A composite reactive resonant shielding coil structure and a robust optimization method thereof

Abstract

The application is suitable for the field of electric vehicle wireless charging technology, and provides a composite reactive resonance shielding coil structure and a robust optimization method thereof. The composite reactive resonance shielding coil structure is composed of nanocrystals, aluminum plates and reactive resonance shielding coils, can weaken the leakage electromagnetic field of the electric vehicle wireless charging (EV-WPT) system, ensure high energy transmission efficiency and system lightweight, and enhance the offset tolerance and shielding effect. The robust optimization method combines the Attention-Unet uncertainty agent model with the multi-objective whale optimization algorithm (MOWOA), takes the system energy transmission efficiency and the maximum value of the induced electric field intensity in the human body as the optimization target, considers the uncertainty factors, and realizes the fast multi-objective optimization design. The application provides practical application significance and theoretical guidance for the electromagnetic shielding design of the EV-WPT system, and provides a new research idea for the safety and protection of human body electromagnetic exposure.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of electric vehicle wireless charging, and particularly relates to a composite reactive resonance shielding coil structure and a robust optimization method thereof. BACKGROUND

[0002] With the continuous development of the automobile industry towards intelligence and green, electric vehicles (EVs) as an important green transportation tool, its industry has gradually moved to the forefront of the world. In the power supply mode of electric vehicles, the traditional wired conduction charging has limitations such as inconvenient plugging of charging wires, easy damage of charging interfaces, and large occupation of charging facilities. In addition, the problems of limited range and inconvenient charging during long-distance travel also limit its use scenarios. Under this background, wireless charging (Wireless Power Transfer, WPT) technology has attracted widespread attention due to its non-contact power transmission advantages. This technology does not require manual plugging of charging wires, not only improving the user experience, but also avoiding safety hazards such as electric shock and short circuit that may occur during the plugging process of the charging wire, while it can work normally in bad weather, reducing the limitations of the charging scene. However, as a kind of loose coupling structure, the working principle of the electric vehicle wireless charging (EV-WPT) system is to realize power transmission through electromagnetic induction between the transmitting coil and the receiving coil. With the increasing demand for charging power, the leakage electromagnetic field dose generated by the system during operation also increases, and the electromagnetic field leaked by the coupling mechanism may pose an electromagnetic safety threat to the surrounding drivers and passengers. Related animal studies have shown that exposure to electromagnetic fields may have a tumor-promoting effect. For example, after mice treated with carcinogens were exposed to electromagnetic fields, the number of lung and liver tumors was significantly higher than that of the control group, and the incidence of breast tumors in mice exposed to low-frequency magnetic fields for a long time also increased. In addition, the position of the human body relative to the WPT device is not fixed (such as the position of the driver, passenger or passing person is not fixed), and the parking position may deviate due to insufficient algorithm accuracy during automatic parking, and these factors will change the internal electromagnetic exposure dose of the human body, further exacerbating the health risks to the human body, so ensuring the safety of human electromagnetic exposure has become a problem to be solved in the development of electric vehicle wireless charging technology.

[0003] To ensure the safety of WPT devices, the safety standards for human exposure to electromagnetic fields are indispensable. Some international organizations have carried out relevant research. For example, ICNIRP 2010 and IEEE C95.1 are both standards for the safety of human exposure to electromagnetic fields around EV-WPT systems. For the leakage magnetic field (LMF) of EV-WPT systems, various shielding measures have been proposed. At present, the shielding technology of EV-WPT systems mainly includes passive shielding and active shielding. Active shielding technology is less used due to its high cost, complex structure, and large space occupation. Passive shielding technology has become the most widely used method due to its characteristics. It is generally based on the structure and properties of the material itself for electromagnetic shielding. The material types are usually divided into ferromagnetic materials and metal conductor materials. Among them, the most common form is to attach ferrite plates and aluminum plates outside the transmitting coil and receiving coil. The magnetic conductivity of high magnetic permeability materials and the eddy current effect of high electrical conductivity are used to prevent the escape of LMF. However, ferromagnetic materials such as ferrite have the problems of heavy weight, strong fragility, and large changes in magnetic properties with temperature. Large-area coverage poses a challenge to the design of compact EV-WPT systems. Metal conductors generate internal eddy currents to offset the magnetic field, but the closer they are to the system, the more obvious the system efficiency will decline. Some researchers have also tried to use metamaterials to construct the shielding structure of EV-WPT systems. However, metamaterials can only achieve shielding effect at a specific frequency, and their shielding performance will quickly decline with the change of system operating frequency, which limits their application in the field of electromagnetic shielding of EV-WPT systems. In addition, various studies have proposed composite structures of ferrite, which can effectively control the magnetic flux and improve the efficiency, but they are still limited by the defects of ferrite itself. Nanocrystalline is considered as a substitute for ferrite, with superior magnetic and mechanical properties. Related research uses nanocrystalline magnetic cores to design magnetic conductive structures, proving that it can achieve higher efficiency and lower leakage electromagnetic field. However, its high cost limits its extensive use, so it is difficult to achieve ideal electromagnetic shielding by laying only these materials.

[0004] The resonant reactive shielding technology uses the energy of the system itself to generate a counteracting magnetic field, saves an external power supply, has good shielding effect, has low influence on system efficiency, and can also improve the lightweight of the WPT device. Related research includes placing a passive shielding coil outside the system to solve the insulation problem of directly connecting the power transmission coil and the shielding coil, or using a capacitor to compensate for the passive shielding coil to achieve resonant reactive shielding and change the equivalent inductance. However, these researches do not consider the uncertainty factors in actual applications, which affects the shielding effect in complex scenarios. The design method that takes into account the inevitable uncertainty factors in the optimization design is called robust design optimization (RDO). In real applications, such as parking position deviation and changes in the distance between the human body and the EV-WPT, uncertainty factors will affect the electromagnetic exposure dose of the human body. However, existing research has not fully integrated RDO into the design of shielding structures, making it difficult to cope with complex situations in actual scenarios.

