Cable electromagnetic radiation response prediction method based on one-dimensional residual attention CNN

Abstract

The present invention discloses a cable electromagnetic radiation response prediction method based on one-dimensional residual attention CNN, which belongs to the field of electromagnetic radiation response waveform prediction. The method determines the physical factors that affect the cable electromagnetic pulse radiation response waveform, samples the physical factors in large quantities, obtains the output waveform of the sampling point through full-wave simulation, divides the simulation data into a training set, a verification set and a test set, constructs a network model based on one-dimensional residual attention CNN, trains, verifies and tests it with simulation data, compares the waveform predicted by the network with the waveform obtained by full-wave simulation, and obtains the cable electromagnetic pulse radiation response prediction model; inputs the sample points of the test set into the trained cable electromagnetic pulse radiation response prediction model for prediction, and performs denormalization on the prediction results to obtain the final prediction results. The method realizes the rapid prediction of the response waveform of the wire harness under high-power microwave irradiation that can better reflect the actual situation.

Description

Technical Field

[0001] The present invention belongs to the field of electromagnetic radiation response waveform prediction, and specifically relates to a cable electromagnetic radiation response prediction method based on one-dimensional residual attention CNN. Background Art

[0002] With the continuous development of electromagnetic pulse weapons and the widespread application of high-power radio communication equipment, both military and civilian electronic equipment face severe electromagnetic interference threats. Among them, high-power microwaves, as a new type of directed energy weapon, damage or destroy the electronic system of the target object by destroying or destroying the electronic system of the target object, combining both "soft" and "hard" killing capabilities. It can be used in areas such as forced parking of vehicles and responding to drone swarm attacks. Its peak power is above 100MW, the radiation wavelength is in the centimeter to millimeter range, and the frequency is between 0.3GHz and 300GHz. It has potential speed, range, flexibility and reusability.

[0003] Studies have shown that the damage state of electronic systems caused by electromagnetic interference is related to the coupling path of electromagnetic interference, the sensitivity of electronic systems and the source of electromagnetic interference. High-power microwaves generate coupling voltage and current on the wiring harness inside the electronic system through field line coupling, causing damage to the terminal equipment of the wire. It can be said that under the action of high-power microwaves, obtaining the coupling voltage and coupling current values ​​on the wiring harness of the electronic system has become a prerequisite for studying the electromagnetic pulse effect of the electronic system and performing electromagnetic protection. At present, the commonly used methods are irradiation test method and electromagnetic simulation method based on mechanism analysis. For the irradiation test method, due to the limitations of test conditions, a large number of repeated tests are often required, the test cost is expensive and the test cycle is long. In addition, the pulse source used in the test cannot provide a sufficient test field strength environment, and it is necessary to predict and extrapolate based on existing data. The electromagnetic simulation method based on mechanism analysis has a long simulation time and complex calculations, and cannot meet the rapid prediction requirements of engineering applications. In addition, there are many factors that affect the waveform of the wiring harness coupling voltage, and the current methods mostly use the high-power microwave modulation signal given in the standard as the excitation source for simulation, which can only be refined to a point and cannot reflect the situation under actual microwave irradiation. Summary of the invention

[0004] In view of the above problems, the purpose of the present invention is to provide a cable electromagnetic radiation response prediction method based on one-dimensional residual attention CNN, which determines the physical factors that affect the electromagnetic pulse response waveform of the wire harness, samples the physical factors in large quantities, obtains the output waveform of the sampling point through full-wave numerical simulation, divides the simulation data into training set, verification set and test set, constructs a network model based on one-dimensional residual attention CNN, and trains, tests and verifies it with simulation data. It can achieve rapid prediction of the response waveform of the wire harness under high-power microwave irradiation that can better reflect the actual situation, saves calculation time, and facilitates engineering application.

