Wireless power transmission system magnetic field prediction method based on improved particle swarm optimization neural network algorithm

Abstract

The invention discloses a wireless power transmission system magnetic field prediction method based on an improved particle swarm optimization neural network algorithm, and the method comprises the steps: building an RBF neural network model with a transmission distance, a power voltage, a load resistance and a spatial three-dimensional coordinate as input features and a wireless power transmission system magnetic field intensity as output; uniformly distributed initial candidate solutions are generated in a high-dimensional space of RBF neural network parameters through Latin hypercube sampling, and search parameters are dynamically adjusted by using an APSO algorithm, so that an optimal parameter combination of an RBF neural network model is determined; the trained RBF neural network model with the optimal parameters reserved is embedded into prediction software, online prediction is carried out according to the parameters of the wireless power transmission system under the actual working condition, corresponding magnetic field intensity distribution is output, and real-time monitoring and early warning of the electromagnetic environment risk of the system are achieved. According to the method, the spatial magnetic field intensity under different working conditions can be accurately predicted, and timely identification and early warning of potential electromagnetic environment risks are realized.

Description

Technical Field

[0001] The present invention relates to wireless power transfer technology, and particularly to a method based on the combination of adaptive Latin hypercube sampling particle swarm optimization (LHS-APSO) and radial basis function (RBF) for predicting the magnetic induction intensity values of a wireless power transfer system (WPT) under different working conditions. Background Art

[0002] WPT technology not only shows broad application prospects in the fields of home, industry, and medical care, but also provides new solutions for fields such as intelligent transportation, the Internet of Things, and intelligent devices. It breaks through the limitation of power transmission relying on wire connections and opens up a new path for power supply. Especially in aspects such as wireless charging of electric vehicles, remote power supply of implantable medical devices, and long-term power supply of Internet of Things sensors, WPT technology shows great potential. With the continuous progress of technology, WPT is expected to completely revolutionize the power supply method and promote the development of energy transmission towards intelligence and flexibility.

[0003] During the operation of the WPT system, the evaluation and optimization of the electromagnetic environment mainly focus on two aspects: electromagnetic compatibility (EMC) and electromagnetic radiation impact. First, electromagnetic interference may occur when the WPT system coexists with other devices, affecting the operation stability of the system and even causing abnormal device functions. Studying the electromagnetic compatibility of the WPT system helps to ensure the efficient operation of the system in different electromagnetic environments and reduce interference with other devices at the same time. Second, the radiation characteristics of the electromagnetic field pose potential risks to the surrounding environment and human health. Especially in high-power transmission scenarios, the local electromagnetic field intensity may reach a relatively high level, and a comprehensive assessment of the safety of electromagnetic radiation is required. Summary of the Invention

[0004] The purpose of the present invention is to propose a magnetic field prediction method for a wireless power transfer system based on an improved particle swarm optimization neural network algorithm.

[0005] The technical solution for achieving the purpose of the present invention is: a magnetic field prediction method for a wireless power transfer system based on an improved particle swarm optimization neural network algorithm, including the following steps:

[0006] Step 1, simulation model construction, data collection and processing:

[0007] Construct a three-dimensional simulation model of the WPT system in Ansys Maxwell, calculate the self-inductance, mutual inductance of the coils and the currents at the transmitting end and receiving end under different working conditions through finite element analysis and LCC-S topology, and then use the built-in field calculator in Ansys Maxwell to export the magnetic field intensity data at each spatial position. After data cleaning and processing, a training sample data set is formed;

[0008] Step 2, Construction of RBF neural network model:

[0009] Taking the transmission distance, power supply voltage, load resistance, and three-dimensional space coordinates as input features, and the magnetic field intensity of the wireless power transmission system as the output, an RBF neural network model is constructed;

[0010] Step 3, Global search and optimization of neural network parameters by LHS optimized adaptive particle swarm algorithm:

[0011] Uniformly distributed initial candidate solutions are generated in the high-dimensional space of RBF neural network parameters through Latin hypercube sampling (LHS), and the APSO algorithm is used to dynamically adjust the search parameters, thereby determining the optimal parameter combination of the RBF neural network model;

[0012] Step 4, Prediction of the magnetic field intensity of the wireless power transmission system:

[0013] The trained RBF neural network model with optimal parameters retained is embedded into the prediction software, and online prediction is carried out according to the parameters under the actual working conditions of the wireless power transmission system, and the corresponding magnetic field intensity distribution is output to realize real-time monitoring and early warning of the electromagnetic environment risk of the system.

[0014] Furthermore, in Step 1, simulation model construction, data collection and processing: A three-dimensional simulation model of the WPT system is constructed in Ansys Maxwell. Through finite element analysis and LCC-S topology, the self-inductance, mutual inductance of the coils, and the currents at the transmitting end and receiving end under different working conditions are calculated. Then, the magnetic field intensity data at each spatial position is exported using the built-in field calculator in Ansys Maxwell, and a training sample data set is formed after data cleaning and processing. The specific method is as follows:

[0015] 1.1. Construction of a three-dimensional electromagnetic simulation model of the wireless power transmission system

[0016] First, a three-dimensional electromagnetic simulation model of the wireless power transmission system is constructed in Ansys Maxwell, including the coil models at the transmitting end and receiving end. A current excitation is added to the coils to obtain the self-inductance and mutual inductance values of the coils varying with the transmission distance;

[0017] Construct the LCC-S circuit topology diagram of the wireless power transmission system. According to Kirchhoff's voltage law, the current-voltage relationship formula is obtained, and there is:

[0018]

[0019] where is the AC power supply; L1 is the compensation inductor at the transmitting end, L p is the self-inductance of the transmitting end coil, L sis the self-inductance of the receiving coil; C1, C p is the transmitter compensation capacitor, C s is the compensation capacitor at the receiving end, M ps is the mutual inductance generated by the transmitting coil and the receiving coil; R eq is the equivalent load impedance; is the input current,

[0020] is the transmitting coil current, is the receiving coil current; ω is the transmission frequency;

[0021] When considering the maximum transmission efficiency of the wireless power transmission system, it works in a resonant state, the inductive reactance and the capacitive reactance cancel each other out, and the impedance presents a pure resistance characteristic, then:

[0022]

[0023] Substituting equation (2) into equation (1) and simplifying it, we get the relationship between the currents of each part:

[0024]

[0025] After substituting mutual inductance, self-inductance, and different power supply voltages and load resistance values ​​into the calculation formula, different current values ​​of the transmitting end and the receiving end are obtained. Then, the current values ​​of the transmitting end and the receiving end are respectively input into the three-dimensional model of the wireless power transmission system as excitations. The magnetic induction intensity values ​​at various spatial positions under different working conditions are obtained through finite element analysis.

