Radiation emission field prediction method for electromagnetic emitter with obstacle in near area

Abstract

The invention provides a radiation emission field prediction method of an electromagnetic emitter with obstacles in a near area. The method comprises the following steps: obtaining an electromagnetic emitter free space equivalent source model based on free space electromagnetic emission characteristics; obtaining an optimized convolutional neural network and an optimized BP neural network based on the free space electromagnetic emission characteristics of the electromagnetic emitter; inputting a to-be-solved magnetic field position into the free space equivalent source model of the electromagnetic emitter, predicting a free space radiation emission magnetic field of the electromagnetic emitter by using the optimized convolutional neural network and the BP neural network, and subtracting the shielding effectiveness of a near-region obstacle from the free space electromagnetic emission magnetic field of the electromagnetic emitter, the radiation emission magnetic field of the electromagnetic emitter with the obstacle in the near area is obtained. According to the field source equivalent modeling method for the obstacles in the near regions based on the optimized CNN network and the optimized BP network cascade neural network, equivalent sources of electromagnetic disturbance sources with the obstacles in various near regions can be solved in the frequency domain, and magnetic field prediction is achieved.

Description

Technical Field

[0001] The present invention relates to the field of electromagnetic prediction technology, and in particular to a method for predicting the radiation emission field of an electromagnetic emitter with obstacles in the vicinity. Background Art

[0002] Various types of motor equipment (such as linear motors, rotary motors, etc.) are used in large quantities in daily life. The motors are excited by high currents and high voltages, so the electromagnetic emissions generated by the motors pose a great threat to the safety and reliability of other equipment and systems. When the motors are running in different states, the characteristics of the disturbance source will change. In addition, there are other equipment around the motors and other equipment, which will affect the electromagnetic emission characteristics of the motors. Therefore, the impact of the motor radiation magnetic field on the electromagnetic environment in the actual environment should be carefully considered to protect sensitive equipment and human health. Since the test method for evaluating the motor radiation characteristics is expensive and requires a lot of maintenance, it is impractical to use measurement methods to estimate motor radiation during the electromagnetic compatibility (EMC) design stage. Another common method for evaluating motor radiation is to use electromagnetic field numerical calculation methods. Numerical simulation can comprehensively consider the edge effects of synchronous motors, the properties of ferromagnetic materials, etc., and can calculate the radiation characteristics of motor equipment under any environment and any working conditions. However, when it comes to multi-source environments and large-scale computing areas, the complex structure of the motor leads to long calculation time and large memory usage in the numerical calculation process.

[0003] At present, the analysis and research of electromagnetic emission based on equivalent sources have attracted much attention. The equivalent source is determined by solving the field source integral equation, so that the radiation emission characteristics of the motor equipment can be reconstructed with the equivalent source with a simple structure, and the far-field radiation can be estimated. Due to the existence of equipment around the motor, the equivalent modeling process becomes more complicated and difficult.

[0004] At present, the methods for solving the equivalent source of motor equipment in the prior art include the least square method, neural network and convolutional neural network. The least square method is to minimize the sum of squares of the error between the solution data and the actual data, and is a linear estimate. The disadvantage of this method is that it will introduce a large nonlinear error into the solution of the equivalent source. The neural network and convolutional neural network methods use known field data to fit the equivalent source. The disadvantage of this method is that the error in the training data affects the accuracy of the solution of the equivalent source, thereby affecting the accuracy of the magnetic field prediction. Summary of the invention

[0005] The embodiment of the present invention provides a method for predicting the radiation emission field of an electromagnetic emitter with obstacles in the near field, so as to effectively perform equivalent modeling on the electromagnetic emission characteristics of various electromagnetic radiation sources with obstacles in the near field in the frequency domain.

[0006] In order to achieve the above object, the present invention adopts the following technical scheme.