[0005] To achieve the above object, the present invention adopts the following technical solution: a cable electromagnetic radiation response prediction method based on one-dimensional residual attention CNN, characterized in that it includes the following steps:

[0006] Step 1: Construct a cable electromagnetic radiation response prediction dataset

[0007] ① Select the physical factors that affect the electromagnetic pulse response waveform of the cable as the input variable, and the electromagnetic pulse response waveform of the cable as the output variable;

[0008] ② Select sample points from the input variable set;

[0009] ③ Use the full-wave simulation algorithm to calculate the coupled voltage and current waveforms of the cable at each sample point;

[0010] ④ Combining each sample point with the coupling voltage waveform corresponding to the sample point to construct a data set; and dividing the data set into a training set, a validation set, and a test set;

[0011] Step 2: Construct a one-dimensional residual attention CNN network model, which includes a convolutional layer, a batch normalization layer, a ReLU activation function layer, five residual attention modules, and a maximum pooling layer;

[0012] Step 3: Use the training set to train the one-dimensional residual attention CNN network model, and use the validation set to validate the network model after each round of training; compare the waveform predicted by the network with the waveform obtained by full-wave simulation to obtain the cable electromagnetic radiation response prediction model;

[0013] Step 4: Input the sample points of the test set into the trained cable electromagnetic radiation response prediction model for prediction, and perform denormalization on the prediction results to obtain the final prediction results.

[0014] Furthermore, the physical factors affecting the electromagnetic response waveform of the cable in step ① include transmission line parameters, excitation source parameters and excitation source waveform, wherein the transmission line parameters include cable length L and cable height H from the ground plane, and the excitation source parameters include transient electric field peak E0, center frequency f0, incident angle ψ, polarization angle α and azimuth angle φ.

[0015] Furthermore, the process of selecting sample points in the input variable set in step ② is: uniformly sampling within the variation range of polarization angle α, incident angle ψ, azimuth angle φ, transient electric field peak value E0 of the excitation source and center frequency f0 of the excitation source, wherein the variation range of polarization angle α is 0 to π, the variation range of incident angle ψ is 0 to π / 2, the variation range of azimuth angle φ is 0 to π, the variation range of transient electric field peak value E0 of the excitation source is 20k / V to 10k / V, and the variation range of center frequency f0 of the excitation source is 0.5GHz to 2.5GHz.

[0016] Furthermore, the process of selecting sample points in the input variable set in step ② is: uniformly sampling within the variation range of the cable length L and the cable height H from the ground plane, wherein the variation range of the cable length L is 50 cm to 100 cm, and the variation range of the cable height H from the ground plane is 2 cm to 10 cm.

[0017] Furthermore, the full-wave simulation algorithm used in step ③ includes a transmission line matrix method, a finite integral method and a finite difference time domain method, and the response voltage and current waveforms of the cable are obtained through three-dimensional full-wave electromagnetic simulation.

[0018] Through the above-mentioned design scheme, the present invention can bring the following beneficial effects: the cable electromagnetic radiation response prediction method based on one-dimensional residual attention CNN provided by the present invention improves the calculation efficiency, and breaks through the previous cable electromagnetic pulse response calculation under the fixed electromagnetic pulse threat that can only be refined to a point, so that the prediction model can better reflect the electromagnetic pulse threat suffered by the cable under actual circumstances, realize the rapid prediction of the cable electromagnetic pulse response, save calculation time, facilitate engineering application, and provide a new solution for electromagnetic pulse response prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a schematic diagram of the cable model under electromagnetic pulse irradiation from the first perspective;

[0020] Figure 2 This is a schematic diagram of the cable model under electromagnetic pulse irradiation from the second viewing angle;

[0021] Figure 3 It is the algorithm flow chart of the cable electromagnetic radiation response prediction method based on one-dimensional residual attention CNN;

[0022] Figure 4 Schematic diagram of the attention module;

[0023] Figure 5 Schematic diagram of the residual attention module;

[0024] Figure 6 For the overall network structure;

[0025] Figure 7This is the prediction result diagram of the validation sample. DETAILED DESCRIPTION

[0026] The purpose of the present invention is to provide a cable electromagnetic radiation response prediction method based on one-dimensional residual attention CNN to improve the computational efficiency and reflect the electromagnetic pulse threat to the cable in actual situations.

