electric vehicles and power supply of industrial automation equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0111] Figure 1 Schematic diagram of the process of optimizing the RBF neural network model using the PSO algorithm.

[0112] Figure 2 MSE curve of the RBF neural network model optimized by PSO algorithm. DETAILED DESCRIPTION

[0113] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0114] The present invention provides a method for predicting the electromagnetic environment of wireless power transmission based on a PSO-RBF neural network, which is implemented by the following steps:

[0115] Step 1, data collection and preprocessing:

[0116] In the actual application environment of wireless power transmission system, electromagnetic field strength data is collected through sensor arrays at multiple spatial positions to ensure that the sensors can cover the entire working area. Each sensor is used to collect the following data:

[0117] Spatial coordinate data (x i ,y i ,z i): represents the position of sensor i in three-dimensional space.

[0118] Electromagnetic field strength data B i : Indicates the electromagnetic field strength value at this location.

[0119] The specific steps are as follows:

[0120] 1.1. Arrangement of measuring points:

[0121] In the working area of ​​the wireless power transmission system, multiple measurement points are arranged at predetermined spatial intervals. Each measurement point is equipped with a high-precision sensor for real-time monitoring of the above-mentioned related data. The following is an explanation of the sensor layout and implementation scheme for the electric vehicle wireless charging scenario. For other application scenarios, the number and array of sensors can be adjusted as needed.

[0122] For the application of wireless charging scenarios for electric vehicles, the main locations for the placement of sensors are: on the ground, above the charging plate, and around the charging equipment.

[0123] a. Ground sensor array: The sensor array is in the shape of a rectangular array.

[0124] Layout Instructions:

[0125] The sensors are arranged on the ground in the form of a rectangular grid in the parking area, with the long and short sides of the grid parallel to the long and wide sides of the parking space, respectively. This ensures that the sensors are evenly distributed throughout the parking area, achieving full coverage and high-resolution monitoring of the electromagnetic field.

[0126] The recommended spacing between each sensor is 50-100 cm, depending on the size of the charging area and the accuracy required for monitoring. The smaller the spacing, the higher the monitoring accuracy, but the number of sensors required and the cost will also increase.

[0127] b. Sensor array above the charging plate: The sensor array is in the shape of a rectangular array.

[0128] Layout Instructions:

[0129] The sensor array is suspended above the charging plate in a rectangular array to improve the spatial coverage of the electromagnetic field. In the rectangular array, the spacing between sensors is 20-30 cm, and the long and short sides of the sensor array are parallel to the long and wide sides of the charging plate respectively. Each sensor is arranged parallel to the edge of the charging plate to form a neat rectangular grid.

[0130] c. Sensor arrangement around the charging device: The sensor array is in the shape of a ring array.

[0131] Layout Instructions:

[0132] A circular sensor array is arranged around the charging device (such as a charging pile or a transmitting coil) to form a concentric circle or elliptical structure to monitor the electromagnetic field strength around the charging device.

[0133] The radius of the ring array is recommended to be determined according to the size of the charging device and the sensor detection range, usually with a radius of 30-50 cm. Two or three circles of sensors can be arranged in the ring, with the inner circle close to the device and the outer circle further extended to ensure all-round monitoring around the device.

[0134] 1.2. Data collection:

[0135] Real-time record of the electromagnetic field strength value B at each measuring point i i and its corresponding spatial coordinates (x i ,y i ,z i ). The data collected by each sensor is comprehensive. For example, the data collected by sensor i is the spatial coordinate data (x i ,y i ,z i ) and electromagnetic field strength data B i , the data collected by sensor j is the spatial coordinate data (x j ,y j ,z j ) and electromagnetic field strength data B j The acquisition frequency and time interval can be set according to the dynamic characteristics of wireless power transmission to ensure the adequacy and effectiveness of the data.

[0136] 1.3. Data cleaning and normalization:

[0137] a. Clean the data collected by sensor i to remove outliers and noise to ensure the validity and integrity of the data. Here, the triple standard deviation method (3-σ criterion) is used to remove outliers. It should be noted that the data collected by each sensor is comprehensive. For example, the data collected by sensor i is the spatial coordinate data (x i ,y i ,z i ) and electromagnetic field strength data B i , the data collected by sensor j is the spatial coordinate data (x j ,y j ,z j ) and electromagnetic field strength data B j , and the corresponding sensor data processing methods are also the same, so the following is an explanation of the data processing of sensor i, and the same is true for other sensors. The specific steps are as follows:

[0138] i. Calculate the mean of the electromagnetic field strength data collected by all sensors and standard deviation σ.