[0026] 1.2. Calculate the coil self-inductance, mutual inductance, and transmitting and receiving currents under different working conditions through finite element analysis and LCC-S topology;

[0027] First, the complex amplitude value of the magnetic induction intensity is extracted through the field calculator provided by Ansys Maxwell. Then, a one-to-one correspondence is established between the power supply voltage, load resistance, transmission distance, and three-dimensional spatial coordinates and the corresponding magnetic induction intensity output through Excel.

[0028] 1.3. After data cleaning, a training sample data set is formed;

[0029] The interquartile range method (IQR) is used to remove outliers from the magnetic field intensity data samples, remove abnormal values ​​and noise, and ensure the validity and integrity of the data;

[0030] a. Calculate the first quartile (Q1) and the third quartile (Q3) of the electromagnetic field strength data collected by the field calculator;

[0031] b. Calculate the interquartile range (IQR), defined as:

[0032] IQR = Q3 - Q1 (4)

[0033] c. Set the rejection threshold and remove all data points outside the following range to ensure data quality:

[0034] [Q1 - 1.5×IQR, Q3 + 1.5×IQR] (5).

[0035] Furthermore, in step 2, construct the RBF neural network model: Use the transmission distance, power supply voltage, load resistance, and three-dimensional space coordinates as input features, and the magnetic field intensity of the wireless power transmission system as the output to construct the RBF neural network model. The specific method is as follows:

[0036] 2.1 Center selection

[0037] For each magnetic field intensity sample, calculate the local density in the corresponding input feature space, and use the truncation distance d c to count the number of samples that satisfy ||x i - x j || ≤ d c to obtain:

[0038]

[0039] where 1() is the indicator function, which takes the value of 1 when the distance between the sample x j and x i does not exceed d c , and 0 otherwise; then calculate the minimum distance between each sample and the sample with higher density for each sample, that is:

[0040]

[0041] For the sample with the maximum density, let δ i be the maximum value among all distances;

[0042] Finally, calculate the index:

[0043] γ i = ρ i × δ i (8)

[0044] and select several points with the highest γ i as the clustering centers, and assign the remaining samples to the nearest center;

[0045] 2.2. Calculation of the width σ i of the RBF neuron

[0046] For any center c i , first find the nearest neighbor c jAnd calculate the Euclidean distance:

[0047] d ij =||c i -c j || (9)

[0048] Subsequently, calculate the width of the RBF neuron:

[0049] σ i =β·d ij (10)

[0050] where β is an adjustable hyperparameter;

[0051] 2.3. Calculation of the output layer weights

[0052] Let x represent the input feature, then the output of the i-th mapping unit is:

[0053]

[0054] where c i is the center of the i-th neuron, and σ i is its width;

[0055] Combine the responses of all samples into a matrix

[0056] Use the orthogonal least squares method to gradually orthogonalize each column of Φ, and successively select the column that contributes the most to the current residual to construct an upper triangular linear system to solve the output layer weight W and calculate the output layer weight w i ;

[0057] Finally, the prediction expression is

[0058]

[0059] where y(x) is the predicted magnetic field intensity of the wireless power transfer system..

[0060] Furthermore, in step 3, the LHS-optimized adaptive particle swarm optimization algorithm is used to globally search for and optimize the neural network parameters: Uniformly distributed initial candidate solutions are generated in the high-dimensional space of the RBF neural network parameters through Latin hypercube sampling (LHS), and the APSO algorithm is used to dynamically adjust the search parameters, thereby determining the optimal parameter combination of the RBF neural network model. The specific method is as follows:

[0061] 3.1. Take the center, width, and weight parameters of the RBF neural network model as particles, and generate uniformly distributed initial candidate solutions in the high-dimensional space of the RBF neural network parameters through Latin hypercube sampling (LHS);

[0062] Using Latin hypercube sampling, the value range of each dimension of the parameter to be optimized is evenly divided, and values are assigned to each particle;

[0063]

[0064] In the formula is the initial position of the i-th particle in the d-th dimension; P max and P min , are the upper and lower bounds of the d-th dimension respectively, and ρ i is a random number, and its value range is [0, 1];

[0065] 3.2. Dynamically adjust the search parameters using the APSO algorithm to determine the optimal parameter combination of the RBF neural network model;

[0066] In the D-dimensional search space, the current velocity and position of the particle are expressed as:

[0067] a i =[a i,1 , a i,2 , …, a i,D (13)

[0068] v i =[v i,1 , v i,2 , …, v i,D (14)

[0069] where i = 1, 2, …, s, s is the number of particle swarms, D is the vector space dimension of K(n + 2), n is the input vector space dimension of the magnetic field strength sample set, and K is the number of neurons in the hidden layer;

[0070] The fitness of the particle is

[0071]

[0072] where y(t) is the predicted magnetic field strength, y d (t) is the actually measured magnetic field strength, and T is the input feature sample of the wireless power transfer system;

[0073] During the search process, each particle will evaluate the quality of the current position according to the fitness function and record its own best position p i (t):

[0074] p i (t)=[p i,1 (t), p i,2 (t), …, p i,D (t)] (16)

[0075] The optimal solution found by all particles in the entire particle swarm is the global optimum g(t):

[0076] g(t) = [g1(t), g2(t), …, g D (t)] (17)

[0077] Based on the individual extreme value p i (t) and the global optimal position g(t), the velocity and position are updated. The update formula for velocity is as follows:

[0078] v i,d (t + 1) = v i,d (t)ω i,d (t) + c1r1(p i,d (t) - a i,d (t)) + c2r2(g d (t) - a i,d (t)) (18)

[0079] where d = 1, 2, …, D; c1, c2 are learning factors; r1, r2 are random numbers in the range [0, 1]; ω i,d is the inertia weight value;

[0080] The diversity is defined as:

[0081] S(t) = f min (a(t)) / f max (a(t)) (19)

[0082] where f min (a(t)) and f max (a(t)) are the minimum and maximum values of the magnetic field strength prediction errors of all particles at time t;

[0083] Based on the diversity, a non - linear regression function is designed:

[0084] γ(t) = (l - S(t)) -1 (20)

[0085] where l is an initialization constant, and l ≥ 2;

[0086] Calculate the difference between the particle and the optimal particle:

[0087] A i (t) = f(g(t)) / f(a i (t)) (21)

[0088] where f(g(t)) is the global optimum, and f(a i (t)) is the fitness value of the i - th particle;

[0089] Adaptive adjustment of inertia weight, which is used to adjust the inertia weight in the velocity formula;

[0090] ω i (t) = γ i (t)(A i (t) + c) (22)

[0091] In the formula, ω i (t) is the inertia weight of the i-th particle at time t; c is a constant.