[0005] In related research on the design of EV-WPT systems, most researchers choose to establish a shielding structure through the finite element method (FEM) to consider both transmission efficiency and shielding effect, and combine it with a multi-objective optimization algorithm. However, the FEM simulation software has high calculation cost, which leads to an increase in the time cost of each optimization iteration. Although some research has proposed a sparse chaos polynomial expansion (PCE) method to quantify the influence of uncertainty for optimization of transmission efficiency, PCE has a "dimension disaster" problem, which also has high calculation cost. Moreover, the optimization design of existing electromagnetic shielding structures is mostly based on deterministic states and does not fully analyze RDO, making it difficult to ensure electromagnetic safety in actual applications. Therefore, the present invention proposes a composite reactive resonant shielding coil structure and a robust optimization method. SUMMARY

[0006] The present invention aims to provide a composite reactive resonant shielding coil structure and a robust optimization method, which aims to solve the problems raised in the background art.

[0007] The object of the present invention is achieved by the following technical solutions:

[0008] A composite reactive resonant shielding coil structure is applied to an electric vehicle wireless charging system, which includes a reactive resonant shielding coil, a nanocrystalline layer, and an aluminum plate. The nanocrystalline layer is pasted on the periphery of the reactive resonant shielding coil, and the aluminum plate is pasted on the nanocrystalline layer and located outside the reactive resonant shielding coil. The reactive resonant shielding coil generates a reverse counteracting magnetic field through a leakage magnetic field to weaken the electromagnetic field intensity in the non-working area of the electric vehicle wireless charging system. The nanocrystalline layer and the aluminum plate are used to attract the leaked electromagnetic field to the periphery of the reactive resonant shielding coil through high magnetic permeability characteristics, to increase the reverse magnetic field intensity and reduce the external magnetic field intensity, thereby reducing the electromagnetic exposure dose of the human body.

[0009] Further, the relative magnetic permeability of the nanocrystalline layer under the working condition of room temperature and 85 kHz is 22000.

[0010] Further, the reactive resonance shielding coils are four, including two transverse reactive resonance shielding coils and two longitudinal reactive resonance shielding coils, which are symmetrically arranged on the transverse sides and longitudinal sides of the receiving coil respectively; the transverse reactive resonance shielding coils and the longitudinal reactive resonance shielding coils are both connected with compensation capacitors.

[0011] Further, the distance between the transverse reactive resonance shielding coil and the edge of the receiving coil is 0-0.05m; the distance between the longitudinal reactive resonance shielding coil and the edge of the receiving coil is 0-0.05m; the compensation capacitor of the transverse reactive resonance shielding coil is 500-1500nF; and the compensation capacitor of the longitudinal reactive resonance shielding coil is 500-1500nF.

[0012] A robust optimization method of the composite reactive resonance shielding coil structure is provided, which comprises:

[0013] An Attention-Unet uncertainty agent model is constructed, in which design variables and uncertainty factors are taken as inputs, and system energy transmission efficiency, maximum induced electric field intensity of human body, maximum induced electric field intensity of brain and maximum induced electric field intensity of lung are taken as outputs; the design variables include the distance between the transverse reactive resonance shielding coil and the edge of the receiving coil, the distance between the longitudinal reactive resonance shielding coil and the edge of the receiving coil, and the compensation capacitors of the transverse and longitudinal reactive resonance shielding coils; and the uncertainty factors include human body transverse offset, human body longitudinal offset, vehicle transverse offset, vehicle longitudinal offset, distance between transmitting coil and receiving coil and human body orientation deflection angle.

[0014] Based on the output of the Attention-Unet uncertainty agent model, a multi-objective whale optimization algorithm is used for multi-objective optimization to solve the Pareto front and obtain the robust optimal parameters (i.e. the optimal distance between the transverse reactive resonance shielding coil and the edge of the receiving coil, the optimal distance between the longitudinal reactive resonance shielding coil and the edge of the receiving coil, and the optimal compensation capacitors of the transverse and longitudinal reactive resonance shielding coils).

[0015] Further, the core framework of the Attention-Unet uncertainty agent model is an Attention-Unet network, which is formed by introducing an Attention Gate mechanism based on a U-Net network; the U-Net network is a full convolutional deep network model based on an encoder-decoder.

[0016] The input data is normalized and then enters the Attention-Unet network.

[0017] First through the encoder, the encoder consists of convolutional layer and down-sampling layer; convolution operation automatically extracts local spatial or sequence features in the input data, and a ReLU activation function is connected after convolution; after passing through the convolutional layer, the down-sampling layer is used to reduce the dimension of the feature map, and the most important features are reserved; after processing by the down-sampling layer, it enters the Attention Gate, which controls the transfer of the encoder features to the decoder; the decoding path first passes through the up-sampling layer, then splices the output of the Attention Gate, and finally passes through the convolution and ReLU activation, and then maps to the output dimension through a convolution to obtain MaxPool Slide by window, keep the maximum value in each window, reduce dimension step by step to compress information and keep the most critical features; after processing by the down-sampling layer, it enters the Attention Gate, which controls the transfer of the encoder features to the decoder; the decoding path first passes through the up-sampling layer, then splices the output of the Attention Gate, and finally passes through the convolution and ReLU activation, and then maps to the output dimension through a convolution to obtain The set of :

[0018] ;

[0019] In the formula, is the prediction output set of the model; W final is the final convolution kernel parameter; is the feature input before entering the convolution mapping; p is the bias.

[0020] Further, the multi-objective whale optimization algorithm updates the population position through three stages of searching for prey, surrounding prey and bubble net attack, and the position update formula includes:

[0021] Search for prey stage:

[0022] ; ;

[0023] In the formula, X ( t +1) is the updated whale position; X r ( t ) is the randomly selected whale position of the population; A and C are coefficient vectors; D is the distance between the whale individual and the optimal solution; t is the current iteration number; X ( t ) is the current whale position;

[0024] Surrounding prey stage:

[0025] ; ;

[0026] In the formula, zis a constant defining the helix shape; X z ( t ) is the current optimal whale position;

[0027] Bubble-net attack phase:

[0028] ; ;

[0029] wherein, l is a random floating point number on the interval (-1, 1).