[0139] ii. Remove more than or less than of data points to ensure data quality.

[0140] b. Normalize the cleaned data and standardize the data of each feature to the interval [0,1]. The normalization formula is as follows:

[0141]

[0142] Among them, d ij ' is the normalized electromagnetic field strength data, the subscript represents the electromagnetic field strength data collected by the sensor in the i-th row and j-th column in the sensor array, d j,min is the minimum value of the electromagnetic field strength in the jth column after data cleaning, d j,max is the maximum value of the electromagnetic field intensity in the jth column after data cleaning;

[0143] c. Get the normalized data matrix D

[0144]

[0145] Wherein, m represents the total number of sensors in a row of the sensor array, and n represents the total number of sensors in a column of the sensor array.

[0146] Step 2: Construction of RBF neural network model

[0147] Construct a radial basis function (RBF) neural network model for electromagnetic field intensity prediction;

[0148] The specific steps are as follows:

[0149] 2.1. Constructing RBF neural network model structure

[0150] Define the structure and parameters of the input layer, hidden layer, and output layer;

[0151] 2.2. Design of hidden layer structure

[0152] The hidden layer consists of multiple radial basis function (RBF) neurons, each of which uses a Gaussian function as the activation function, which is expressed as:

[0153]

[0154] Among them, c i is the center position of the ith neuron, σ is the expansion parameter, r = ||vc i || is the Euclidean distance between the input vector and the neuron center. Here v is the spatial coordinate data collected by sensor i.

[0155] 2.3. Model output calculation

[0156] Predicted values ​​of electromagnetic field strength The weighted sum of all RBF neurons is calculated as:

[0157]

[0158] Among them, w i is the output layer weight parameter, N is the number of RBF neurons, and (x, y, z) is the spatial coordinates of the electromagnetic field strength that needs to be predicted.

[0159] Step 3: Construct a PSO-RBF neural network wireless power transmission electromagnetic environment prediction model

[0160] The PSO algorithm is used to optimize the RBF neural network parameters to construct a PSO-RBF neural network wireless power transmission electromagnetic environment prediction model (including initializing the particle population; mapping the particle position information; calculating the particle fitness value; updating the individual extreme value of the particle; updating the global extreme value of the particle; updating the particle's own position and speed; recording the global extreme value and ending PSO; and constructing the RBF neural network from the global extreme value).

[0161] The details are as follows:

[0162] 3.1.RBF neural network model training

[0163] The RBF neural network model training process includes: determining the center c of the RBF neuron i and width σ i , calculate the weight w of the output layer i To minimize the prediction error. Specific steps:

[0164] a. Determine the center c i and width σ i

[0165] RBF neuron center c i Determination: Use K-means clustering algorithm to divide the spatial coordinate data (x, y, z) collected by all sensors into M clusters. The center of each cluster is used as the center of an RBF neuron c i The specific algorithm steps are as follows:

[0166] Randomly select M spatial coordinate data points as the initial center.

[0167] Calculate the Euclidean distance from each spatial coordinate data point to all centers, and assign the corresponding spatial coordinate data point to the cluster where the nearest center is located according to the size of the Euclidean distance.

[0168] Recalculate the center of each cluster, that is, update where n i is the number of data points in the i-th cluster, and l is the spatial coordinate data selected as the initial center.

[0169] The above process is repeated until all centers no longer change significantly or the maximum number of iterations is reached.

[0170] b. RBF neuron width σ i Calculation: After determining the center, calculate the width σ of each RBF neuron i Usually, σ i Set to the average distance between two neighboring centers, or calculated according to the following formula:

[0171]

[0172] Among them, ||c i -c j || represents the Euclidean distance between the i-th and j-th centers. This setting method ensures that each RBF neuron has an appropriate response width to the input within its neighborhood.

[0173] c. Calculate the output layer weight w i

[0174] Construct the output matrix of the RBF network: Calculate the spatial coordinate data of all sensor inputs through the response of RBF neurons to form a matrix Φ, whose elements φ ij Represents the output of the i-th data point through the j-th RBF neuron:

[0175]

[0176] Among them, v i is the spatial coordinate of the ith input.