[0007] A method for predicting the radiation emission field of an electromagnetic emitter with obstacles in the vicinity comprises:

[0008] The free-space electromagnetic emission characteristics of an electromagnetic emitter with obstacles near it are constructed by generalizing Huygens' principle, vector potential function and Maxwell's equations, and the free-space equivalent source model of the electromagnetic emitter is obtained.

[0009] Based on the free-space electromagnetic emission characteristics of the electromagnetic emitter, the optimized convolutional neural network and optimized BP neural network training are carried out for the purpose of equivalent source extraction and magnetic field intensity prediction, and the optimized convolutional neural network and BP neural network are obtained;

[0010] The position of the magnetic field to be determined is input into the free-space equivalent source model of the electromagnetic emitter, and the free-space radiation emission magnetic field of the electromagnetic emitter is predicted using the optimized convolutional neural network and the optimized BP neural network. The shielding effectiveness of the near-zone obstacles is subtracted from the free-space electromagnetic emission magnetic field of the electromagnetic emitter to obtain the radiation emission magnetic field of the electromagnetic emitter with obstacles in the near zone.

[0011] Preferably, the free-space electromagnetic emission characteristics of an electromagnetic emitter with obstacles in the near zone are constructed by using the generalized Huygens principle, vector potential function and Maxwell equations to obtain a free-space equivalent source model of the electromagnetic emitter, including:

[0012] According to Huygens' generalization principle, the equivalent source model is located on the surface of an air box with a size comparable to that of the original emission source. The surface of the air box is divided into N networks, and the center of the network is regarded as the equivalent source position. The current density in each grid is J xN , J yN , J zN , the magnetic flux density is M xN 、M yN 、M zN ;

[0013] The relationship between the equivalent current source, magnetic current source and magnetic field is obtained by using the vector potential function and Maxwell equations. The magnetic field generated by the current source in free space is described as three independent mathematical models:

[0014]

[0015] x, y, z are the field position coordinates, x′, y′, z′ are the source position coordinates, r is the field source distance, k is the wave number of electromagnetic wave propagation, H Js,x (r), H Js,y (r) and H Js,z (r) are the x, y, and z components of the magnetic field radiated by the equivalent current source, △Si is the area of ​​the network where the equivalent source is located;

[0016] The magnetic field generated by the magnetic current source in free space is described as:

[0017]

[0018] In the above formula:

[0019]

[0020] μ is the magnetic permeability of the medium, ω is the angular frequency, H M,x (r), H M,y (r) and H M,z (r) are the x, y, and z components of the magnetic field radiated by the equivalent magnetic current source;

[0021] The radiation emission characteristics of the radiation source at the free space position r are calculated by the superposition principle. The isotropic components of the total radiation magnetic field can be described by an equivalent source:

[0022]

[0023] Among them, H x , H y , H z They represent the magnetic field components in the x, y, and z directions respectively.

[0024] Preferably, the free-space electromagnetic emission characteristics of the electromagnetic emitter are used to perform optimized convolutional neural network and optimized BP neural network training for the purpose of equivalent source extraction and magnetic field intensity prediction to obtain optimized convolutional neural network and BP neural network, including:

[0025] Taking the prediction of the x component of the magnetic field as an example, the relationship between the x component of the magnetic field and the equivalent source is:

[0026]

[0027] in,

[0028]

[0029] According to equations (10)-(15), when the position coordinates x of the equivalent source are known i ,y i ,z i , the area of ​​each grid of the equivalent source model Δs i , propagation constant k and the distance r between the field sources, the weight coefficient C of the equivalent source is obtained gi (r)(g=1,2,3,4,5,i=1,2,...,m), using C gi(r)(g=1,2,3,4,5,i=1,2,...,m) is used as the input of the convolutional neural network, and the corresponding magnetic field data H is measured in the field x As the target output of the convolutional neural network.

[0030] Co-training is used to generate a semi-supervised algorithm to optimize the convolutional neural network. The training data is randomly divided into two groups, and two convolutional neural networks are trained separately. The predicted data of one convolutional neural network is used as the target output of the other network. The other convolutional neural network is trained continuously, and the one with better performance among the two convolutional neural networks is selected as the first level of the cascade neural network to obtain the optimized convolutional neural network.