[0030] Compared with the prior art, the present application has the beneficial effects that:

[0031] The composite reactive resonant shielding coil structure is composed of nanocrystals, aluminum plates and reactive resonant shielding coils, which can ensure high energy transmission efficiency while realizing system lightweight, effectively weaken the leakage electromagnetic field of the EV-WPT system, enhance the system offset tolerance and shielding effect, and provide a safer and more reliable electromagnetic environment for the surrounding users. The robust optimization method combines the Attention-Unet uncertainty agent model with the multi-objective whale optimization algorithm (MOWOA), takes the system energy transmission efficiency and the maximum value of the induced electric field intensity in the human body as the optimization target, considers the uncertainty factors, realizes fast multi-objective optimization design, improves the robustness of the composite reactive resonant shielding coil structure, and makes the system energy transmission efficiency and the human body electromagnetic exposure dose more stable and safe under the influence of uncertainty. Simulation experiments show that, under the consideration of uncertainty factors, the mean value of the maximum value of the induced electric field intensity in the human body E max is reduced by 48.38%, the mean value of the maximum value of the induced electric field intensity in the lung E max is reduced by 41.73%, the mean value of the energy transmission efficiency of the WPT system is improved by 6.46%, and the probability of the maximum value of the induced electric field intensity in the human body exceeding the standard limit is reduced to 0%, which proves the effectiveness of the composite reactive resonant shielding coil structure and its design method. The present application provides practical application significance and theoretical guidance for electromagnetic shielding design of the EV-WPT system, and provides a new research idea for human body electromagnetic exposure safety and protection. BRIEF DESCRIPTION OF DRAWINGS

[0032] Figure 1 is a schematic diagram of an EV-WPT system.

[0033] Figure 2 is a schematic diagram of a composite reactive resonant shielding coil structure.

[0034] Figure 3 is an equivalent circuit diagram of an EV-WPT system.

[0035] Figure 4 For the position relationship between the receiving coil and the reactive resonance shielding coil.

[0036] Figure 5 For the Attention-Unet network structure.

[0037] Figure 6 For the overall structure of the system; wherein (a) is the exposure scene, (b) is the human body and electric vehicle offset.

[0038] Figure 7 For the human body simulation model; wherein (a) is the human body trunk, (b) is the organ.

[0039] Figure 8 For the comparison of the results of each method (Attention-Unet uncertainty agent model, FEM simulation, traditional BP neural network and GRU network); wherein (a) is the system energy transmission efficiency η , (b) is the lung E max , (c) is the brain E max , (d) is the human body E max .

[0040] Figure 9 For the Pareto frontier diagram.

[0041] Figure 10 For the magnetic field distribution of the EV-WPT system; wherein (a) is the magnetic field distribution of the EV-WPT system before shielding, (b) is the magnetic field distribution of the EV-WPT system after shielding.

[0042] Figure 11 For the electric field distribution of the human body trunk; wherein (a) is the electric field distribution of the human body trunk before shielding, (b) is the electric field distribution of the human body trunk after shielding.

[0043] Figure 12 For the electric field distribution of the lung; wherein (a) is the electric field distribution of the lung before shielding, (b) is the electric field distribution of the lung after shielding.

[0044] Figure 13 For the probability distribution function (PDF); wherein (a) is the PDF of the system energy transmission efficiency η , (b) is the PDF of the lung E max , (c) is the PDF of the human body E max , (d) is the PDF of the brain E max .

[0045] Figure 14 The results are from MOAT; where (a) represents the effect of input variables on the system energy transfer efficiency. η The degree of influence, (b) is the effect of input variables on the lungs E max The degree of influence, (c) represents the impact of the input variable on the human body. E max The degree of influence, (d) is the effect of the input variable on the brain. E max The extent of the impact. Detailed Implementation

[0046] In order to provide a clearer understanding of the technical features, objectives and beneficial effects of the present invention, the technical solution of the present invention will now be described in detail below, but it should not be construed as limiting the scope of implementation of the present invention.

[0047] This invention provides a composite reactive resonant shielded coil structure and its robust optimization method, including:

[0048] 1. Composite reactive resonant shielded coil structure and its equivalent circuit;

[0049] The EV-WPT system is based on the principle of magnetic coupling resonance, using an electromagnetic field as a medium, and transmitting through a transmitting coil. Tx With receiving coil Rx Wireless power transmission is achieved through resonance between the transmitting coil and the electromagnetic field. First, AC power is input from the mains, converted to DC by the transmitter's rectifier circuit, and then converted back to AC by a high-frequency inverter circuit. The output frequency of the inverter circuit is the resonant frequency of the transmitting resonant circuit. At this point, the transmitting circuit operates in a resonant state under the action of the transmitter's compensation circuit, exhibiting purely resistive behavior. According to Faraday's law of electromagnetic induction, the transmitting coil... Tx A high-frequency electromagnetic field with a resonant frequency is excited into space, and is received by a receiving coil. Rx During reception, under the action of the receiving end compensation circuit, the receiving end circuit also operates in a resonant state, exhibiting pure resistivity. The receiving end rectifier circuit and chopper circuit convert high-frequency electrical energy into DC power required by the load. To ensure maximum efficiency of the EV-WPT system during normal operation, this invention employs a bilateral SS compensation circuit. An EV-WPT system with a reactive resonant circuit is as follows... Figure 1 As shown.

[0050] in, U S As a source of motivation, C 1 is the transmitting coil Tx Series compensation capacitor at the end, L 1 is the transmitting coil Tx The coil self-inductance at the end, L 2 is the receiving coil RxThe coil self-inductance of the transmitting coil, L 3 The coil self-inductance of the reactive resonance shielding coil, M 12 The coil mutual-inductance between the transmitting coil Tx and the receiving coil, Rx M 13 The coil mutual-inductance between the transmitting coil Tx and the reactive resonance shielding coil, M 23 The coil mutual-inductance between the receiving coil Rx and the reactive resonance shielding coil, C 2 The coil self-inductance of the receiving coil Rx The series compensation capacitance of the transmitting coil, R L The load resistance, U 1 The transmitting coil voltage, U 2 The receiving coil voltage, C 3 The reactive resonance shielding coil compensation capacitance. Generally, the EV-WPT system only considers the coil equivalent internal resistance. In the system in the resonant state, each element satisfies the following relationship:

[0051] Formula 1: ;

[0052] In the formula, ω is the angular frequency.