[0092] Furthermore, in step 4, prediction of the magnetic field intensity of the wireless power transfer system: Embed the trained RBF neural network model with optimal parameters retained into the prediction software, and perform online prediction based on the parameters under the actual working conditions of the wireless power transfer system, output the corresponding magnetic field intensity distribution, and realize real-time monitoring and early warning of the electromagnetic environment risk of the system. Among them, the specific method for realizing real-time monitoring and early warning of the electromagnetic environment risk of the system is as follows:

[0093] a. Region division

[0094] Divide the entire monitoring area into several uniform grid cells according to a predetermined size. For example, each grid covers a certain area or a square area with a fixed side length;

[0095] b. Statistical grid data

[0096] For each grid cell, calculate the average field intensity of all points within the grid:

[0097]

[0098] where B i is the field intensity of the i-th measurement point within the grid, and n is the number of points within the grid;

[0099] c. Threshold judgment

[0100] According to the ICNIRP or GB 8702-2014 standard, divide the prediction results with a safe magnetic field intensity value of 27 μT:

[0101] i. Low risk level: The predicted value is lower than 10 μT, indicating a very low exposure risk and no need to take additional protective measures;

[0102] ii. Medium risk level: The predicted value is between 10 μT and 20 μT, indicating a certain risk. It is recommended to monitor the exposure time or use protective measures;

[0103] iii. High risk level: The predicted value is close to or exceeds 27 μT, indicating a high risk and the need to take immediate protective measures or reduce exposure.

[0104] A magnetic field prediction system for a wireless power transfer system based on an improved particle swarm optimization neural network algorithm, implementing the magnetic field prediction method for the wireless power transfer system based on the improved particle swarm optimization neural network algorithm, achieving magnetic field prediction for the wireless power transfer system based on the improved particle swarm optimization neural network algorithm, and separately executing steps 1 to 4 in four modules.

[0105] A computer device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the magnetic field prediction method for the wireless power transfer system based on the improved particle swarm optimization neural network algorithm, achieving magnetic field prediction for the wireless power transfer system based on the improved particle swarm optimization neural network algorithm.

[0106] A computer-readable storage medium stores a computer program. When the computer program is executed by a processor, it implements the magnetic field prediction method for the wireless power transfer system based on the improved particle swarm optimization neural network algorithm, achieving magnetic field prediction for the wireless power transfer system based on the improved particle swarm optimization neural network algorithm.

[0107] Compared with the prior art, the significant advantages of the present invention are as follows: First, the improved particle swarm optimization algorithm is combined with the RBF neural network, and the global search is used to optimize the model parameters, which can effectively reduce the training error and accurately fit the complex electromagnetic field distribution, thereby achieving high-precision prediction of the magnetic field intensity of the wireless power transfer system. Second, the graphical software constructed by MATLAB APPDesigner realizes real-time data acquisition, processing, and prediction; and divides the regional risk according to the preset threshold, which can timely detect potential electromagnetic environment risks and provide dynamic early warning for the safe operation of the system.

[0108] Third, the present invention is applicable to wireless power transfer systems with various coupling mechanism coils (such as DD coils and circular coils), can effectively predict the magnetic field intensity under different working conditions, has high versatility and practical value, and is applicable to electromagnetic environment evaluation and optimal control in complex and variable environments such as electric vehicle wireless charging and industrial automation equipment power supply. Description of the Drawings

[0109] Figure 1 Magnetic field intensity cloud map around the WPT model under the built DD coil

[0110] Figure 2 LCC-S circuit topology diagram

[0111] Figure 3 Process schematic diagram of the LHS-APSO-RBF neural network model.

[0112] Figure 4The predicted result of the magnetic induction intensity around the circular coil.

[0113] Figure 5 It is the interface diagram of the prediction software designed by APP Designer of Matlab. Specific implementation manners

[0114] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0115] A magnetic field prediction method for a wireless power transmission system based on an improved particle swarm optimization neural network algorithm of the present invention is realized through the following steps:

[0116] Step 1, simulation model construction, data collection and processing:

[0117] Build a three-dimensional simulation model of the WPT system in Ansys Maxwell. Through finite element analysis and LCC-S topology, calculate the self-inductance, mutual inductance of the coils and the currents at the transmitting end and receiving end under different working conditions (transmission distance, power supply voltage, load resistance), and then substitute the calculated current excitation into the three-dimensional model of the wireless power transmission system. Use the built-in field calculator of Ansys Maxwell to export the magnetic field intensity data at each spatial position, and form a training sample data set after data cleaning and processing.

[0118] Specifically as follows:

[0119] 1.1. Construction of the wireless power transmission system model:

[0120] Build a three-dimensional electromagnetic simulation model of the wireless power transmission system. First, establish the coil models of the transmitting end and the receiving end in Ansys Maxwell respectively. Add current excitation to the coils in the Ansys Maxwell software to obtain the self-inductance and mutual inductance values of the coils varying with the transmission distance.

[0121] Through the LCC-S circuit topology diagram of the wireless power transmission system (see attached Figure 2 ), according to Kirchhoff's voltage law, obtain the current-voltage relationship formula, then there is:

[0122]

[0123] Where is the AC power supply; L1 is the compensation inductor at the transmitting end, L p is the self-inductance of the transmitting end coil, L s is the self-inductance of the receiving end coil; C1, C p is the compensation capacitor at the transmitting end, C sis the compensation capacitor at the receiving end, M ps is the mutual inductance generated by the transmitting coil and the receiving coil; R eq is the equivalent load impedance; is the input current,

[0124] is the transmitting coil current, is the receiving coil current; ω is the transmission frequency.