[0031] Using C gi (r)(g=1,2,3,4,5,i=1,2,...,m) is used as the input of the BP neural network. The predicted value of the first-level convolutional neural network, the field measurement data and the numerical simulation data are used as the target output of the BP neural network. The BP neural network is trained. The input matrices are normalized. The matrix size is M×1 and the number of hidden layers is N. The BP neural network is optimized using the co-teaching algorithm. The training data are grouped and two BP neural networks are trained separately. The training data with smaller error is selected from one of the BP neural networks for the training of the other BP neural network. The better performance of the two BP neural networks is selected as the second level of the cascade neural network. The error threshold of the training data and the predicted data is set to 20% of the target output. The network credibility increases with the increase of the number of iterations. The error threshold is reduced by 0.1% of the target output after each iteration of training.

[0032] In the process of optimizing convolutional neural networks and back propagation neural networks, the training loss equation is defined to measure the gap between the model output and the actual result. The mean square error is used to quantify the training error. The mean square error E is expressed as:

[0033]

[0034] Among them, H i is the expected output magnetic field strength, a j (L) is the output of the jth neuron in the Lth layer;

[0035] According to the training effect, the parameters of the number of convolution kernels in each layer of the convolutional neural network, the number of convolutional layers, the number of fully connected layers, and the number of grids in the equivalent source model are adjusted until an acceptable mean square error E is obtained through training; according to the training effect, the parameters of the number of neurons in each layer of the back propagation neural network, the number of hidden layers, and the number of grids in the equivalent source model are adjusted until an acceptable mean square error E is obtained through training, thereby obtaining the optimized BP neural network.

[0036] Preferably, the method comprises inputting the position of the magnetic field to be determined into the free space equivalent source model of the electromagnetic emitter, predicting the free space radiation emission magnetic field of the electromagnetic emitter by using the optimized convolutional neural network and the optimized BP neural network, and subtracting the shielding effectiveness of the near-zone obstacle from the free space electromagnetic emission magnetic field of the electromagnetic emitter to obtain the radiation emission magnetic field of the electromagnetic emitter with obstacles in the near zone, including:

[0037] The magnetic field position to be determined is input into the free space equivalent source model of the electromagnetic emitter, and the free space equivalent source weight coefficient C of the electromagnetic emitter is calculated by using the optimized convolutional neural network and the optimized BP neural network. gi (r) Extract and predict the free space radiation emission magnetic field H of the electromagnetic emitter x , H y , H z , further using formula (17) to obtain the free space radiation emission magnetic field H of the electromagnetic emitter J,M ;

[0038]

[0039] H x , H y , H z Respectively represent the magnetic field components in the x, y, and z directions;

[0040] Obtain the radiation emission field H of the electromagnetic emitter with obstacles in the near area, and use formula (18) to obtain the radiation emission magnetic field H (dB) of the electromagnetic emitter with obstacles in the near area

[0041] H(dB)=H J,M (dB)-SE H (dB) (18)

[0042] Among them, H J,M is the free-space radiation emission magnetic field of the electromagnetic emitter, SE H is the shielding effectiveness of the magnetic field generated by the obstacle. The shielding effectiveness can be obtained by numerical simulation or measurement.

[0043] It can be seen from the technical solutions provided by the above-mentioned embodiments of the present invention that the field source equivalent modeling method for near-zone obstacles based on the optimized CNN network and the optimized BP network cascade neural network proposed in the present invention can obtain the equivalent sources of electromagnetic disturbance sources of various near-zone obstacles in the frequency domain and realize magnetic field prediction.