[0027] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0028] like Figure 3 As shown, the method comprises the following steps:

[0029] Step 1: Construct a cable electromagnetic pulse radiation response prediction dataset

[0030] ① Select the physical factors that affect the electromagnetic pulse response waveform of the cable as the input variable, and the electromagnetic pulse response waveform of the cable as the output variable;

[0031] The method of selecting the physical factors that affect the electromagnetic pulse response waveform of the cable described in step ① is:

[0032] The factors that affect the electromagnetic pulse response waveform of the cable mainly include transmission line parameters, excitation source parameters and excitation source waveform. The transmission line parameters mainly include cable length L and cable height H from the ground plane. The excitation source parameters mainly include transient electric field peak E0, center frequency f0, incident angle ψ, polarization angle α and azimuth angle φ. The excitation source waveform E(t) is determined by the excitation source parameters. The time series of the excitation source waveform is obtained by calculating the excitation source expression. Different excitation sources correspond to different excitation source expressions. Figure 1 and Figure 2 The schematic diagrams of the cable model under electromagnetic pulse irradiation at two viewing angles are shown respectively. In this embodiment, the cable line type is selected as a single core line, and the electromagnetic pulse excitation source is a high-power microwave specified in the IEC standard. The excitation source waveform expression is E(t)=E o e -βt sin(ω o t), where E0 is the peak value of the transient electric field, ω o =2πf o , f0 is the center frequency, β=10 8 rad / s is the damping constant, and t is the time. The cable terminal impedance, cable radius, conductivity of the cable material, relative dielectric constant, etc. are all non-random parameters, and their influence on the electromagnetic pulse response waveform of the cable can be determined. Therefore, they can be determined as fixed parameters and are not used as inputs to the prediction model.

[0033] like Figure 1 and Figure 2As shown, a rectangular coordinate system xyz is established, and the xoy plane is the ground plane, where is the direction of the electric field, is the magnetic field direction, is the wave vector direction, L is the cable length, and Z1 and Z2 are the cable terminal impedances.

[0034] ②Select sample points from the input variable set

[0035] For excitation source parameters, such as Figure 1 and Figure 2 As shown, the polarization angle α represents the angle between the electric field and the incident plane, and its range is 0~π; the incident angle ψ represents the direction of the wave vector The angle between the wave vector and the xoy plane varies from 0 to π / 2. The azimuth angle φ represents the angle between the projection of the wave vector on the xoy plane and the x-axis, which varies from 0 to π. It is the propagation direction of the incident plane electromagnetic wave, that is, the normal direction of the incident plane. The transient electric field peak E0 of the excitation source varies from 20k / V to 100k / V, and the center frequency f of the excitation source is o The range of variation is 0.5GHz to 2.5GHz. The incident angle ψ, azimuth angle φ and polarization angle α of the electromagnetic pulse are adjusted by changing the incident plane wave propagation direction vector V and the electric field vector E, that is: V = (-sin(ψ)×cos(φ),-sin(ψ)×sin(φ),-cos(ψ))

[0036] E=(cos(φ)×cos(ψ)×cos(α)-sin(φ)×sin(α), sin(φ)×cos(ψ)×cos(α)+cos(φ)×sin(α), -sin(ψ)×cos(α))

[0037] The cable length L ranges from 50 cm to 100 cm, and the cable height H ranges from 2 cm to 10 cm. Both are sampled evenly within the range.

[0038] In this embodiment, the transient electric field peak value E0, the center frequency f o , polarization angle α, incident angle ψ and azimuth angle φ, as well as the cable length L and the cable height H from the ground plane training sample sampling points are:

[0039] E0=20h,h=1,2,...,5

[0040] f0=0.5i,i=1,2,...,5

[0041] α=18°×j, j=-5,-4,...,0,...,4,5

[0042] ψ=18°×k, k=0,1,...,5

[0043] φ=18°×l,l=-5,-4,...,0,...,4,5

[0044] L=50+10u,u=0,1,...,5

[0045] H=2v,v=1,...,5

[0046] Among them, h, i, j, k, l, u and v are serial numbers.

[0047] It should be noted that after the excitation source parameters are determined, the electromagnetic pulse waveform is obtained through the excitation source expression, and it is used as one of the physical factors affecting the electromagnetic pulse response waveform of the cable and is used as an input variable.