[0177] The least squares method is used to calculate the weight w i :Solve the weight w by minimizing the objective function error E i , the error is defined as:

[0178] E=‖Y-ΦW‖ 2 (15)

[0179] Among them, Y is the actual output target vector, that is, the electromagnetic field strength, Φ is the matrix composed of the response values ​​of all spatial coordinate data, and W is the weight vector to be solved. Using the least squares method, we can get: W = (Φ T Φ) -1 Φ T Y

[0180] 3.2. Use the particle swarm optimization algorithm (PSO) to optimize the parameters of the RBF neural network. The specific implementation steps are as follows:

[0181] a. Initialize particle population: Initialize a number of particles in the parameter space, each particle represents a set of parameters of the RBF neural network (c i ,σ i ,w i ), and set the initial speed and position at the same time.

[0182] b. Define the fitness function: Use mean square error (MSE) as the fitness function, and the goal is to minimize MSE:

[0183]

[0184] c. Particle velocity and position update: According to the PSO algorithm rules, iteratively update the particle velocity v i (t) and position x i (t):

[0185] v i (t+1)=w·v i (t)+c 1 ·r 1 ·(p best -x i (t))+c 2 ·r 2 ·(g best -x i (t))(11)

[0186] x i (t+1) = x i (t) + v i (t+1) (12)

[0187] Among them, the parameters in formula (11), i.e. the speed update formula, are explained as follows:

[0188] v i (t+1): The velocity of particle i at time t+1. The velocity determines the moving direction and step size of the particle in the search space.

[0189] w·v i (t): Inertia term, which indicates the influence of the particle's velocity at the previous moment on the current velocity. w is the inertia weight, which is used to adjust the particle's dependence on history. A larger w value is conducive to global search, while a smaller w value is conducive to local search.

[0190] c 1 ·r 1 ·(p best -x i(t)): cognitive term, indicating that the particle has its own historical optimal position p best to update the speed. c1 is the learning factor, which is used to adjust the degree to which the particle learns from its own historical optimal position. r1 is a random number between [0,1], which increases the randomness of the search.

[0191] c 2 ·r 2 ·(g best -x i (t)):Social term, which means the particle has the best historical position g in the group best to update the speed. c2 is the learning factor, which is used to adjust the degree to which the particle learns from the historical optimal position of the group. r2 is also a random number between [0,1].

[0192] Formula (12), i.e. the parameters of the position update formula, are explained as follows:

[0193] x i (t+1): The position of particle i at time t+1. The position represents a potential solution for the particle in the search space.

[0194] x i (t): The position of particle i at time t.

[0195] v i (t+1): The velocity of particle i at time t+1. This velocity is calculated by the velocity update formula and determines how the particle moves to the new position.

[0196] d. Convergence judgment: Repeat the update process until the convergence condition is met. For example, the maximum number of iterations or the change in the fitness function value is less than the set threshold, and the optimal RBF neural network parameter set is obtained.

[0197] Model Validation

[0198] In order to verify the generalization ability of the model, the k-fold cross validation method is used. The specific steps are as follows:

[0199] a. Divide the training data set into k subsets, use k-1 subsets to train the RBF neural network model each time, and use the remaining subset for verification. Repeat this process k times.

[0200] b. Calculate the mean square error (MSE) for each validation: Among them, N i is the number of data points for the i-th validation, B j is the actual value, is the predicted value.

[0201] c. The validation error of the final model is the average of all k validation errors:

[0202] d. If E cv If the value exceeds the set threshold, it is necessary to return to 3.2 to optimize the model parameters and repeat the training and verification process until the model reaches the expected accuracy requirement.

[0203] Step 4: Electromagnetic field prediction and result output

[0204] a. Electromagnetic field prediction

[0205] Once the RBF neural network model is trained, the optimized RBF neural network model is used to predict the electromagnetic field strength. In the area to be tested, the spatial coordinates (x, y, z) of the location to be tested are known and put into the model to calculate the electromagnetic field strength at the corresponding location:

[0206]

[0207] Where N is the number of RBF neurons, w i 、c i and σ i are the parameters after training.