[0053] In the EV-WPT system, the high-frequency coupling electromagnetic field around the coupling coil is the core to realize the energy transmission. The current generated by the leakage magnetic field of the reactive resonance shielding coil can generate a reverse cancellation magnetic field, thereby weakening the electromagnetic field in the non-working area. For the EV-WPT system, the transmission of electromagnetic waves in the shielding structure directly determines the shielding effect of the shielding structure. The traditional reactive resonance shielding coil has good shielding effect in the coil center and the internal area, but the shielding effect around the coil boundary and the external area is poor. Because the magnetic force line is closed and cannot be truncated, only the guided magnetic force line can be used to reduce the magnetic field escaping to the outside of the transmission mechanism, so as to realize the shielding of the magnetic field. In the magnetic field, there is a relationship:

[0054] Formula 2: ;

[0055] In the formula, represents the magnetic flux, F m represents the magnetic motive force, R m represents the magnetic resistance. In order to overcome the above problems, the present application puts forward a kind of composite reactive resonance shielding coil structure (see Figure 2 ​), which is composed of a reactive resonance shielding coil, a nanocrystalline layer and an aluminum plate. The nanocrystalline layer is pasted on the periphery of the reactive resonance shielding coil, and the aluminum plate is pasted on the nanocrystalline layer and located outside the reactive resonance shielding coil. The reactive resonance shielding coil is wound by copper wire, and the copper material has the characteristics of good electrical conductivity and low material price. The leakage electromagnetic field outside the reactive resonance shielding coil is weakened by the high magnetic permeability ultra-thin nanocrystalline layer and the aluminum plate. First, the nanocrystalline layer converges the magnetic field to guide a part of the magnetic field in the non-working area into the working area; then the aluminum plate shields the other leakage electromagnetic field, and reduces the electromagnetic field strength in the non-working area through eddy current effect. Under the working condition of room temperature and 85 kHz, the relative magnetic permeability of the nanocrystalline layer can reach 22000. Under the same condition, the relative magnetic permeability of the traditional magnetic material ferrite can only reach 3300. The nanocrystalline layer layer attracts the leakage electromagnetic field around the original reactive resonance shielding coil to the coil, further improves the reverse magnetic field strength and reduces the external magnetic field strength, thereby achieving better shielding effect and reducing the human body electromagnetic exposure dose.

[0056] For convenience of calculation, the power supply and the conversion circuit are respectively equivalent to two voltage sources, and the equivalent internal resistance of each coil is too small to be ignored, so Figure 1 can be further simplified as Figure 3 the equivalent circuit shown. Among them, the inductance L 3 and the compensation capacitor C 3 are simplified as an equivalent inductance L eq . The formula is as follows:

[0057] Formula 3: ;

[0058] The loop voltage equation is:

[0059] Formula 4: ;

[0060] In the formula, is the loop voltage source of the transmitting coil; is the loop voltage source of the receiving coil; j is the imaginary unit; is the loop current of the transmitting coil; is the loop current of the receiving coil; is the loop current of the reactive resonance shielding coil.

[0061] When the reactive resonance shielding coil is working, let M 13 = αM 12 , M 23 = βM 12 , wherein αand β are all proportional parameters, whose purpose is to scale M 13 and M 23 are all proportional parameters, whose purpose is to scale M 12 are all proportional parameters, whose purpose is to scale

[0062] Equation 5: ;

[0063] From equation 5, if α and β are very small, the coupling effect between the reactive resonance shielding coil and the transmitting and receiving coils is very small, and it can be obtained that the original system is not affected after the reactive resonance shielding coil works, and can be regarded as constants. The reactive resonance circuit only generates current by the action of the leakage magnetic field. Therefore, equation 4 can be simplified to equation 6:

[0064] Equation 6: ;

[0065] According to the Biot-Savart law, the magnetic field strength generated by the current in the coil is proportional to the current. Therefore, the shielding effect is determined by . The size of is mainly determined by the mutual inductance M 13 , M 23 and the equivalent inductance L eq . The size of M 13 , M 23 and L eq is determined by the structure of the coil itself, the distance d between the reactive resonance shielding coil and the EV-WPT system, and the compensation capacitor C 3. Once the structure of the coil is determined, the shielding effect of the reactive resonance shielding coil is only determined by the distance d between the reactive resonance shielding coil and the EV-WPT system, and the compensation capacitor C 3.

[0066] However, in practice, only one equivalent inductance L eqYes, it exists. When all coupling mechanisms except the reactive resonant shielding coil are determined, the aforementioned parameters are closely related to the number of coil turns, coil size, and compensation capacitor parameters. Therefore, finding the optimal parameter combination through numerous trials is unrealistic. Furthermore, due to medical ethics concerns, current research on human electromagnetic exposure mainly relies on FEM simulation software. However, due to the fine meshing of simulation software, the time cost of each calculation is enormous. Therefore, this invention proposes a surrogate model based on Attention-Unet, which can effectively shorten the computation time cost, generalize the human electromagnetic exposure scenario with relatively less data, and obtain accurate results. In practical applications, various uncertainties are unavoidable. This requires the composite reactive resonant shielding coil structure designed in this invention to enhance its tolerance to various uncertainties, reduce the electromagnetic exposure dose to the human body around the EV-WPT system, and thus have greater practical application value.

[0067] The design variables and their distribution characteristics, as well as the uncertainty factors and their distribution characteristics of the reactive resonant shielded coil proposed in this invention, are shown in Tables 1 and 2, respectively. This is because the EV-WPT system receiving coil... Rx The structure is symmetrical, and the positions of the four reactive resonant shielding coils are also symmetrical. Therefore, the parameters of the two horizontal reactive resonant shielding coils and the two vertical reactive resonant shielding coils should be the same. The positional relationship between the receiving coil and the reactive resonant shielding coils is as follows: Figure 4 As shown in the figure. Here, U represents a uniform distribution. Since the offset between the coils in the EV-WPT system is equally likely as the offset of the human body in all directions, it is usually considered a uniform distribution in engineering applications.