[0048] A total of 544,500 training sample data were collected.

[0049] E0=20h v ,h v =1,3,5

[0050] f0=0.5i v ,i v =1,2,...,5

[0051] α=2°+36°×j v ,j v =-2,-1,...,2

[0052] ψ=2°+36°×k v ,k v =0,1,2

[0053] φ=2°+36°×l v ,l v =-2,-1,...,2

[0054] L=52+20u v ,u v =0,1,2

[0055] H=0.2+2v v ,v v =1,3,5

[0056] where h v 、i v 、j v , k v , l v 、u v and v v It is a serial number.

[0057] A total of 6075 validation sample data were collected.

[0058] E0=20h t ,h t =1,3,5

[0059] f0=0.5i t ,i t =1,2,...,5

[0060] α=4°+36°×j t ,j t =-2,-1,...,2

[0061] ψ=4°+36°×k t ,k t =0,1,2

[0062] φ=4°+36°×l t ,l t =-2,-1,...,2

[0063] L=54+20u t ,u t =0,1,2

[0064] H=0.4+2v t ,v t =1,3,5

[0065] where h t 、i t 、j t , k t , l t 、u t and v t It is a serial number.

[0066] A total of 6075 test sample data were collected.

[0067] The impedance of both ends of the cable is 50Ω, the cable radius is r=1mm, and the conductivity is σ g =100S / m, dielectric constant is ε rg =2.3.

[0068] ③ Use full-wave simulation software to calculate the response voltage and current waveforms on the cable at each sample point. In this embodiment, the transmission line matrix method (Transmission-Line Modeling, TLM) is used in the full-wave simulation software to calculate the output waveform at each sample point.

[0069] ④ Combine each sample point with the coupling voltage waveform corresponding to the sample point to construct a data set; and divide the data set into a training set, a validation set, and a test set.

[0070] Step 2: Construct a one-dimensional residual attention CNN network model, which includes a convolutional layer, a batch normalization layer, a ReLU activation function layer, five residual attention modules, and a maximum pooling layer.

[0071] The first layer is a one-dimensional convolution layer with a kernel size of 8 and a stride of 1. It uses 64 convolution kernels and uses zero padding to keep the output length unchanged. The ReLU activation function is used for nonlinear mapping. The next layer is the series connection of five residual attention modules, where each residual attention module has the following structure: Figure 5 As shown in Figure 2. Each residual attention module consists of a one-dimensional convolution layer with a convolution kernel size of 15 and a stride of 1, using 64 convolution kernels, the same ReLU activation function, and an attention module. The structure of the attention module is shown in Figure 2. Figure 4 As shown in the figure, the global average pooling layer and the global maximum pooling layer are first used, and the results of the two pooling methods are added point by point to achieve the feature fusion of the two methods. Then there is a fully connected layer with a total of 16 neurons, which plays a role in dimensionality reduction. The activation function uses the ReLU function, followed by a fully connected layer. The number of neurons is the same as the number of input channels of the module, which is 64. The Sigmoid activation function is used to normalize the output structure, and the output result is multiplied by the input of the module to obtain the output of the attention module. The last layer of the network structure is the maximum pooling layer to complete the result output. The overall network structure is as follows Figure 6 As shown, a total of 318736 parameters are included.

[0072] Step 3: Use the training set to train the one-dimensional residual attention CNN model, and use the validation set to verify the network model after each round of training; compare the waveform predicted by the network with the waveform obtained by full-wave simulation to obtain the cable electromagnetic pulse radiation response prediction model.

[0073] It should be noted that the input data needs to be normalized before training, and the data is normalized to the interval [-1,1]. The loss function used in the training process is the mean square error loss, the optimization algorithm is the Adam algorithm, and the initial learning rate is 0.01.

[0074] Step 4: Input the sample points of the test set into the trained cable electromagnetic pulse response prediction model for prediction, and perform inverse normalization on the prediction results to obtain the final prediction results. The prediction results are compared with the electromagnetic pulse response voltage waveform of the cable obtained by the full-wave simulation software to achieve verification and obtain the prediction model.