[0044] Additional aspects and advantages of the present invention will be given in part in the following description, which will become obvious from the following description, or may be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0046] Figure 1 A schematic diagram of the implementation principle of a method for predicting the radiation emission field of an electromagnetic emitter with obstacles in the vicinity based on an optimized convolutional network and an optimized BP network cascade neural network provided by an embodiment of the present invention;

[0047] Figure 2 A processing flow chart of a method for predicting the radiation emission field of an electromagnetic emitter with obstacles in the vicinity based on an optimized convolutional network and an optimized BP network cascade neural network provided by an embodiment of the present invention;

[0048] Figure 3 A structural diagram of a BP (Backpropagation) neural network provided in an embodiment of the present invention;

[0049] Figure 4 A structural diagram of a CNN neural network provided by an embodiment of the present invention;

[0050] Figure 5 A processing flow chart of collaborative teaching provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0051] The embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be interpreted as limiting the present invention.

[0052] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. In addition, the "connection" or "coupling" used herein may include wireless connection or coupling. The term "and / or" used herein includes any unit and all combinations of one or more associated listed items.

[0053] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless defined as herein.

[0054] To facilitate understanding of the embodiments of the present invention, several specific embodiments will be further explained below with reference to the accompanying drawings, and each embodiment does not constitute a limitation on the embodiments of the present invention.

[0055] The embodiment of the present invention uses the radiation emission characteristics of the electromagnetic emitter with obstacles in the near field to improve the efficiency of electromagnetic environment analysis. At the same time, the existence of obstacles in the near field will reduce the solution accuracy of the equivalent source model. Therefore, the present invention proposes that the equivalent modeling of obstacles in the near field is divided into two steps, first reconstructing the radiation emission characteristics of the emission source in free space, and then considering the influence of obstacles. The prediction method proposed in the present invention can perform equivalent modeling of the electromagnetic emission characteristics of various electromagnetic radiation sources with obstacles in the near field in the frequency domain.

[0056] The embodiment of the present invention is based on the free space electromagnetic emission characteristics of the electromagnetic emitter, takes the matrix C calculated by the field source position as the input of the optimized cascade neural network, and respectively uses the optimized convolutional neural network and the optimized BP neural network to extract the equivalent source weight coefficient, thereby reducing the influence of the known magnetic field data error, realizing the prediction of the free space magnetic field intensity, and providing a reliable tool for the study of the electromagnetic emission characteristics of the free space emission source.

[0057] The embodiment of the present invention is based on the shielding effectiveness of obstacles and the impact of equivalent obstacles on the radiation emission characteristics of the emission source. It uses the equivalent source model of the free space emission source and the shielding effectiveness of obstacles to predict the magnetic field strength in the near zone where obstacles exist, and provides a reliable tool for studying the electromagnetic environment characteristics of various near-field areas with obstacles in the frequency domain.

[0058] The schematic diagram of the implementation principle of a method for predicting electromagnetic emission of a transmitter with obstacles in the vicinity based on an optimized convolutional network and an optimized BP network cascade neural network proposed in an embodiment of the present invention is shown in FIG. Figure 1 As shown in FIG. 1 , the method ensures accurate analysis and classification of various electromagnetic disturbances and time-frequency characteristics of communication signals. The method of the present invention can be applied to the electromagnetic environment of high-speed maglev trains. The processing flow of the method is as follows: Figure 2 As shown, the processing steps include the following:

[0059] Step S10: construct the free-space electromagnetic emission characteristics of the electromagnetic emitter with obstacles in the near area through the generalized Huygens principle, vector potential function and Maxwell equations, and obtain the free-space equivalent source model of the electromagnetic emitter. The electromagnetic emission characteristics include the relationship expression between the equivalent source and the electromagnetic emission magnetic field.

[0060] In the modeling process, the position of the equivalent source determined according to Huygens' generalization principle is located on the surface of an air box with a size equivalent to that of the emitter. To reduce the computational complexity, the surface of the air box is divided into N networks, and the center of the network is regarded as the equivalent source position. The current density in each grid is J xN , J yN , J zN , the magnetic flux density is M xN 、M yN 、M zN .