[0208] b. Result output and analysis

[0209] According to the International Commission on Non-Ionizing Radiation Protection (ICNIRP) or IEEE C95.1 standard, the prediction results are classified as follows:

[0210] i. Low risk level (green zone): The predicted value is 10% lower than the international standard limit, indicating that the exposure risk is extremely low and no additional protective measures are required;

[0211] ii. Medium risk level (yellow zone): The predicted value is between 10% and 50% of the international standard limit, indicating a certain risk.

[0212] Risk, it is recommended to monitor exposure time or use protective measures;

[0213] iii. High risk level (red zone): The predicted value is close to or exceeds the international standard limit, indicating a high risk and requiring immediate protective measures or reduction of exposure.

[0214] Example

[0215] In order to verify the effectiveness and reliability of the electromagnetic environment prediction method of wireless power transmission based on PSO-RBF neural network in the present invention, a typical wireless power transmission system was selected, and the electromagnetic environment in its three-dimensional space was simulated by using MAXWELL simulation software. The specific implementation steps are as follows:

[0216] Simulation environment construction: Use MAXWELL electromagnetic field simulation software to build a typical three-dimensional model of wireless power transmission system. The model includes the following main components:

[0217] Transmitter coil: A circular copper coil is used as the transmitter. The coil size is 30 cm in diameter, 2 mm in thickness, and the operating frequency is set to 100 kHz.

[0218] Receiving coil: Corresponding to the transmitting coil, the receiving coil has the same size and material as the transmitting coil. It is placed 50 cm away from the transmitting end and is used to receive wireless energy.

[0219] Space environment setting: An air medium environment is set in the three-dimensional space between the transmitting and receiving coils, while considering the influence of obstacles in the surrounding environment (such as metal plates, walls, etc.) on the propagation of electromagnetic fields.

[0220] Data collection and simulation settings, the specific steps are as follows:

[0221] In the simulation model, multiple sensor points are arranged to collect electromagnetic field strength data at different locations in three-dimensional space. The sensor points are arranged at intervals of 10 cm, and the array is arranged in a rectangular array, with 10 in length and width, that is, a 10×10 matrix, and a total of 100 sensor points are set to ensure coverage of the entire simulation area.

[0222] Data type collected: Each sensor point collects electromagnetic field strength (B i ), spatial coordinate data (x i ,y i ,z i ).

[0223] Simulation parameter settings: Set the simulation step size of the MAXWELL simulation software to 1us and the total simulation time to 100ms to obtain sufficient timing data.

[0224] Simulation run: Run the simulation program to obtain the electromagnetic field intensity data of each sensor point at different time steps, and export the data to the data processing module.

[0225] Data preprocessing: Preprocess the exported raw simulation data: remove outliers and normalize the data.

[0226] The PSO-RBF neural network model of the present invention is used to predict the electromagnetic field strength of the collected simulation data:

[0227] Model training: The processed data is input into the RBF neural network model, the input layer is set to the feature data after dimension reduction, the hidden layer is multiple RBF neurons, and the output layer is used to output the predicted value of the electromagnetic field strength.

[0228] Parameter optimization: The particle swarm optimization (PSO) algorithm is used to optimize the parameters of the RBF neural network (including the neuron center position, width, and output layer weight), and the model parameters are dynamically adjusted with the goal of minimizing the mean square error (MSE).

[0229] Prediction verification: Use the optimized RBF neural network model to predict the electromagnetic field strength of the new simulation data, and compare and analyze the predicted results with the actual results of the MAXWELL simulation software to calculate the average relative error and prediction accuracy.

[0230] Result analysis: According to the training model, the training error is 0.0041191; the test error is 0.0060781. In the attached figure, the MSE curve of the PSO algorithm optimizing the RBF neural network model is given.

[0231] Read the trained RBF-PSO neural network parameters and display the relevant Center, SP, and W values;

[0232]

[0233] Compared with the traditional method based on electromagnetic wave propagation model, the electromagnetic environment prediction model based on PSO-RBF established by the present invention can significantly reduce the calculation time under the same calculation conditions, while improving the real-time and adaptability of the prediction.