[0075] In this embodiment, 544,500 training samples are used to train the model, 6075 groups of validation samples are used to validate the model, adjust the hyperparameters in the model, and perform a preliminary assessment of the model capability.

[0076] Figure 7 The prediction results of the cable electromagnetic radiation response prediction model of the one-dimensional residual attention CNN at the test sample point are as follows: the polarization angle α of the incident plane wave is -68°, the azimuth angle φ is 4°, the incident angle ψ is -40°, the transient electric field peak value E0 of the excitation source is 20 kV / m, and the center frequency f of the excitation source is 1. o is 0.5GHz; the cable length L is 54cm, the height H of the cable from the ground plane is 2.4cm; and the waveform of the high-power microwave corresponding to the test sample point.

[0077] On a 6-core 16G personal commercial desktop computer, the full-wave numerical calculation time is about 30s, while the cable electromagnetic radiation response prediction model using a one-dimensional residual attention CNN takes about 10ms to make a prediction.

[0078] The cable electromagnetic radiation response prediction method based on one-dimensional residual attention CNN provided by the present invention improves the calculation efficiency and breaks through the previous cable electromagnetic pulse response calculation under the fixed electromagnetic pulse threat that can only be refined to a point, so that the prediction model can better reflect the electromagnetic pulse threat suffered by the cable under actual circumstances, and realizes the rapid prediction of the cable electromagnetic pulse response, which provides a new solution for the electromagnetic pulse response prediction.

Claims

1. A cable electromagnetic radiation response prediction method based on one-dimensional residual attention CNN, characterized in that: The steps include: Step 1: Construct a cable electromagnetic radiation response prediction dataset ① Select the physical factors that affect the electromagnetic pulse response waveform of the cable as the input variable, and the electromagnetic pulse response waveform of the cable as the output variable; ② Select sample points from the input variable set; ③ Use the full-wave simulation algorithm to calculate the coupled voltage and current waveforms of the cable at each sample point; ④ Combining each sample point with the coupling voltage waveform corresponding to the sample point to construct a data set; and dividing the data set into a training set, a validation set, and a test set; Step 2: Construct a one-dimensional residual attention CNN network model, which includes a convolutional layer, a batch normalization layer, a ReLU activation function layer, five residual attention modules, and a maximum pooling layer; Step 3: Use the training set to train the one-dimensional residual attention CNN network model, and use the validation set to validate the network model after each round of training; compare the waveform predicted by the network with the waveform obtained by full-wave simulation to obtain the cable electromagnetic radiation response prediction model; Step 4: Input the sample points of the test set into the trained cable electromagnetic radiation response prediction model for prediction, and perform denormalization on the prediction results to obtain the final prediction results; The physical factors that affect the electromagnetic response waveform of the cable in step ① include transmission line parameters, excitation source parameters and excitation source waveform, where the transmission line parameters include cable length and the height of the cable from the ground plane , the excitation source parameters include the transient electric field peak , center frequency , angle of incidence , polarization angle and azimuth ; The process of selecting sample points in the input variable set in step ② is as follows: , angle of incidence , azimuth , the transient electric field peak value of the excitation source and the center frequency of the excitation source The polarization angle is uniformly sampled within the range of The range of variation is , angle of incidence The range of variation is , azimuth The range of variation is , the transient electric field peak value of the excitation source The range of variation is , the center frequency of the excitation source The range of variation is .

2. The cable electromagnetic radiation response prediction method based on one-dimensional residual attention CNN according to claim 1 is characterized in that: The process of selecting sample points in the input variable set in step ② is as follows: and the height of the cable from the ground plane The cable length is uniformly sampled within the range of The range of variation is , Cable height from ground plane The range of variation is .

3. The cable electromagnetic radiation response prediction method based on one-dimensional residual attention CNN according to claim 1 is characterized in that: The full-wave simulation algorithm used in step ③ includes the transmission line matrix method, the finite integration method and the time-domain finite difference method, and the response voltage and current waveforms of the cable are obtained through three-dimensional full-wave electromagnetic simulation.

Images