[0125] When considering the maximum transmission efficiency of the wireless power transfer system, the wireless power transfer system operates in the resonant state. The inductive reactance and capacitive reactance in the LCC - S circuit topology cancel each other out, and the impedance of the wireless power transfer system exhibits a pure resistance characteristic. Then, there is:

[0126]

[0127] Substituting Equation (2) into (1) and simplifying, the relationship between the currents of each part in the circuit topology diagram can be obtained:

[0128]

[0129] After substituting the mutual inductance, self - inductance, and different power supply voltages and load resistance values into the formula for calculation, the current values of different transmitting and receiving ends can be obtained. Then, the current values of the transmitting and receiving ends are respectively used as excitation inputs to the three - dimensional model of the wireless power transfer system, and the magnetic induction intensity values at each spatial position under different working conditions are obtained through finite element analysis.

[0130] 1.2. Data Acquisition

[0131] Extract the complex amplitude value of the magnetic induction intensity through the field calculator built in Ansys Maxwell. In the field calculator, first set the boundary and step size of the measurement area and the three - dimensional coordinates. After completing the data acquisition, export the magnetic field intensity data. Subsequently, organize the data through Excel to establish a one - to - one correspondence between each input parameter (power supply voltage, load resistance, transmission distance, and three - dimensional spatial coordinates) and the corresponding magnetic induction intensity output.

[0132] 1.3. Data Processing

[0133] Use the interquartile range method (IQR) to remove outliers from the magnetic field intensity data samples, removing abnormal values and noise to ensure the validity and integrity of the data

[0134] a. Calculate the first quartile (Q1) and the third quartile (Q3) of the electromagnetic field intensity data collected by the field calculator;

[0135] b. Calculate the interquartile range (IQR), defined as:

[0136] IQR = Q3 - Q1 (4)

[0137] c. Set the rejection threshold and remove all data points outside the following range to ensure data quality:

[0138] [Q1 - 1.5×IQR, Q3 + 1.5×IQR] (5)

[0139] Step 2, Construction of the RBF neural network model:

[0140] Based on the processed data, construct an RBF neural network model with the transmission distance, power supply voltage, load resistance, and three-dimensional space coordinates as input features and the magnetic field strength of the wireless power transmission system as the output. This model includes an input layer, a hidden layer, and an output layer, and preliminarily describes the non-linear mapping relationship between the input parameters and the magnetic field strength.

[0141] The training process of the RBF neural network model includes: determining the center c of the RBF neurons i and the width σ i , and calculating the weight w of the output layer i to minimize the prediction error.

[0142] Specifically as follows:

[0143] 2.1 Center selection

[0144] For each magnetic field strength sample, first calculate the local density in its corresponding input feature space (transmission distance, power supply voltage, load resistance, and space coordinates), and use the truncation distance d c to count the number of samples that satisfy ||x i - x j || ≤ d c to obtain:

[0145]

[0146] where 1() is the indicator function, which takes the value of 1 when the distance between the sample x j and x i does not exceed d c , and 0 otherwise. Then calculate the minimum distance between each sample and the sample with higher density, that is:

[0147]

[0148] For the sample with the maximum density, its δ i can be set to the maximum value among all distances;

[0149] Finally, calculate the index:

[0150] γ i = ρi ×δ i (8)

[0151] and select the point with the highest γ i as the clustering center, and assign the remaining samples to the nearest center. The selected clustering center is the key representative point in the distribution of magnetic field strength data.

[0152] 2.2. Calculation of the width σ of the RBF neuron i

[0153] For any center c i , first find the nearest neighbor c j among all centers and calculate the Euclidean distance:

[0154] d ij =||c i -c j || (9)

[0155] Subsequently, set the width to a proportion of this distance:

[0156] σ i =β·d ij (10)

[0157] where β is an adjustable hyperparameter. In this way, the centers in the dense region will have a smaller expansion range, while the sparse region allows a larger expansion range, thus more precisely characterizing the local distribution characteristics of the data;

[0158] 2.3. Calculation of the output layer weights

[0159] Let x represent the input (such as transmission distance, power supply voltage, load resistance, spatial coordinates), then the output of the i-th mapping unit is:

[0160]

[0161] where c i is the center of the i-th neuron, and σ i is its width;

[0162] Combine the responses of all samples into a matrix

[0163] Orthogonalize each column of Φ step by step using the orthogonal least squares method, and successively select the column that contributes the most to the current residual to construct an upper triangular linear system to solve the output layer weight W and calculate the output layer weight w i ;

[0164] Finally, the prediction expression of the RBF model is

[0165] ​

[0166] Among them, y(x) is the predicted magnetic field intensity of the wireless power transfer system.

[0167] Step 3: The LHS-optimized adaptive particle swarm optimization algorithm performs global search and optimization on the neural network parameters

[0168] Uniformly distributed initial candidate solutions are generated in the high-dimensional space of the RBF neural network parameters through Latin hypercube sampling (LHS), providing a good starting point for adaptive particle swarm optimization (APSO). The improved APSO algorithm dynamically adjusts the search parameters to achieve a balance between global search and local exploitation, thereby determining the optimal parameter combination, which is used for subsequent RBF network training and verification.

[0169] Specifically as follows:

[0170] 3.1 Adaptive particle swarm optimization algorithm

[0171] The PSO algorithm regards each particle as a parameter combination (including the center, width, and connection weight) of an RBF to be optimized in the magnetic field intensity prediction model of the wireless power transfer system. The position and velocity of the particle are represented by vectors respectively. In the D-dimensional search space, the current velocity and position of the particle can be expressed as:

[0172] a i =[a i,1 , a i,2 , …, a i,D (13)

[0173] v i =[v i,1 , v i,2 ,..., v i,D (14)

[0174] where i = 1, 2, …, s, s is the number of particle swarms, D is the vector space dimension of K(n + 2), n is the input vector space dimension of the magnetic field intensity sample set, and K is the number of hidden layer neurons.

[0175] During the optimization process, the parameter combination represented by each particle is substituted into the previously constructed RBF prediction model. The magnetic field intensity is predicted through the optimized RBF model and compared with the actual magnetic field intensity samples, thereby calculating the fitness value of the particle.

[0176]

[0177] where y(t) is the magnetic field intensity predicted by the network, y d (t) is the actually measured magnetic field intensity, and T is the input feature sample of the wireless power transfer system.