[0234] In summary, the present invention combines the global search capability of the PSO algorithm and the nonlinear mapping capability of the RBF neural network, and has high prediction accuracy and real-time performance. Compared with the traditional method based on the electromagnetic wave propagation model, the present method does not rely on complex physical models and precise environmental parameters, and can achieve high-precision prediction by learning historical data and sensor data when the data is relatively incomplete. In addition, by optimizing the parameters of the neural network, the present invention can establish an efficient prediction model to achieve real-time prediction and dynamic adjustment of the electromagnetic field intensity under complex and changeable environments, and provide technical support for the safety and efficiency of the wireless power transmission system. Therefore, the method of the present invention is more adaptable and practical, and is particularly suitable for wireless power transmission scenarios that require rapid response and dynamic adjustment.

[0235] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0236] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for predicting electromagnetic environment of wireless power transmission based on PSO-RBF neural network, characterized in that: The steps include: Step 1: Data collection and preprocessing Multiple sensors are arranged in the target area of ​​the wireless power transmission system to collect spatial coordinate data (x i ,y i ,z i ) and electromagnetic field strength data B i , and preprocess the collected data, including data cleaning and normalization; Step 2: Construction of RBF neural network model With spatial coordinates as input and electromagnetic field intensity as output, a radial basis function neural network model for electromagnetic field intensity prediction is constructed; Step 3: Construct a PSO-RBF neural network wireless power transmission electromagnetic environment prediction model Construct a PSO-RBF neural network radio power transmission electromagnetic environment prediction model, use the PSO algorithm to optimize the RBF neural network parameters, train the RBF neural network after PSO optimization, and obtain a trained RBF neural network model through multiple iterations and cross-validation to predict the electromagnetic field strength; Step 4: Electromagnetic field prediction In the area to be measured, the electromagnetic field strength at the corresponding position is predicted according to the spatial coordinates of the position to be measured.

2. The method for predicting electromagnetic environment of wireless power transmission based on PSO-RBF neural network according to claim 1 is characterized in that: Step 1: Data collection and preprocessing. The specific method is as follows: 1.

1. Arrangement of measuring points: In the working area of ​​the wireless power transmission system, multiple measuring points are arranged at predetermined spatial intervals, and sensors are installed at each measuring point for real-time monitoring of electromagnetic field intensity data and spatial coordinate data; 1.

2. Data collection: Real-time record of the electromagnetic field strength value B at each measuring point i i and its corresponding spatial coordinates (x i ,y i ,z i ), forming a unified representation of spatial coordinate data and electromagnetic field intensity: d i =(x i1 ,y i1 ,z i1 ,B i1 ,x i2 ,y i2 ,z i2 ,B i2 ,...,x im ,y im ,z im ,B im ); 1.

3. Data cleaning and normalization: a. Use the triple standard deviation method to eliminate outliers, clean the data collected by sensor i, remove outliers and noise, and ensure the validity and integrity of the data; i. Calculate the mean of the electromagnetic field strength data collected by all sensors and standard deviation σ; ii. Remove more than or less than data points to ensure data quality; b. Normalize the electromagnetic field strength data after cleaning to the interval [0,1]. The normalization formula is as follows: Among them, d ij ' is the normalized electromagnetic field strength data, the subscript represents the electromagnetic field strength data collected by the sensor in the i-th row and j-th column in the sensor array, d j,min is the minimum value of the electromagnetic field strength in the jth column after data cleaning, d j,max is the maximum value of the electromagnetic field intensity in the jth column after data cleaning; c. Get the normalized data matrix D Wherein, m represents the total number of sensors in a row of the sensor array, and n represents the total number of sensors in a column of the sensor array.