[0061] The relationship between the equivalent current source, magnetic current source and magnetic field is obtained by using the vector potential function and Maxwell equations. Among them, the magnetic field generated by the current source in free space can be described as three independent mathematical models:

[0062]

[0063] x, y, z are the field position coordinates, x′, y′, z′ are the source position coordinates, r is the field source distance, k is the wave number of electromagnetic wave propagation, H Js,x (r), H Js,y (r) and H Js,z (r) are the x, y, and z components of the magnetic field radiated by the equivalent current source, △S i is the area of ​​the network where the equivalent source is located.

[0064] The magnetic field generated by a magnetic current source in free space can be described as:

[0065]

[0066] In the above formula:

[0067]

[0068] μ is the magnetic permeability of the medium, ω is the angular frequency, H M,z (r), H M,y (r) and H M,z (r) are the x, y, and z components of the magnetic field radiated by the equivalent magnetic current source.

[0069] The radiation emission characteristics of the radiation source at the free space position r are calculated by the superposition principle. The isotropic components of the total radiation magnetic field can be described by an equivalent source:

[0070]

[0071] Among them, H x , H y , H z They represent the magnetic field components in the x, y, and z directions respectively.

[0072] Step S20: Based on the free-space electromagnetic emission characteristics of the electromagnetic emitter, an optimized convolutional neural network and an optimized BP neural network are trained for the purpose of equivalent source amplitude extraction and magnetic field intensity prediction to obtain optimized convolutional neural networks and BP neural networks.

[0073] An optimized convolutional neural network structure provided by an embodiment of the present invention is as follows: Figure 3 As shown, the structure of a BP neural network is as follows Figure 4 shown.

[0074] Taking the prediction of the x component of the magnetic field as an example, the relationship between the x component of the magnetic field and the equivalent source is:

[0075]

[0076] in,

[0077]

[0078] According to equations (10)-(15), when the position coordinates x of the equivalent source are known i ,y i ,z i , the area of ​​each grid of the equivalent source model Δs i , propagation constant k and the distance r between the field sources, the weight coefficient C of the equivalent source is obtained gi (r)(g=1,2,3,4,5,i=1,2,...,m), using C gi(r)(g=1,2,3,4,5,i=1,2,...,m) is used as the input of the convolutional neural network, and the corresponding magnetic field data H is measured in the field x As the target output of the convolutional neural network.

[0079] The co-training semi-supervised algorithm is used to optimize the convolutional neural network. The co-training semi-supervised optimization algorithm randomly divides the training data into two groups and trains two convolutional neural networks separately. In addition, the prediction data of one network is used as the target output of the other network to continue training the other network. Finally, the better performing of the two convolutional neural networks is selected as the first level of the cascade neural network.

[0080] The input matrices are all normalized, the matrix size is M×1, and the size of each convolution kernel in the convolution layer is 3×1. After being processed by the k1 convolution kernels of the first layer, the processed information is transformed using the Sigmoid activation function, and after being processed by the second layer of convolution kernels with a size of 3×1, the processed information is transformed using the ReLU activation function.

[0081] The Sigmoid and ReLU activation functions are used to introduce nonlinear factors in the training process to improve the expressiveness of the model. After the Sigmoid and ReLU function excitation layers, the information is averaged and pooled to reduce the model size, increase the calculation speed, and improve the robustness of the extracted features. Repeat the convolution layer → nonlinear activation function excitation layer → pooling layer arrangement to achieve feature extraction of equivalent source amplitude. A fully connected layer is added after the last pooling layer to map the information extracted by the equivalent source amplitude feature extraction layer to the magnetic field intensity space as the output layer to achieve magnetic field intensity prediction.

[0082] Using C gi (r)(g=1,2,3,4,5,i=1,2,...,m) is used as the input of the optimized BP neural network, and the predicted value of the first-level convolutional neural network, the field measurement data and the numerical simulation data are used as the target output of the BP network to train the network. The input matrices are all normalized, the matrix size is M×1, and the number of hidden layers is N. The Sigmoid activation function is used between the hidden layer and the input layer to transform the processed information, and the Tanh activation function is used between the hidden layer and the output layer to transform the processed information. The Sigmoid and Tanh activation functions are used to introduce nonlinear factors in the training process to improve the expression ability of the model. The hidden layer realizes the feature extraction of the equivalent source amplitude, and the output layer realizes the prediction of the magnetic field intensity.

