A method and system for evaluating electromagnetic interference effects of a drone data link

Abstract

The application relates to a method and system for evaluating electromagnetic interference effects of a UAV data link, and relates to the technical field of electromagnetic interference effect evaluation. The method comprises the following steps: adopting an SSA algorithm to optimize hyperparameters of a double-channel convolutional neural network according to a data set of a UAV data link to be trained, to obtain an SSA-DCNN model; and training the SSA-DCNN model according to the data set of the UAV data link to be trained, to obtain an SSA-DCNN prediction model, wherein the SSA-DCNN prediction model is used to determine actual interference performance grades of the UAV data link under electromagnetic interference, and the data set comprises an electromagnetic interference time-frequency spectrum, a data link performance parameter histogram and actual interference performance grades under electromagnetic interference. The application can improve the convergence speed and evaluation accuracy of the model.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electromagnetic interference effect evaluation, in particular to a UAV data link electromagnetic interference effect evaluation method and system. BACKGROUND

[0002] With the progress of UAV technology and cost reduction, UAV systems have developed rapidly in military, agriculture, forestry, law enforcement and other application fields. UAVs use data link equipment to control and transfer data. For over-the-horizon long-range UAVs, the reliability of data link communication is particularly important. However, there may be intentional or unintentional high-power electromagnetic interference in space, which threatens the stability of data link communication. Therefore, higher requirements are put forward for the intelligence and autonomy of UAV data link.

[0003] Electromagnetic interference effect modeling of data link usually uses a single and specific interference type to extract interference features. However, in a complex electromagnetic environment, the interference type is diverse and difficult to characterize with fixed parameters. In addition, the degree of interference on the data link is not only related to the type and power of the interference source, but also related to the working distance of the data link. Current electromagnetic interference effect modeling methods mainly include simulation analysis method, mechanism modeling method and machine learning method. The simulation analysis method and mechanism modeling method depend on detailed prior knowledge of the structure of the tested system, and the scalability of the model is limited, which makes it difficult to be applied in actual scenarios full of changes. With the development of artificial intelligence in various fields, machine learning methods have attracted widespread attention and application in the fields of electromagnetic compatibility and communication.

[0004] Convolutional neural networks are proposed by Hubel and Wiesel through the study of local sensitivity and directional selection of cat cerebral cortex neurons. Since the network avoids complex preprocessing of images and has certain deep learning ability, it can directly input original images, so it has been widely applied. However, with the increase of network layers, the complexity of hyperparameter dimension also increases greatly, and certain methods need to be taken to select the optimal combination of hyperparameters. Generally speaking, there are many methods for hyperparameter optimization, such as manual tuning, grid search and Bayesian optimization method, etc. However, when adjusting the hyperparameters of the convolutional neural network, the convergence speed and evaluation accuracy of these methods are low, which causes the accuracy of the interference performance level obtained by subsequent interference evaluation to be low. SUMMARY

[0005] The purpose of the present application is to provide a UAV data link electromagnetic interference effect evaluation method and system. The electromagnetic interference effect evaluation method provided by the present application can improve the convergence speed and evaluation accuracy of the model. The interference performance level of the UAV data link obtained by evaluation can be used to evaluate the interference degree of the UAV data link, and provide decision basis for subsequent anti-interference measures.

[0006] To achieve the above object, the present application provides the following scheme:

[0007] An unmanned aerial vehicle data link electromagnetic interference effect evaluation method, comprising:

[0008] Obtaining a data set of a to-be-trained unmanned aerial vehicle data link, the data set comprising: an electromagnetic interference time-frequency spectrum, a data link performance parameter histogram, and an actual interference performance level under electromagnetic interference;

[0009] According to the data set of the to-be-trained unmanned aerial vehicle data link, the super parameters of a double-channel convolutional neural network are optimized by using an SSA algorithm to obtain an SSA-DCNN model;

[0010] According to the data set of the to-be-trained unmanned aerial vehicle data link, the SSA-DCNN model is trained to obtain an SSA-DCNN prediction model, which is used to determine the actual interference performance level of the unmanned aerial vehicle data link under electromagnetic interference.

[0011] Optionally, the data set of the to-be-trained unmanned aerial vehicle data link is obtained, specifically comprising:

[0012] Obtaining IQ data, performance parameters, and actual interference performance levels under electromagnetic interference of electromagnetic signals of the to-be-trained unmanned aerial vehicle data link, the performance parameters comprising signal gain control, signal-to-noise ratio, and bit error rate;

[0013] According to the IQ data of the electromagnetic signals, an electromagnetic interference time-frequency spectrum is obtained;

[0014] According to the performance parameters, a data link performance parameter histogram is obtained.

[0015] Optionally, the double-channel convolutional neural network comprises: a convolutional network, an addition layer, a third full connection layer, and a regression layer connected in sequence; the convolutional network comprises a first convolutional module and a second convolutional module; the first convolutional module comprises a first convolutional layer, a first maximum pooling layer, a third convolutional layer, a first RELU activation function, and a first full connection layer connected in sequence; the second convolutional module comprises a second convolutional layer, a second maximum pooling layer, a fourth convolutional layer, a second RELU activation function, and a second full connection layer connected in sequence; the first full connection layer and the second full connection layer are connected with the addition layer.