3. The method for predicting electromagnetic environment of wireless power transmission based on PSO-RBF neural network according to claim 1 is characterized in that: For the application of wireless charging scenarios for electric vehicles, the sensor layout locations are: ground sensor array, sensor array above the charging plate, and sensor array around the charging equipment, among which: a. Ground sensor array The sensor array is in the shape of a rectangular array, and the layout is as follows: The sensors are arranged on the ground in the form of a rectangular grid in the parking area, with the long and short sides of the grid parallel to the long and wide sides of the parking space, respectively, to ensure that the sensors are evenly distributed throughout the parking area, achieving full coverage and high-resolution monitoring of the electromagnetic field; The distance between each sensor is 50-100 cm. The specific distance depends on the size of the charging area and the accuracy required for monitoring. The smaller the distance, the higher the monitoring accuracy, but the number of sensors required and the cost will also increase. b. Sensor array above the charging plate The sensor array is in the shape of a rectangular array, and the layout is as follows: The sensor array is suspended above the charging plate in a rectangular array to improve the spatial coverage of the electromagnetic field. In the rectangular array, the spacing between sensors is 20-30 cm. The long and short sides of the sensor array are parallel to the long and wide sides of the charging plate, respectively. Each sensor is arranged parallel to the edge of the charging plate to form a neat rectangular grid. c. Sensor array around charging equipment The sensor array is in the shape of a ring array, and the layout is as follows: A circular sensor array is arranged around the charging device to form a concentric circle or elliptical structure to monitor the electromagnetic field strength around the charging device. The radius of the circular array is determined according to the size of the charging device and the sensor detection range. The radius is 30-50 cm. 2-3 circles of sensors are arranged in the ring, with the inner circle close to the device and the outer circle further extended to ensure all-round monitoring around the device.

4. The method for predicting electromagnetic environment of wireless power transmission based on PSO-RBF neural network according to claim 1 is characterized in that: Step 2: Construction of RBF neural network model. The specific method is as follows: 2.

1. Constructing RBF neural network model structure The structure includes input layer, hidden layer and output layer; 2.

2. Design of hidden layer structure The hidden layer consists of multiple radial basis function neurons, each neuron uses a Gaussian function as the activation function, and its expression is: Among them, c i is the center position of the ith neuron, σ is the expansion parameter, r = ||vc i || is the Euclidean distance between the input vector and the neuron center, and v is the spatial coordinate data collected by sensor i; 2.

3. Model output calculation Predicted values ​​of electromagnetic field strength The weighted sum of all neurons is calculated as: Among them, w i is the output layer weight parameter, N is the number of RBF neurons, and (x, y, z) is the spatial coordinates of the electromagnetic field strength that needs to be predicted.

5. The method for predicting electromagnetic environment of wireless power transmission based on PSO-RBF neural network according to claim 1, characterized in that: Step 3: construct a PSO-RBF neural network wireless power transmission electromagnetic environment prediction model. The specific method is as follows: 3.1.RBF neural network model training The RBF neural network model training process includes: determining the center c of the RBF neuron i and width σ i , calculate the weight w of the output layer i To minimize the prediction error, as follows: a. Determine the center c i RBF neuron center c i Determination: Using the K-means clustering algorithm, the spatial coordinate data (x, y, z) collected by all sensors are divided into M clusters, and the center of each cluster is used as the center c of an RBF neuron i , the algorithm is as follows: Randomly select M spatial coordinate data points as the initial center; Calculate the Euclidean distance from each spatial coordinate data point to all centers, and assign the corresponding spatial coordinate data point to the cluster where the nearest center is located according to the size of the Euclidean distance; Recalculate the center of each cluster, that is, update where n i is the number of spatial coordinate data points in the i-th cluster, l is the spatial coordinate data selected as the initial center; Repeat the above process until all centers no longer change significantly or the maximum number of iterations is reached; b. RBF neuron width σ i Calculation Calculate the width σ of each RBF neuron i , σ i Set to the average distance between two adjacent centers, or calculated according to formula (13): Among them, ||c i -c j || represents the Euclidean distance between the i-th and j-th centers; c. Calculate the output layer weight w i Calculate the spatial coordinate data of all sensor inputs through the response of RBF neurons to form a matrix Φ, whose elements φ ij Represents the output of the i-th spatial coordinate data point through the j-th RBF neuron: Among them, v i is the spatial coordinate of the i-th input; The least squares method is used to calculate the weight w i :Solve the weight w by minimizing the objective function error E i , the error is defined as: E=‖Y-ΦW‖ 2 (15) Among them, Y is the actual output target vector, that is, the electromagnetic field strength, Φ is the matrix composed of the response values ​​of all spatial coordinate data, and W is the weight vector to be solved. The least squares method is used to solve: W = (Φ T Φ) -1 Φ T Y; 3.