[0083] The back propagation neural network is optimized using a co-teaching algorithm. The processing flow of a co-teaching algorithm provided by an embodiment of the present invention is as follows: Figure 5 As shown. The collaborative teaching algorithm is to group the training data and train two BP neural networks separately. One of the BP neural networks is used to select the training data with smaller error for the training of the other network to reduce the impact of the training set error. Finally, the two back-propagation neural networks with better performance are selected as the second level of the cascade neural network. The error threshold of the training data and the predicted data is set to 20% of the target output. As the number of iterations increases, the network credibility increases, so the error threshold is reduced by 0.1% of the target output after each iteration of training.

[0084] In the process of optimizing convolutional neural networks and back propagation neural networks, the training loss equation is defined to measure the gap between the model output and the actual result. The mean square error is used to quantify the training error. The mean square error E is expressed as:

[0085]

[0086] Among them, H i is the expected output magnetic field strength, a j (L) is the output of the jth neuron in the Lth layer.

[0087] According to the training effect, the parameters of the number of convolution kernels in each layer of the convolutional neural network, the number of convolutional layers, the number of fully connected layers, and the number of grids in the equivalent source model are adjusted until an acceptable mean square error E is obtained through training; according to the training effect, the parameters of the number of neurons in each layer of the back propagation neural network, the number of hidden layers, and the number of grids in the equivalent source model are adjusted until an acceptable mean square error E is obtained through training, thereby obtaining the optimized BP neural network.

[0088] After the optimization cascade neural network training is completed, the normalized weight coefficient matrix C is input gi The matrix can be used to output the x, y, and z components of the magnetic field intensity at the corresponding position through the neural network, and finally the components of the magnetic field intensity are integrated into the total magnetic field intensity at the desired position:

[0089]

[0090] H x , H y , H z They represent the magnetic field components in the x, y, and z directions respectively.

[0091] Step S30, input the position of the magnetic field to be determined in the free space equivalent source model of the electromagnetic emitter, subtract the shielding effectiveness of the near-field obstacles from the free space electromagnetic emission magnetic field of the electromagnetic emitter, and obtain the radiation emission magnetic field of the electromagnetic emitter with obstacles in the near field.

[0092] When there are obstacles near the electromagnetic emitter, the equivalent source model of the free space of the electromagnetic emitter is used to consider the influence of the obstacles on the radiation emission characteristics of the electromagnetic emitter. For obstacles with extremely small conductivity and negligible eddy current effect, the magnetic field shielding effectiveness is used to equal the influence of the obstacles.

[0093] The embodiment of the present invention is based on the free space electromagnetic emission characteristics of the electromagnetic emitter, and uses an optimized convolutional neural network and an optimized BP neural network to calculate the free space equivalent source weight coefficient C of the electromagnetic emitter. gi (r) Extract and predict the free space radiation emission magnetic field H of the electromagnetic emitter x , H y , H z , and further use formula (18) to get H J,M .

[0094] The radiation emission field H of the electromagnetic emitter with obstacles in the near area is obtained:

[0095] H(dB)=H J,M (dB)-SE H (dB) (18)

[0096] Among them, H J,M is the free-space radiation emission magnetic field of the electromagnetic emitter, SE H is the shielding effectiveness of the magnetic field generated by the obstacle. The shielding effectiveness can be obtained by numerical simulation or measurement.