[0068] Table 1 Design variables and their distribution

[0069]

[0070] Table 2. Uncertainty Factors and Their Distribution

[0071]

[0072] 2. A robust optimization method based on the Attention-Unet uncertain surrogate model;

[0073] The U-Net network is a fully convolutional deep network model based on an encoder-decoder architecture. It enables efficient learning using a small number of labeled samples in processing one-dimensional high-order data. The encoder extracts features through convolution and downsampling, deeply extracting high-dimensional information; the decoder reconstructs local information and restores the original data structure through upsampling and skip connections. The combination of these two approaches achieves multi-scale feature extraction. Furthermore, this invention introduces the Attention Gate (AG) mechanism on top of the U-Net network, forming the Attention-Unet network. This mechanism can calculate weights for each channel or skip connection, strengthening important features, suppressing redundant features, and dynamically adjusting the model's focus. Moreover, addressing the overfitting problem common in traditional U-Net regression networks, the Attention Gate mechanism automatically sparsifies feature representations, enhancing the model's generalization ability, exhibiting stronger robustness to noise or irrelevant features, and indicating the variables the model focuses on.

[0074] This invention uses the Attention-Unet network as the core framework to construct the Attention-Unet uncertainty surrogate model, and takes the design variables of the reactive resonant shielded coil in Table 1 as input variables. X =[ x 1, x 2,..., x N ],in N This refers to the dimensions of the optimization variables. Since there are a total of 4 optimization variables in the table, therefore... N =4. Since the ICNIRP 2010 guidelines explicitly state that the induced electric field strength is a limit for the human body and brain, and considering the lungs' proximity to the EV-WPT system, their large size, and existing research indicating that exposure to extremely low-frequency magnetic fields may lead to symptoms such as decreased respiratory rate and prolonged respiratory cycle, this invention also considers the electromagnetic exposure dose to the human lungs. In summary, this invention uses simulation to measure the system's energy transfer efficiency. η The maximum induced electric field strength in the human body, brain, and lungs. E max As the regression variable in the model, let it be denoted as Y =[ y 1, y 2, y 3, y 4], of which y 1, y 2, y 3, y 4 represents the system energy transfer efficiency. η The maximum value of the induced electric field intensity of the human bodyE max The maximum value of the induced electric field intensity in the brain E max Maximum value of induced electric field intensity in the lungs E max Input data is normalized to improve model accuracy and enhance the model's ability to generalize to new data.

[0075] Formula 7: ;

[0076] Formula 8: ;

[0077] In the formula, For the first n The first set of data i Data normalization results; For the first n The first set of data i One belongs to X The original data for the category; For the first i indivual X A collection of class data; The value is the normalized value; For the first n The first group of data j One set of raw data; For the first j indivual Y A collection of class data; ε It is a very small constant used to prevent the denominator from being 0; max and min represent the maximum and minimum value functions, respectively.

[0078] The input data is normalized before entering the Attention-Unet network, whose specific structure is as follows: Figure 5 As shown. First, the data passes through an encoder, which consists of convolutional layers and downsampling layers. The convolutional operation automatically extracts local spatial or sequence features from the input data. For the one-dimensional data in this invention, it's equivalent to sliding a small window to extract pattern fragments, thereby enhancing nonlinear expressive power. Furthermore, a ReLU activation function is applied after convolution, as shown in the following equation:

[0079] Formula 9: ;

[0080] In the formula, For the first l The output features of the layer; Represents the ReLU activation function; For the first l Layer weights; * indicates a one-dimensional convolution operation; For the first lOutput feature of the 1st layer; Bias of the 1st layer. After the convolutional layer, the down-sampling layer compresses the information step by step by reducing the dimension, but still retains the most critical features. l Bias of the 1st layer. After the convolutional layer, the down-sampling layer compresses the information step by step by reducing the dimension, but still retains the most critical features. MaxPool Sliding window, keeping the maximum value in each window, reducing dimension step by step compresses information, but still retains the most critical features.

[0081] Formula 10: ;

[0082] In the formula, Feature of the 1st layer after down-sampling. l Feature of the 1st layer after down-sampling.

[0083] After processing by the down-sampling layer, enter the Attention Gate, which controls which features of the Encoder are passed to the Decoder. Let m Feature of the Encoder, g Gate signal of the Decoder, then the attention coefficient τ As follows:

[0084] Formula 11: ;

[0085] In the formula, W x Feature weight matrix of the Encoder; x Output feature of the Encoder; W g Gate signal weight matrix of the Decoder; p Bias.

[0086] Then the output of the Attention Gate is as follows:

[0087] Formula 12: ;

[0088] In the formula, x gated Output of the Attention Gate.

[0089] The decoding path first passes through the up-sampling layer:

[0090] Formula 13: ;

[0091] In the formula, Output feature of the 1st layer after up-sampling; l One-dimensional transpose convolution; ConvTransposelD Output feature of the 1st layer after up-sampling; Output feature of the 1st layer after up-sampling; l Output feature of the 1st layer after up-sampling.

[0092] Then splice with the output of the Attention Gate:

[0093] Formula 14: ;

[0094] In the formula, Concat The splicing operation combines features from different sources to enrich feature information.

[0095] Finally, through convolution and ReLU activation:

[0096] Formula 15: ;

[0097] In the formula, BN Batch normalization is represented; ConvlD It is a one-dimensional convolution operation.

[0098] Then, through a convolution mapping to the output dimension, the set of is obtained, that is, .

[0099] Formula 16: ;

[0100] In the formula, The set of predicted outputs of the model; W final The final convolution kernel parameter; The feature input before entering the convolution mapping; p The bias.

[0101] On the one hand, the present application can obtain uncertainty proxy models conforming to various scenes by taking uncertainty factors as input variables. On the other hand, the Attention Gate mechanism is integrated into the U-Net to realize spatial features at different positions and improve the fitting ability of the model. The pseudo code of the Attention-Unet uncertainty proxy model method is shown in Algorithm 1. The Attention-Unet uncertainty proxy model method proposed in the present application can reduce the number of training groups and quickly obtain an uncertainty proxy model to assist in multi-objective robust optimization of a reactive resonance shielding coil.