2. Optimizing the parameters of RBF neural network using particle swarm optimization algorithm a. Initialize particle population: Initialize a number of particles in the parameter space, each particle represents a set of parameters of the RBF neural network (c i ,σ i ,w i ), and set the initial speed and position at the same time; b. Define the fitness function: Use mean square error (MSE) as the fitness function, and the goal is to minimize MSE: c. Particle velocity and position update: According to the PSO algorithm rules, iteratively update the particle velocity v i (t) and position x i (t): v i (t+1)=w·v i (t)+c1·r1·(p best -x i (t))+c2·r2·(g best -x i (t)) (11) x i (t+1)=x i (t)+v i (t+1) (12) Among them, the parameters in formula (11), i.e. the speed update formula, are explained as follows: v i (t+1): The velocity of particle i at time t+1. The velocity determines the moving direction and step size of the particle in the search space. w·v i (t): Inertia term, which indicates the influence of the particle's velocity at the previous moment on the current velocity. w is the inertia weight, which is used to adjust the particle's dependence on history. A larger w value is conducive to global search, while a smaller w value is conducive to local search. c1·r1·(p best -x i (t)): cognitive term, indicating that the particle has its own historical optimal position p best to update the speed. c1 is the learning factor, which is used to adjust the degree to which the particle learns from its own historical optimal position. r1 is a random number between [0,1], which increases the randomness of the search. c2·r2·(g best -x i (t)):Social term, which means the particle has the best historical position g in the group best to update the speed. c2 is the learning factor, which is used to adjust the degree to which the particle learns from the historical optimal position of the group. r2 is also a random number between [0,1]. Formula (12), i.e. the parameters of the position update formula, are explained as follows: x i (t+1): The position of particle i at time t+1. The position represents a potential solution for the particle in the search space. x i (t): The position of particle i at time t. v i (t+1): The velocity of particle i at time t+1. This velocity is calculated by the velocity update formula and determines how the particle moves to the new position. d. Convergence judgment: Repeat the update process until the convergence conditions are met and the optimal RBF neural network parameter set is obtained; 3.

3. Training and verification of RBF neural network model parameters In order to verify the generalization ability of the model, the k-fold cross validation method is used, as follows: a. Divide the training data set into k subsets, use k-1 subsets to train the RBF neural network model each time, and use the remaining subset for verification, and repeat this process k times; b. Calculate the mean square error MSE for each validation: Among them, N i is the number of data points for the i-th validation, B j is the actual value, is the predicted value; c. The validation error of the final model is the average of all k validation errors: d. If E cv If the value exceeds the set threshold, the algorithm returns to step 3.2 to optimize the parameters and repeat the training and verification process until the expected accuracy is achieved.

6. The method for predicting electromagnetic environment of wireless power transmission based on PSO-RBF neural network according to claim 1, characterized in that: Step 4: electromagnetic field prediction and result output. The specific method is as follows: Given the spatial coordinates (x, y, z) of the position to be measured, the electromagnetic field strength at the corresponding position can be calculated: Where N is the number of RBF neurons, w i 、c i and σ i are the parameters after training.

7. The method for predicting electromagnetic environment of wireless power transmission based on PSO-RBF neural network according to claim 1, characterized in that: Also includes: According to ICNIRP or IEEE C95.1 standards, the prediction results are classified as follows: i. Low risk level: The predicted value is less than 10% of the international standard limit, indicating that the exposure risk is extremely low and no additional protective measures are required; ii. Medium risk level: The predicted value is between 10% and 50% of the international standard limit, indicating that there is a certain risk, and it is recommended to monitor the exposure time or use protective measures; iii. High risk level: The predicted value is close to or exceeds the international standard limit, indicating a high risk and requiring immediate protective measures or reduction of exposure.

8. A wireless power transmission electromagnetic environment prediction system based on PSO-RBF neural network, characterized in that: Implement the method for predicting the electromagnetic environment of wireless power transmission based on the PSO-RBF neural network as described in any one of claims 1 to 7 to realize the prediction of the electromagnetic environment of wireless power transmission based on the PSO-RBF neural network.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method for predicting the electromagnetic environment of wireless power transmission based on the PSO-RBF neural network according to any one of claims 1 to 7 is implemented to realize the prediction of the electromagnetic environment of wireless power transmission based on the PSO-RBF neural network.

10. A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the method for predicting the electromagnetic environment of wireless power transmission based on a PSO-RBF neural network according to any one of claims 1 to 7 is implemented to realize the prediction of the electromagnetic environment of wireless power transmission based on a PSO-RBF neural network.

Images