[0097] In summary, the embodiment of the present invention proposes that the equivalent modeling of obstacles in the near zone is divided into three steps. First, the equivalent source model of the radiation emission of the emitter in the presence of obstacles is reconstructed. Secondly, the equivalent source weight coefficient is extracted using the optimized convolutional neural network and the optimized BP neural network, and the radiation emission characteristics of the emitter in the presence of obstacles are calculated. Finally, the influence of obstacles is considered. The shielding effectiveness of the obstacle is obtained by numerical simulation or measurement. Based on the radiation emission characteristics of the emitter in the presence of obstacles, the shielding effectiveness of the obstacle is processed to obtain the radiation emission characteristics of the emitter in the absence of obstacles. The prediction method proposed in the present invention can perform equivalent modeling of the electromagnetic emission characteristics of electromagnetic radiation sources in various near-field areas with obstacles in the frequency domain.

[0098] Those skilled in the art can understand that the accompanying drawings are only schematic diagrams of an embodiment, and the modules or processes in the accompanying drawings are not necessarily required to implement the present invention.

[0099] It can be known from the description of the above implementation methods that those skilled in the art can clearly understand that the present invention can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present invention or certain parts of the embodiments.

[0100] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The device and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0101] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for predicting the radiation emission field of an electromagnetic emitter with obstacles in the vicinity, characterized in that: include: The free-space electromagnetic emission characteristics of an electromagnetic emitter with obstacles near it are constructed by generalizing Huygens' principle, vector potential function and Maxwell's equations, and the free-space equivalent source model of the electromagnetic emitter is obtained. Based on the free-space electromagnetic emission characteristics of the electromagnetic emitter, the optimized convolutional neural network and optimized BP neural network training are carried out for the purpose of equivalent source extraction and magnetic field intensity prediction, and the optimized convolutional neural network and BP neural network are obtained; The position of the magnetic field to be determined is input into the free-space equivalent source model of the electromagnetic emitter, and the free-space radiation emission magnetic field of the electromagnetic emitter is predicted using the optimized convolutional neural network and the optimized BP neural network. The shielding effectiveness of the near-zone obstacles is subtracted from the free-space electromagnetic emission magnetic field of the electromagnetic emitter to obtain the radiation emission magnetic field of the electromagnetic emitter with obstacles in the near zone.

2. The method according to claim 1, characterized in that The free-space electromagnetic emission characteristics of an electromagnetic emitter with obstacles in the near zone are constructed by using the generalized Huygens principle, vector potential function and Maxwell equations to obtain a free-space equivalent source model of the electromagnetic emitter, including: According to Huygens' generalization principle, the equivalent source model is located on the surface of an air box with a size comparable to that of the original emission source. The surface of the air box is divided into N networks, and the center of the network is regarded as the equivalent source position. The current density in each grid is J xN , J yN , J zN , the magnetic flux density is M xN 、M yN 、M zN ; The relationship between the equivalent current source, magnetic current source and magnetic field is obtained by using the vector potential function and Maxwell equations. The magnetic field generated by the current source in free space is described as three independent mathematical models: x, y, z are the field position coordinates, x′, y′, z′ are the source position coordinates, r is the field source distance, k is the wave number of electromagnetic wave propagation, H Js,x (r), H Js,y (r) and H Js,z (r) are the x, y, and z components of the magnetic field radiated by the equivalent current source, △S i is the area of ​​the network where the equivalent source is located; The magnetic field generated by the magnetic current source in free space is described as: In the above formula: μ is the magnetic permeability of the medium, ω is the angular frequency, H M,x (r), H M,y (r) and H M,z (r) are the x, y, and z components of the magnetic field radiated by the equivalent magnetic current source; The radiation emission characteristics of the radiation source at the free space position r are calculated by the superposition principle. The isotropic components of the total radiation magnetic field can be described by an equivalent source: Among them, H x , H y , H z They represent the magnetic field components in the x, y, and z directions respectively.