[0102]

[0103] In addition, the present application adopts MOWOA (Multi-Objective Whale Optimization Algorithm) for robust optimization design. As a meta-heuristic algorithm, MOWOA searches for optimal solutions by simulating the hunting behavior of whales, can effectively search in the entire solution space, and has fast convergence speed; at the same time, it balances the weights of each target through chaotic mapping and adaptive weight mechanism, and has high optimization efficiency due to its simple structure and fewer parameters. Its optimization process mainly consists of three stages, as follows:

[0104] (1) Searching for prey: At the beginning of the algorithm, the whale individual searches for the potential position of the prey according to the fitness of its own position and the surrounding environment, as shown in the following formula:

[0105] Formula 17: ;

[0106] Formula 18: ;

[0107] In the formula, X ( t +1) is the updated whale position; X r ( t ) is the randomly selected whale position of the population; A and C are the coefficient vectors; D is the distance between the whale individual and the optimal solution; t is the current iteration number; X ( t ) is the current whale position.

[0108] (2) Surrounding prey: In the middle of the algorithm, the whale individual will swim towards the prey position (better solution). In this process, the whale will form a bubble net underwater, then quickly swim to the water surface to capture the prey, as shown in the following formula:

[0109] Formula 19: ;

[0110] Formula 20: ;

[0111] In the formula, z is a constant that defines the spiral shape, X z ( t ) is the optimal whale position obtained so far.

[0112] (3) Bubble net attack: In the later stage of the algorithm, the whale captures the prey through a specific foraging behavior (bubble net attack), as shown in the following formula:

[0113] Formula 21: ;

[0114] Formula 22: ;

[0115] In the formula, l is a random floating-point number in the interval (-1, 1).

[0116] MOWOA realizes efficient search for the optimal solution through the above three stages, and is especially suitable for multi-objective optimization scenarios. First, the population is initialized and evaluated, and then the Attention-Unet uncertainty surrogate model is embedded as the objective function in the optimization process, and through the search prey, surround prey and bubble net attack of MOWOA, better population individuals are selected, so as to output the Pareto front, and finally the robust optimal structure of the composite reactive resonance shielding coil is obtained. The present application combines the Attention-Unet uncertainty surrogate model with MOWOA to realize efficient and accurate robust optimization design of the structure of the composite reactive resonance shielding coil.

[0117] In order to realize efficient robust optimization design of the reactive resonance shielding coil in the EV-WPT system and protect the safety of human electromagnetic exposure, the present application uses the surrogate model based on the Attention-Unet method as the optimization objective function of MOWOA. The MOWOA pseudo code is shown in Algorithm 2. Wherein, p is a random number.

[0118]

[0119] The specific implementation of the present application is described in detail in combination with specific embodiments.

[0120] Embodiment 1: Simulation and optimization verification of the composite reactive resonance shielding coil of the EV-WPT system;

[0121] I. Fine simulation model building

[0122] The present application establishes a simulation model of the EV-WPT system and the composite reactive resonance shielding coil, and the specific parameters are as follows: the vehicle size is 4m x 2m x 1.5m (consistent with the size of a family car); the transmitting coil Tx size is 0.32m x 0.32m, the receiving coil Rx size is 0.65m x 0.5m, both are 18 turns, both are Litz coils, and the wire diameter is 0.007m; the system operating frequency is 85kHz, and the operating power is 11kW. Since the receiving coil Rx is fixed on the chassis of the electric vehicle, the composite reactive resonance shielding coil is placed in four directions of the receiving coil Rx to shield the LMF in each direction. The human exposure scenario and uncertainty factor labeling are shown in Figure 6 (a) and (b). In addition, the present application establishes a fine human body simulation model of a human body and part of the organs (brain, lung) which is 1.75m high (the height of an ordinary male), as shown in Figure 7 (a) and (b).

[0123] The simulation model established by the present application is based on an Intel 13900K processor, and each calculation takes 11 minutes and 45 seconds to obtain the corresponding simulation results. Based on the original data of FEM simulation, the Attention-Unet uncertainty surrogate model is established using the parameters in Table 1 and Table 2, and the model output is the system energy transmission efficiency η , the maximum value of the induced electric field intensity of three parts of the human body (lungs, human body, brain) E max .

[0124] II. Verification of the Attention-Unet uncertainty surrogate model;

[0125] In order to verify the prediction accuracy of the Attention-Unet uncertainty surrogate model, 30 groups of test data are selected from the parameters in Table 1 and Table 2. When using the Attention-Unet uncertainty surrogate model, FEM simulation, traditional back propagation (BP) neural network and gated recurrent unit (GRU) network are used as control groups to calculate the η , and the maximum value of the induced electric field intensity of three parts of the human body E max . The comparison results are shown in Figure 8 (a)-(d), and the prediction performance is measured by the mean absolute error (MAE), as shown in Table 3.

[0126] Table 3 MAE of different methods

[0127]

[0128] Figure 8 The data in Table 3 shows that the Attention-Unet uncertainty surrogate model is consistent with the results of FEM simulation, while the fitting performance of BP neural network and GRU neural network is significantly worse. In the Attention-Unet part of Table 3, the MAE of the system energy transmission efficiency η is 0.0031, which proves that the Attention-Unet uncertainty surrogate model has accurate prediction performance. On the other hand, the MAE of the maximum value of the induced electric field intensity of three parts of the human body E max is 0.0112, 0.0514 and 0.0113 respectively, which is slightly higher than the MAE of the system energy transmission efficiency η , because there is strong nonlinearity and complexity between the LMF value and the uncertainty factor, but this is still acceptable accuracy. Therefore, the Attention-Unet uncertainty surrogate model established by the present application can replace FEM simulation as the objective function of robust optimization design, realizing the equivalent replacement of simulation.