[0178] During the search process, each particle evaluates the quality of its current position according to the fitness function and records its own best position p i (t):

[0179] p i (t)=[p i,1 (t),p i,2 (t),…,p i,D (t)](16)

[0180] The optimal solution found by all particles in the entire particle swarm is the global optimum g(t):

[0181] g(t)=[g1(t),g2(t),…,g D (t)](17)

[0182] The particle updates its velocity and position based on the individual extreme value p i (t) and the global optimal position g(t). The velocity update formula is as follows:

[0183] v i,d (t + 1)=v i,d (t)ω i,d (t)+c1r1(p i,d (t)-a i,d (t))+c2r2(g d (t)-a i,d (t))(18)

[0184] Where d = 1, 2, …, D; c1, c2 are learning factors; r1, r2 are random numbers in the range of [0, 1] to enhance the search ability; ω i,d is the inertia weight value.

[0185] Secondly, through an adaptive strategy, the inertia weight in the velocity formula is adjusted. The inertia weight is optimized based on a non-linear strategy of population diversity. Diversity is defined as:

[0186] S(t)=f min (a(t)) / f max (a(t)) (19)

[0187] Where f min (a(t)) and f max (a(t)) are the minimum and maximum values of the magnetic field intensity prediction errors of all particles at time t.

[0188] Based on diversity, a non-linear regression function is designed to adjust the inertia weight to make it more suitable for the flight state. The function is as follows:

[0189] γ(t)=(l - S(t)) -1(20)

[0190] Among them, l is an initialization constant, and l ≥ 2.

[0191] In addition, the spatial states of each particle are different, and the inertia weight needs to be adaptively adjusted according to the state of the particle to guide the flight of each particle. The difference between the particle and the optimal particle can well reflect the current optimal difference of the particle, so as to guide the particle flight. The difference between the particle and the optimal particle is expressed as:

[0192] A i (t) = f(g(t)) / f(a i (t)) (21)

[0193] In the formula, f(g(t)) is the global optimum, and f(a i (t)) is the fitness value of the i-th particle.

[0194] Therefore, there is an adaptive strategy as

[0195] ω i (t) = γi i (t)(A i (t) + c) (22)

[0196] In the formula, ω i (t) is the inertia weight of the i-th particle at time t; c is a constant used to improve the global search ability.

[0197] The initial values of the parameters in the above APSO algorithm will directly affect the subsequent search performance. Therefore, the Latin Hypercube Sampling (LHS) method in Section 3.2 below needs to be used for initialization to ensure the uniform distribution of the initial positions of the particles.

[0198] 3.2. Latin Hypercube Optimization Algorithm

[0199] To solve the problem of uneven distribution of initial particles in the traditional APSO algorithm, the present invention adopts the Latin Hypercube Sampling (LHS) method to optimize the distribution of initial particles. Using Latin Hypercube Sampling, for the parameter space of the RBF magnetic field prediction model, the value range of each dimension of the parameter to be optimized is evenly divided, and values are assigned to each particle to fit the value of the input vector:

[0200]

[0201] In the formula is the initial position of the i-th particle in the d-th dimension; P max and P min , are the upper and lower bounds of the d-th dimension respectively, and ρ i is a random number with a value range of [0, 1].

[0202] Through the LHS, the particle swarm was evenly divided, and the LHS-APSO-RBF model combining Latin hypercube sampling and APSO-RBF was obtained.

[0203] 3.3. Verification of the Effect of the Optimization Algorithm

[0204] To verify the effectiveness and optimization performance of the above LHS-APSO-RBF method, the following verification experiments were further designed.

[0205] a. Verifying the LHS-APSO-RBF neural network using a non-linear function:

[0206] y = 0.5e -0.5x sin(5x)

[0207] In this experiment, 500 sets of training samples between [0, 2] were selected for the x coordinate, and 80% of the data was used as the training group, while 20% of the data was used as the test group samples. 5, 10, and 15 neurons were respectively selected and substituted into the LHS-APSO-RBF model obtained above for verification, and the predicted values output were relatively close to the actual values.

[0208] b. Testing LHS-APSO-RBF using non-linear system identification. The non-linear system is composed of the following formula:

[0209] y(t + 1) = 0.72y(t) + 0.025y(t - 1)u(t - 1) + 0.01u 2 (t - 2) + 0.2u(t - 3)(25)

[0211] Among them, the non-linear system determines the output y(t + 1) through the inputs y(t) and u(t). And the training inputs are divided into two parts. One part is evenly distributed between [-2, 2], while the other part of the input is determined by the sinusoid function 1.05×sin(t / 45). The test input u(t) is a piecewise function, as shown in formula (3.20):

[0212]

[0213] This system is often used to verify the performance of neural networks. There are 800 sets of training samples and 200 sets of test samples. 5, 10, and 15 neurons are respectively used in the hidden layer. To verify the performance of the proposed method, LHS-APSO-RBF is compared with other RBF neural network structures, as shown in Table 1.

[0214] Table 1 Comparison Table of LHS-APSO-RBF and Other RBF Neural Networks

[0215]

[0216] The LHS-APSO-RBF can approximate non-linear functions well, and has better effects than other algorithms, and can better optimize the parameters of the neural network.

[0217] Step 4, Prediction of the magnetic field intensity of the wireless power transfer system

[0218] Embed the trained RBF neural network model with the optimal parameters reserved into the prediction software, and perform online prediction based on the parameters under the actual working conditions of the wireless power transfer system (including transmission distance, power supply voltage, load resistance, and spatial coordinates of the measurement position), and output the corresponding magnetic field intensity distribution to realize the real-time monitoring and early warning of the electromagnetic environment risk of the system.

[0219] Specifically as follows:

[0220] 4.1. Build the software program

[0221] a. Parameter input module

[0222] The user enters the key working condition parameters of the wireless power transfer system through the graphical interface, including transmission distance, power supply voltage, load resistance, and three-dimensional spatial coordinates of the measurement point;

[0223] b. Data preprocessing module

[0224] Standardize and verify the input parameters to ensure that the data format and range meet the requirements of subsequent processing;

[0225] c. Model parameter calling module

[0226] Automatically load and call the optimal model parameters obtained through the improved particle swarm optimization-RBF neural network training in advance, and these parameters have been stored inside the software;

[0227] d. Prediction calculation module

[0228] According to the preprocessed input data and the loaded model parameters, use the improved PSO-RBF model to calculate the magnetic field intensity at the corresponding position in real time

[0229] 4.2. Magnetic field calculation

[0230] Given the spatial coordinates (x, y, z) of the measurement position, the transmission distance h, the power supply voltage U, and the load resistance R, calculate the electromagnetic field intensity at the corresponding position:

[0231]

[0232] Among them, N is the number of RBF neurons, w i , c i and σ iAre the parameters after training.