3. The method according to claim 2, characterized in that The free space electromagnetic emission characteristics of the electromagnetic emitter are used to perform the optimized convolutional neural network and optimized BP neural network training for the purpose of equivalent source extraction and magnetic field intensity prediction, and the optimized convolutional neural network and BP neural network are obtained, including: Taking the prediction of the x component of the magnetic field as an example, the relationship between the x component of the magnetic field and the equivalent source is: in, According to equations (10)-(15), when the position coordinates x of the equivalent source are known i ,y i ,z i , the area of ​​each grid of the equivalent source model Δs i , propagation constant k and the distance r between the field sources, the weight coefficient C of the equivalent source is obtained gi (r)(g=1,2,3,4,5,i=1,2,...,m), using C gi (r)(g=1,2,3,4,5,i=1,2,...,m) is used as the input of the convolutional neural network, and the corresponding magnetic field data H is measured in the field x As the target output of the convolutional neural network; The convolutional neural network is optimized by using a semi-supervised algorithm generated by co-training. The training data is randomly divided into two groups, and two convolutional neural networks are trained separately. The predicted data of one convolutional neural network is used as the target output of the other network. The other convolutional neural network is trained continuously, and the one with better performance is selected as the first level of the cascade neural network to obtain the optimized convolutional neural network. Using C gi (r)(g=1,2,3,4,5,i=1,2,...,m) is used as the input of the BP neural network. The predicted value of the first-level convolutional neural network, the field measurement data and the numerical simulation data are used as the target output of the BP neural network. The BP neural network is trained. The input matrices are normalized. The matrix size is M×1 and the number of hidden layers is N. The BP neural network is optimized using the co-teaching algorithm. The training data are grouped and two BP neural networks are trained separately. The training data with smaller error is selected from one of the BP neural networks for the training of the other BP neural network. The better performance of the two BP neural networks is selected as the second level of the cascade neural network. The error threshold of the training data and the predicted data is set to 20% of the target output. The network credibility increases with the increase of the number of iterations. The error threshold is reduced by 0.1% of the target output after each iteration of training. In the process of optimizing convolutional neural networks and back propagation neural networks, the training loss equation is defined to measure the gap between the model output and the actual result. The mean square error is used to quantify the training error. The mean square error E is expressed as: Among them, H i is the expected output magnetic field strength, a j (L) is the output of the jth neuron in the Lth layer; According to the training effect, the parameters of the number of convolution kernels in each layer of the convolutional neural network, the number of convolutional layers, the number of fully connected layers, and the number of grids in the equivalent source model are adjusted until an acceptable mean square error E is obtained through training; according to the training effect, the parameters of the number of neurons in each layer of the back propagation neural network, the number of hidden layers, and the number of grids in the equivalent source model are adjusted until an acceptable mean square error E is obtained through training, thereby obtaining the optimized BP neural network.

4. The method according to claim 3, characterized in that The method comprises: inputting the position of the magnetic field to be determined into the free space equivalent source model of the electromagnetic emitter, predicting the free space radiation emission magnetic field of the electromagnetic emitter by using the optimized convolutional neural network and the optimized BP neural network, and subtracting the shielding effectiveness of the near-zone obstacle from the free space electromagnetic emission magnetic field of the electromagnetic emitter to obtain the radiation emission magnetic field of the electromagnetic emitter with obstacles in the near zone, including: The magnetic field position to be determined is input into the free space equivalent source model of the electromagnetic emitter, and the free space equivalent source weight coefficient C of the electromagnetic emitter is calculated by using the optimized convolutional neural network and the optimized BP neural network. gi (r) Extract and predict the free space radiation emission magnetic field H of the electromagnetic emitter x , H y , H z , further using formula (17) to obtain the free space radiation emission magnetic field H of the electromagnetic emitter J,M ; H x , H y , H z Respectively represent the magnetic field components in the x, y, and z directions; Obtain the radiation emission field H of the electromagnetic emitter with obstacles in the near area, and use formula (18) to obtain the radiation emission magnetic field H (dB) of the electromagnetic emitter with obstacles in the near area H(dB)=H J,M (dB)-SE H (dB) (18) Among them, H J,M is the free-space radiation emission magnetic field of the electromagnetic emitter, SE H It is the shielding effectiveness of the magnetic field generated by the obstacle. The shielding effectiveness can be obtained through numerical simulation or measurement.

Images