[0129] III. Robust optimization based on Attention-Unet uncertainty surrogate model

[0130] According to the optimization parameters set in Table 1 and the uncertainty factors in Table 2, the tested Attention-Unet uncertainty surrogate model is used as the objective function of optimization, and the Pareto front of the optimization result is obtained as shown in Figure 9 The robust optimal parameters of the composite reactive resonance shielding coil obtained by the two methods (Attention-Unet, FEM) and the calculation time of the two methods are shown in Table 4.

[0131] Table 4 Robust optimal parameters and calculation time

[0132]

[0133] The data in Table 4 shows that the robust optimization design method based on the Attention-Unet uncertainty surrogate model can reduce the calculation time by 80.06% compared to 10,000 times of FEM simulation while ensuring the calculation accuracy. It can be seen that the method of the present application can improve the calculation efficiency of the optimization process and significantly reduce the time cost of the optimization design.

[0134] IV. Effectiveness verification of the composite reactive resonance shielding coil

[0135] In order to verify the effectiveness of the robust optimal structure and design method of the composite reactive resonance shielding coil obtained in the foregoing work, and considering the actual human electromagnetic exposure scenario, the present application uses the electromagnetic field intensity of three parts of the human body E max as the measurement standard of the shielding effect of the composite reactive resonance shielding coil. According to the ICNIRP 2010, the limit value of the public induced electric field intensity E lim The calculation formula is:

[0136] Equation 23: ;

[0137] In the formula, f is the frequency of the electromagnetic field. The working frequency of the EV-WPT system of the present application is 85 kHz, so E lim = 11.475 V / m.

[0138] The magnetic flux density distribution around the EV-WPT system before and after placing the composite reactive resonance shielding coil is shown in Figure 10 Figure 10 ​The magnetic flux density distribution without shielding is shown in Fig. 2(a), which shows that the magnetic flux density is high in the middle region (red region) and gradually decreases towards the periphery, and the magnetic flux density at the edge decays relatively slowly; Figure 10 The magnetic flux density distribution with shielding is shown in Fig. 2(b), which shows that the high magnetic flux density region (red region) is significantly reduced, the magnetic flux density at the edge decreases more significantly, and the overall magnetic flux density distribution is more concentrated and the external leakage is reduced. Taking the induced electric field intensity in human tissue as the analysis object, the induced electric field intensity in human tissue before and after shielding is compared, as shown in Fig. 3(a) and (b). Figure 11 Figure 12 As shown in Fig. 2(a) and (b), the induced electric field distribution is more obvious in the leg (belonging to the human body trunk) closest to the EV-WPT system. The induced electric field intensity in the three parts of the human body (lungs, human body trunk, and brain) before and after shielding is shown in Fig. 3(a) and (b). E max As shown in Table 5.

[0139] Table 5 Shielding effect and system energy transmission efficiency before and after shielding

[0140]

[0141] Table 5 data shows that before shielding, the induced electric field intensity in the human body trunk E max exceeds the standard, the induced electric field intensity in the brain E max does not exceed the standard; after shielding, the induced electric field intensity in the human body trunk E max decreases by 61.72%, the induced electric field intensity in the brain E max decreases by 36.19%, and the induced electric field intensity in the lungs E max decreases by 40.39%, all of which meet the standard limit. In addition, through the compensation of the composite reactive resonance shielding coil, the system energy transmission efficiency η increases by 11.79%.

[0142] However, in actual application, whether it is the position of the vehicle or the position of the human body, the error cannot be avoided, which will lead to uncertainty in the electromagnetic exposure dose of the human body. According to the robust optimal structure of the aforementioned composite reactive resonance shielding coil and the distribution of uncertainty factors, the probability distribution functions (PDFs) of the induced electric field intensity in the human body trunk E max 、 brain E max and lungs E max are calculated respectively using the Attention-Unet uncertainty proxy model, as shown in Fig. 4(a)-(d), and the statistical characteristics are shown in Table 6. Figure 13 ​​

[0143] Table 6 Statistical characteristics

[0144]

[0145] Figure 13 As shown in the data in Table 6, under the influence of uncertainty factors, if there is no composite reactive resonance shielding coil, E max there is a 43.09% probability of exceeding the standard limit, at which time the human body trunk E max The average is 12.67V / m. After shielding, even under the influence of uncertainty factors, the human body trunk E max will not exceed the standard limit, the average is 6.54V / m, and the system energy transmission efficiency η The average is improved by 6.46%, E max The probability of exceeding the standard limit is reduced to 0%. As can be seen, in the case of considering uncertainty factors, the composite reactive resonance shielding coil can still reduce the average of the human body trunk E max by 48.38%, and the average of the lung E max by 41.73%, which proves the effectiveness of the composite reactive resonance shielding coil proposed in the present application.

[0146] Five, influence analysis of input parameters (MOAT method);

[0147] In order to qualitatively measure the influence degree of input parameters, and screen out the input variables that have greater influence on the above four indexes (system energy transmission efficiency η , human body trunk E max , brain E max , lung E max ), the present application adopts the MOAT (Morris-One-at-a-Time) method to substitute into the Attention-Unet uncertainty agent model for solving, and the results are shown in (a)-(d) of Figure 14 . From Figure 14It can be seen that, for the system energy transmission efficiency, the composite reactive resonance shielding coil capacitance is the most influential factor; and for the human body (including the human body, brain, and lungs), the position of the parked vehicle has the greatest impact, therefore, the driver or passenger should maintain a certain distance around the EV-WPT system. In addition, since the composite reactive resonance shielding coil can enhance the coupling between the coils of the EV-WPT system, it can effectively reduce the LMF, which also indirectly proves the effectiveness of the composite reactive resonance shielding coil proposed in the application.