[0233] 4.3. Result Prediction and Analysis

[0234] a. Region Division

[0235] Divide the entire monitoring area into several uniform grid cells according to a predetermined size. For example, each grid covers a certain area or a square area with a fixed side length;

[0236] b. Statistical Grid Data

[0237] For each grid cell, calculate the average field strength of all points within the grid:

[0238]

[0239] where B i is the field strength of the i-th measurement point within the grid, and n is the number of points within the grid;

[0240] c. Threshold Judgment

[0241] According to the ICNIRP or GB 8702 - 2014 standard, divide the prediction results using the safe magnetic field strength value of 27 μT:

[0242] i. Low - risk level: The predicted value is below 10 μT, indicating an extremely low exposure risk and no additional protective measures are required;

[0243] ii. Medium - risk level: The predicted value is between 10 μT and 20 μT, indicating a certain risk. It is recommended to monitor the exposure time or use protective measures;

[0244] iii. High - risk level: The predicted value is close to or exceeds 27 μT, indicating a high risk and immediate protective measures need to be taken or exposure reduced.

[0245] The present invention also proposes a magnetic field prediction system for a wireless power transmission system based on an improved particle swarm optimization neural network algorithm. Implement the magnetic field prediction method for the wireless power transmission system based on the improved particle swarm optimization neural network algorithm to achieve magnetic field prediction for the wireless power transmission system based on the improved particle swarm optimization neural network algorithm, and execute steps 1 - 4 in four modules respectively.

[0246] A computer device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the magnetic field prediction method for the wireless power transmission system based on the improved particle swarm optimization neural network algorithm to achieve magnetic field prediction for the wireless power transmission system based on the improved particle swarm optimization neural network algorithm.

[0247] A computer-readable storage medium stores a computer program thereon. When the computer program is executed by a processor, the magnetic field prediction method of the wireless power transfer system based on the improved particle swarm optimization neural network algorithm is implemented, and the magnetic field prediction of the wireless power transfer system based on the improved particle swarm optimization neural network algorithm is realized.

[0248] Embodiment

[0249] To verify the effectiveness and reliability of the electromagnetic environment prediction method of the wireless power transfer based on the PSO-RBF neural network of the present invention, a typical wireless power transfer system is selected, and the electromagnetic environment in its three-dimensional space is simulated by using the Ansys Maxwell simulation software. The specific implementation steps are as follows:

[0250] Simulation model construction: Use the Ansys Maxwell electromagnetic field simulation software to construct a three-dimensional model of a typical wireless power transfer system.

[0251] Coil model: The inner diameter of the circular coil model built in Ansys Maxwell is 30 mm, the wire spacing is 2.5 mm, the wire diameter is 1.1 mm, the number of turns is 12 turns, and the coil material is set to copper. The transmission frequency is set to 85 kHz.

[0252] Space environment setting: Set an air medium environment in the three-dimensional space between the transmitter coil and the receiver coil.

[0253] Self / mutual inductance calculation: The transmission distance between the transmitter coil and the receiver coil is 10 mm, 25 mm, 40 mm. And, in Ansys Maxwell, current excitation is added to both coils to obtain the relationship between self-inductance / mutual inductance and the transmission distance;

[0254] Table 2 Self-inductance / mutual inductance table under different transmission distances

[0255]

[0256] Current amplitude calculation: According to different circuit parameters, calculate the transmitter current and the receiver current in different cases;

[0257] Table 3 Transmitter current and receiver current tables under different cases

[0258]

[0259]

[0260] Finite element analysis setting: Load the calculated current values of the transmitter and the receiver into the model, set the true step size to 1 us, and the total simulation time to 100 ms to obtain enough sample data.

[0261] Data collection and processing: The specific steps are as follows:

[0262] Data collection: After performing finite element analysis on the simulation model of the wireless power transmission system, the overall magnetic field distribution of the wireless power transmission system was obtained. The value ranges of the x, y, and z three-dimensional coordinates were set in the field calculator of Ansys Maxwell. The value ranges of the x and y axes were [-120mm, 120mm], and the value range of the z axis was [-10mm, 190mm]. The step lengths of the three were all 40mm, and the data of the magnetic induction intensity under different working conditions and positions were obtained.

[0263] Data processing: The exported magnetic induction intensity data was imported into Excel for preliminary processing, and the one-to-one correspondence between the input quantities such as transmission distance, power supply voltage, load resistance and spatial three-dimensional coordinates and the corresponding magnetic induction intensity output was sorted out. The interquartile range method (IQR) was used to remove outliers from the data collected by the field calculator, and outliers and noise were removed to ensure the validity and integrity of the data. After the processed data was obtained, 75% of it was divided into a training group and 25% into a test group.

[0264] The neural network model of the present invention is used to predict the electromagnetic field strength of the collected simulation data:

[0265] Model training: The RBF neural network model is constructed based on the magnetic field strength data collected by the wireless power transmission system. The input features are transmission distance, power supply voltage, load resistance and spatial coordinates, and the output is magnetic field strength.

[0266] Parameter optimization: The Latin hypercube sampling (LHS) method is used to initialize the adaptive particle swarm optimization (APSO). Taking the prediction error (RMSE) as the target, the center, width and weight parameters of the RBF neural network are optimized to obtain the optimal parameters.

[0267] Prediction verification: The optimized model is used to predict the magnetic field strength under new working conditions, compared with the Ansys Maxwell simulation results, and the prediction error is calculated to verify the prediction accuracy of the model.

[0268] Result analysis: According to the training model, the training error is 0.006231, and the error comparison between the predicted value and the actual value of the improved algorithm is given in the attached figure.

[0269] Table 4 Error comparison between the predicted value and the actual value of the improved algorithm

[0270]

[0271]

[0272] Integrate the trained RBF neural network model into the software to achieve the following functions:

[0273] The user inputs the working parameters of the wireless power transfer system (such as transmission distance, power supply voltage, load resistance, and three-dimensional coordinates of the measurement points) on the interface;

[0274] Adopt the grid statistics method for all prediction points, calculate the average or maximum field strength within each grid, and conduct regional risk division based on a preset threshold (for example, low risk < 10 μT, medium risk 10–27 μT, high risk ≥ 27 μT).

[0275] In summary, the present invention proposes a method for predicting the magnetic field strength of wireless power transfer based on Latin hypercube sampling adaptive particle swarm optimization (LHS-APSO) and radial basis function (RBF) neural network. Compared with the traditional numerical simulation method, the present invention can quickly construct a high-precision prediction model using limited simulation data, effectively reduce the simulation calculation amount, and avoid a large number of repeated electromagnetic field calculations under complex working conditions. Through the global optimization of the RBF model parameters by the LHS-APSO algorithm, the generalization ability and prediction accuracy of the network are improved. The established model has strong robustness and real-time performance, and can accurately and efficiently predict the magnetic field distribution of the wireless power transfer system under different working conditions, thus providing a reliable theoretical basis and technical support for the design of wireless charging systems, electromagnetic safety assessment, and real-time optimization.

[0276] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, 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, it should be considered as the scope described in this specification.

[0277] The above-described embodiments only represent several implementation manners of the present application. The description is relatively specific and detailed, but it should not be construed as a limitation to the scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

[0278] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, 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, it should be considered as the scope described in this specification.

[0279] The above-described embodiments merely represent several implementation manners of the present application. The description thereof is relatively specific and detailed, but it should not be construed as a limitation on the scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all fall within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims

1. A magnetic field prediction method for a wireless power transmission system based on an improved particle swarm optimization neural network algorithm, characterized in that, It includes the following steps: Step 1, simulation model construction, data collection and processing: Build a three-dimensional simulation model of the WPT system in Ansys Maxwell. Calculate the self-inductance, mutual inductance of the coils and the currents at the transmitting end and receiving end under different working conditions through finite element analysis and the LCC-S topology. Then use the built-in field calculator in Ansys Maxwell to export the magnetic field intensity data at each spatial position, and form a training sample data set after data cleaning and processing; Step 2, construction of the RBF neural network model: Construct an RBF neural network model with the transmission distance, power supply voltage, load resistance, and three-dimensional spatial coordinates as input features and the magnetic field intensity of the wireless power transmission system as the output; Step 3, global search and optimization of the neural network parameters by the LHS-optimized adaptive particle swarm algorithm: Generate uniformly distributed initial candidate solutions in the high-dimensional space of the RBF neural network parameters through Latin hypercube sampling, and use the APSO algorithm to dynamically adjust the search parameters to determine the optimal parameter combination of the RBF neural network model; Step 4, prediction of the magnetic field intensity of the wireless power transmission system: Embed the trained RBF neural network model with the optimal parameters retained into the prediction software, perform online prediction according to the parameters of the wireless power transmission system under actual working conditions, output the corresponding magnetic field intensity distribution, and realize real-time monitoring and early warning of the electromagnetic environment risk of the system.

2. The magnetic field prediction method for a wireless power transfer system based on an improved particle swarm optimization neural network algorithm according to claim 1, wherein Step 1, simulation model construction, data collection and processing: Build a three-dimensional simulation model of the WPT system in Ansys Maxwell. Calculate the self-inductance, mutual inductance of the coils and the currents at the transmitting end and receiving end under different working conditions through finite element analysis and the LCC-S topology. Then use the built-in field calculator in Ansys Maxwell to export the magnetic field intensity data at each spatial position, and form a training sample data set after data cleaning and processing. The specific method is as follows: 1.

1. Construct a three-dimensional electromagnetic simulation model of the wireless power transmission system First, construct a three-dimensional electromagnetic simulation model of the wireless power transmission system in Ansys Maxwell, including the coil models at the transmitting end and receiving end, add current excitation to the coils, and obtain the self-inductance and mutual inductance values of the coils varying with the transmission distance; Construct the LCC-S circuit topology diagram of the wireless power transmission system. According to Kirchhoff's voltage law, obtain the current-voltage relationship formula, then there is: Among them is an AC power supply; L1 is the compensation inductor at the transmitting end, L p is the self-inductance of the transmitting coil, L s is the self-inductance of the receiving coil; C1, C p is the compensation capacitor at the transmitting end, C s is the compensation capacitor at the receiving end, M ps is the mutual inductance generated by the transmitting coil and the receiving coil; R eq is the equivalent load impedance; is the input current, is the current of the transmitting coil, is the current of the receiving coil; ω is the transmission frequency; When considering the maximum transmission efficiency of the wireless power transmission system, working in the resonant state, the inductive reactance and capacitive reactance cancel each other out, and the impedance presents a pure resistance characteristic, then there is: Substitute Equation (2) into (1) and simplify to obtain the relationship formula of the currents of each part: Substitute the mutual inductance, self-inductance, and different power supply voltage and load resistance values into the formula for calculation to obtain different current values at the transmitting end and receiving end. Then, respectively input the current values at the transmitting end and receiving end as excitations into the three-dimensional model of the wireless power transmission system, and obtain the magnetic induction intensity values at each spatial position under different working conditions through finite element analysis; 1.

2. Calculate the self-inductance, mutual inductance of the coils and the currents at the transmitting end and receiving end under different working conditions through finite element analysis and the LCC-S topology; First, extract the complex amplitude value of the magnetic induction intensity through the field calculator built into Ansys Maxwell. Subsequently, establish a one-to-one correspondence between the power supply voltage, load resistance, transmission distance, three-dimensional spatial coordinates, and the corresponding output of the magnetic induction intensity through Excel; 1.

3. After data cleaning and processing, a training sample data set is formed; Use the interquartile range method (IQR) to remove outliers from the magnetic field intensity data samples, removing outliers and noise to ensure the validity and integrity of the data; a. Calculate the first quartile (Q1) and the third quartile (Q3) of the electromagnetic field intensity data collected by the field calculator; b. Calculate the interquartile range (IQR), defined as: IQR = Q3 - Q1 (4) c. Set the rejection threshold to remove all data points outside the following range to ensure data quality: [Q1 - 1.5×IQR, Q3 + 1.5×IQR] (5).

3. The magnetic field prediction method for a wireless power transmission system based on an improved particle swarm optimization neural network algorithm according to claim 1, characterized in that Step 2, RBF neural network model construction: Use the transmission distance, power supply voltage, load resistance, and three-dimensional spatial coordinates as input features, and the magnetic field intensity of the wireless power transmission system as the output to construct an RBF neural network model. The specific method is as follows: 2.1 Center selection For each magnetic field strength sample, calculate the local density in the corresponding input feature space using a truncation distance d c to count the number of samples that satisfy ‖x i - x j ‖ ≤ d c to obtain: where 1() is the indicator function, when the sample x j and x i are no more than d c apart, the value is 1, otherwise 0; then calculate the minimum distance between each sample and the sample with higher density, that is: For the densest sample, let δ i be the maximum of all distances; Finally, calculate the index: γ i = ρ i × δ i (8) And select several points with the highest γ i as the clustering centers, and assign the remaining samples to the nearest center; 2.