[0148] The application proposes a composite reactive resonance shielding coil structure applied to an EV-WPT system, which is composed of a reactive resonance shielding coil, a nanocrystalline layer, and an aluminum plate, and can realize the reduction of magnetic flux density in the non-working area while ensuring high system energy transmission efficiency. To weaken the influence of uncertain factors on the electromagnetic exposure safety of the human body, the application proposes a fast multi-objective optimization method based on Attention-Unet deep learning for robust optimization design of the composite reactive resonance shielding coil structure. Through simulation test verification, after applying the composite reactive resonance shielding coil structure, the maximum induced electric field intensity of the human body and the lungs E max respectively decreased by 61.72% and 40.39%, and the system energy transmission efficiency increased by 11.79%. At the same time, the application establishes an uncertainty proxy model of the electromagnetic exposure dose of the human body and the system energy transmission efficiency, and integrates MOWOA to realize robust optimization considering uncertain factors, which significantly improves the optimization efficiency, and the calculation time is shortened from 1958.3h to 156.8h. To evaluate the influence of the LMF of the EV-WPT system on the electromagnetic exposure safety of the human body, the application constructs a refined human body simulation model, and quantifies the influence degree of the input parameters on the system energy transmission efficiency and the human body E max through the MOAT method. Simulation experiments show that, under the consideration of uncertain factors, the composite reactive resonance shielding coil structure can still reduce the mean value of the human body E max by 48.38%. In addition, without using the composite reactive resonance shielding coil structure, E max the probability of exceeding the ICNIRP2010 limit is 43.09%; while applying the structure, E max the exceeding probability is reduced to 0%, and the mean value of the system energy transmission efficiency is increased by 6.46%, which further proves the effectiveness and robustness of the composite reactive resonance shielding coil structure.

[0149] The above are only preferred embodiments of the present application, it should be pointed out that, for those skilled in the art, without departing from the concept of the present application, can also make several variations and improvements, these should also be considered as the protection scope of the present application, these will not affect the effect and the practicality of the patent of the present application.

Claims

1. A robust optimization method for a composite reactive resonant shield coil structure, characterized by, The application relates to a composite reactive resonance shielding coil structure applied to an electric vehicle wireless charging system. The Attention-Unet uncertainty agent model is constructed to take design variables and uncertainty factors as inputs and take system energy transmission efficiency, maximum induced electric field intensity of a human body trunk, maximum induced electric field intensity of a brain and maximum induced electric field intensity of a lung as outputs; the design variables include the distance between a transverse reactive resonance shielding coil and the edge of a receiving coil, the distance between a longitudinal reactive resonance shielding coil and the edge of the receiving coil and compensation capacitors of the transverse and longitudinal reactive resonance shielding coils; the uncertainty factors include human body transverse offset, human body longitudinal offset, vehicle transverse offset, vehicle longitudinal offset, the distance between a transmitting coil and the receiving coil and a human body orientation deflection angle; Based on the output of the Attention-Unet uncertainty agent model, a multi-objective whale optimization algorithm is used for multi-objective optimization to solve a Pareto front and obtain robust optimal parameters. The composite reactive resonance shielding coil structure applied to the electric vehicle wireless charging system comprises a reactive resonance shielding coil, a nanocrystalline layer and an aluminum plate, the nanocrystalline layer is pasted on the periphery of the reactive resonance shielding coil, and the aluminum plate is pasted on the nanocrystalline layer and located outside the reactive resonance shielding coil; the reactive resonance shielding coil generates a reverse offset magnetic field through a leakage magnetic field to weaken the electromagnetic field intensity of a non-working area of the electric vehicle wireless charging system; the nanocrystalline layer and the aluminum plate are used for attracting the leaked electromagnetic field to the periphery of the reactive resonance shielding coil through the high magnetic permeability characteristics to improve the reverse magnetic field intensity and reduce the external magnetic field intensity and reduce the human body electromagnetic exposure dose.

2. The robust optimization method of claim 1, wherein, The relative magnetic permeability of the nanocrystalline layer under room temperature and 85 kHz working conditions is 22000.

3. The robust optimization method of claim 1, wherein, The reactive resonance shielding coil is provided with four reactive resonance shielding coils, including two transverse reactive resonance shielding coils and two longitudinal reactive resonance shielding coils, which are symmetrically arranged on the transverse two sides and the longitudinal two sides of the receiving coil; the transverse reactive resonance shielding coil and the longitudinal reactive resonance shielding coil are both connected with compensation capacitors.

4. The robust optimization method of claim 3, wherein, The distance between the transverse reactive resonance shielding coil and the edge of the receiving coil is 0-0.05 m; the distance between the longitudinal reactive resonance shielding coil and the edge of the receiving coil is 0-0.05 m; The compensation capacitor of the transverse reactive resonance shielding coil is 500-1500 nF; the compensation capacitor of the longitudinal reactive resonance shielding coil is 500-1500 nF.

5. The robust optimization method of claim 1, wherein, The core framework of the Attention-Unet uncertainty agent model is an Attention-Unet network formed by introducing an Attention Gate mechanism on the basis of a U-Net network; the U-Net network is an encoder-decoder-based full convolution deep network model; After input data is normalized, the data enters the Attention-Unet network; First through the encoder, the encoder consists of convolutional layers and down-sampling layers; the convolutional operation automatically extracts the local spatial or sequential features in the input data, and a ReLU activation function is connected after the convolution; after passing through the convolutional layer, the down-sampling layer retains the maximum value in each window through MaxPool sliding, reduces the dimension step by step to compress the information, and retains the most key features; after processing through the down-sampling layer, it enters the Attention Gate to control the transfer of the encoder features to the decoder; the decoding path first passes through the up-sampling layer, then splices with the output of the Attention Gate, and finally passes through the convolution and ReLU activation, and then maps to the output dimension through a convolution to obtain The set of : ； wherein is a set of predicted outputs of the model; W final is the final convolution kernel parameter; is the feature input before entering the convolution mapping; and p is the bias.

6. The robust optimization method of claim 1, wherein, The multi-objective whale optimization algorithm updates the population position through three stages of searching for prey, surrounding prey and bubble net attack, and the position updating formula comprises: The searching for prey stage: ； ； where X(t+1) is the updated whale position; X r (t) is a randomly selected whale position from the population; A and C are coefficient vectors; D is the distance between the whale individual and the optimal solution; t is the current iteration number; X(t) is the current whale position; The surrounding prey stage: ； ； where z is a constant defining the helix shape; X z (t) is the best whale position obtained so far; The bubble net attack stage: ； ； In the formula, l is a random floating point number in the interval (-1, 1).

Images