2. RBF neuron width σ i Calculation For any center c i , first find the nearest neighbor c among all centers j and calculate the Euclidean distance: d ij = ||c i -c j || (9) Subsequently, calculate the width of the RBF neuron: σ i = β·d ij (10) where β is an adjustable hyperparameter; 2.

3. Calculation of the output layer weight Let x represent the input feature, then the output of the i-th mapping unit is: where c i is the center of the i-th neuron, and σ i is its width; Combine the responses of all samples into a matrix Orthogonally normalize each column of Φ step by step using the orthogonal least squares method, and successively select the column that contributes the most to the current residual to construct an upper triangular linear system to solve for the output layer weight W and calculate the output layer weight w i ; Finally, the prediction expression is where y(x) is the predicted magnetic field intensity of the wireless power transmission system..

4. The magnetic field prediction method for a wireless power transfer system based on an improved particle swarm optimization neural network algorithm according to claim 1, characterized in that, Step 3, Use the LHS optimization adaptive particle swarm algorithm to globally search for and optimize the neural network parameters: Generate uniformly distributed initial candidate solutions in the high-dimensional space of the RBF neural network parameters through Latin hypercube sampling, and use the APSO algorithm to dynamically adjust the search parameters to determine the optimal parameter combination of the RBF neural network model. The specific method is as follows: 3.

1. Use the center, width, and weight parameters of the RBF neural network model as particles to generate uniformly distributed initial candidate solutions in the high-dimensional space of the RBF neural network parameters through Latin hypercube sampling; Use Latin hypercube sampling to evenly divide the value range of each parameter dimension to be optimized and assign values to each particle; where is the initial position of the i-th particle in the d-th dimension; P max and P min are the upper and lower bounds of the d-th dimension respectively, and ρ i is a random number with a value range of [0, 1]; 3.

2. Use the APSO algorithm to dynamically adjust the search parameters to determine the optimal parameter combination of the RBF neural network model; In the D-dimensional search space, the current velocity and position of the particle are expressed as: a i = [a i,1 , a i,2 , …, a i,D (13)v i = [v i,1 , v i,2 ,..., v i,D (14) where i = 1, 2,..., s, s is the number of particle swarms, D = K(n + 2) is the vector space dimension, n is the input vector space dimension of the magnetic field intensity sample set, and K is the number of hidden layer neurons; The fitness of the particle is where y(t) is the predicted magnetic field strength, and y d (t) is the actually measured magnetic field strength, and T is the input characteristic sample of the wireless power transmission system; During the search process, each particle evaluates the quality of its current position according to the fitness function and records its own best position p i (t): p i (t) = [p i,1 (t), p i,2 (t), …, p i,D (t)](16) The optimal solution found by all particles in the entire particle swarm is the global optimum g(t): g(t) = [g1(t), g2(t), …, g D (t)](17) Based on the individual extreme value p i (t) and the global optimal position g(t) to update the velocity and position. The update formula for the velocity is as follows: v i,d (t + 1) = v i,d (t)ω i,d (t) + c1r1(p i,d (t) - a i,d (t)) + c2r2(g d (t) - a i,d (t))(18) where d = 1, 2, …, D; c1, c2 are learning factors; r1, r2 are random numbers in the range of [0, 1]; ω i,d is the inertia weight value; Define the diversity as: S(t) = f min (a(t)) / f max (a(t))(19) where f min (a(t)) and f max (a(t)) are the minimum and maximum values of the predicted magnetic field strength errors of all particles at time t; Based on the diversity, design a non-linear regression function: γ(t) = (l - S(t)) -1 (20) Where l is an initialization constant and l ≥ 2; Calculate the difference between the particle and the optimal particle: A i (t) = f(g(t)) / f(a i (t)) (21) where f(g(t)) is the global optimum, and f(a i (t)) is the fitness value of the i-th particle; Adaptively adjust the inertia weight, which is used to adjust the inertia weight in the velocity formula; ω i γ(t) = i γ(t)(A i (t) + c) (22) where ω i (t) is the inertial weight of the i-th particle at time t; c is a constant.

5. The magnetic field prediction method for a wireless power transmission system based on an improved particle swarm optimization neural network algorithm according to claim 1, characterized in that Step 4, Magnetic field intensity prediction of the wireless power transfer system: Embed the trained RBF neural network model with the optimal parameters retained into the prediction software, and perform online prediction based on the parameters of the wireless power transfer system under actual working conditions, output the corresponding magnetic field intensity distribution, and realize the real-time monitoring and early warning of the electromagnetic environment risk of the system. Among them, the specific method for realizing the real-time monitoring and early warning of the electromagnetic environment risk of the system is as follows: a. Region division Divide the entire monitoring area into several uniform grid cells according to a predetermined size. For example, each grid covers a certain area or a square area with a fixed side length; b. Statistical grid data For each grid cell, calculate the average field intensity of all points within the grid: Among which B i is the field strength of the i-th measurement point in this grid, and n is the number of points in this grid; c. Threshold judgment According to the ICNIRP or GB 8702-2014 standard, divide the prediction results with a magnetic field intensity of the safety value of 27 μT: i. Low risk level: The predicted value is lower than 10 μT, 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 μT and 20 μT, 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 27 μT, indicating a high risk, and immediate protective measures need to be taken or the exposure needs to be reduced.

6. A magnetic field prediction system for a wireless power transfer system based on an improved particle swarm optimization neural network algorithm, which implements the magnetic field prediction method for a wireless power transfer system based on an improved particle swarm optimization neural network algorithm according to any one of claims 1-5, and realizes the magnetic field prediction of a wireless power transfer system based on an improved particle swarm optimization neural network algorithm, and separately executes steps 1-4 in four modules.

7. A computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the magnetic field prediction method for a wireless power transfer system based on an improved particle swarm optimization neural network algorithm according to any one of claims 1-5, and realizes the magnetic field prediction of a wireless power transfer system based on an improved particle swarm optimization neural network algorithm.

8. A computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the magnetic field prediction method for a wireless power transfer system based on an improved particle swarm optimization neural network algorithm according to any one of claims 1-5, and realizes the magnetic field prediction of a wireless power transfer system based on an improved particle swarm optimization neural network algorithm.

Images