[0016] Optionally, the super parameters of the double-channel convolutional neural network are optimized by using the SSA algorithm to obtain the SSA-DCNN model according to the data set of the to-be-trained unmanned aerial vehicle data link, specifically comprising:

[0017] input the electromagnetic interference time-frequency spectrogram into a first convolutional layer of the dual-channel convolutional neural network, input the data link performance parameter histogram into a second convolutional layer of the dual-channel convolutional neural network, to obtain a test interference performance level;

[0018] The super parameter of the dual-channel convolutional neural network is optimized by using an SSA algorithm to obtain an SSA-DCNN model, with the root mean square error of the test interference performance level and the actual interference performance level under electromagnetic interference being minimized as the target.

[0019] An unmanned aerial vehicle data link electromagnetic interference effect evaluation system comprises:

[0020] An acquisition module is configured to acquire a data set of a to-be-trained unmanned aerial vehicle data link, the data set comprising an electromagnetic interference time-frequency spectrogram, a data link performance parameter histogram, and an actual interference performance level under electromagnetic interference.

[0021] A super parameter adjustment module is configured to optimize the super parameter of a dual-channel convolutional neural network by using an SSA algorithm according to the data set of the to-be-trained unmanned aerial vehicle data link to obtain an SSA-DCNN model.

[0022] A training module is configured to train the SSA-DCNN model according to the data set of the to-be-trained unmanned aerial vehicle data link to obtain an SSA-DCNN prediction model, the SSA-DCNN prediction model being configured to determine the actual interference performance level of the unmanned aerial vehicle data link under electromagnetic interference.

[0023] Optionally, the acquisition module specifically comprises:

[0024] An acquisition unit is configured to acquire IQ data of an electromagnetic signal of a to-be-trained unmanned aerial vehicle data link, performance parameters, and an actual interference performance level under electromagnetic interference, the performance parameters comprising signal gain control, signal-to-noise ratio, and bit error rate.

[0025] A time-frequency spectrogram determination unit is configured to obtain an electromagnetic interference time-frequency spectrogram according to the IQ data of the electromagnetic signal.

[0026] A histogram determination unit is configured to obtain a data link performance parameter histogram according to the performance parameters.

[0027] Optionally, the double-channel convolutional neural network comprises a convolutional network, an addition layer, a third fully connected layer and a regression layer connected in sequence; the convolutional network comprises a first convolutional module and a second convolutional module; the first convolutional module comprises a first convolutional layer, a first max-pooling layer, a third convolutional layer, a first RELU activation function and a first fully connected layer connected in sequence; the second convolutional module comprises a second convolutional layer, a first max-pooling layer, a fourth convolutional layer, a second RELU activation function and a second fully connected layer connected in sequence; the first fully connected layer and the second fully connected layer are connected with the addition layer.

[0028] Optionally, the hyperparameter adjustment module specifically comprises:

[0029] The initialization unit is configured to input the electromagnetic interference time-frequency spectrogram into a first convolutional layer of the double-channel convolutional neural network and input the data link performance parameter histogram into a second convolutional layer of the double-channel convolutional neural network to obtain a test interference performance level.

[0030] The hyperparameter adjustment unit is configured to adopt an SSA algorithm to optimize hyperparameters of the double-channel convolutional neural network to obtain an SSA-DCNN model, with the root mean square error of the test interference performance level and an actual interference performance level under electromagnetic interference being minimized as a target.

[0031] According to the embodiments of the present application, the following technical effects are achieved: the present application comprises obtaining a data set of a to-be-trained unmanned aerial vehicle data link, the data set comprising an electromagnetic interference time-frequency spectrogram, a data link performance parameter histogram and an actual interference performance level under electromagnetic interference; an SSA algorithm is adopted to optimize hyperparameters of a double-channel convolutional neural network according to the data set of the to-be-trained unmanned aerial vehicle data link to obtain an SSA-DCNN model; the SSA-DCNN model is trained according to the data set of the to-be-trained unmanned aerial vehicle data link to obtain an SSA-DCNN prediction model, which is used to determine the actual interference performance level of the unmanned aerial vehicle data link under electromagnetic interference; and the optimization of the hyperparameters of the double-channel convolutional neural network by the SSA algorithm can improve the convergence speed and evaluation accuracy of the model. BRIEF DESCRIPTION OF DRAWINGS

[0032] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0033] Figure 1 The flow chart of the unmanned aerial vehicle data link electromagnetic interference effect evaluation method provided by the embodiments of the present application is shown in the drawings.

[0034] Figure 2 A schematic diagram of SSA-DCNN model prediction provided by the embodiment of the present application;

[0035] Figure 3 A structure block diagram of an injection data link electromagnetic interference collection system provided by the embodiment of the present application;

[0036] Figure 4 A flowchart of SSA-DCNN model prediction provided by the embodiment of the present application;

[0037] Figure 5 A comparison diagram of loss function values of DCNN model and SSA-DCNN model;

[0038] Figure 6 A comparison diagram of root mean square error values of DCNN model and SSA-DCNN model. DETAILED DESCRIPTION

[0039] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0040] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0041] Sparrow Search Algorithm (SSA) is a bionic group design optimization method based on the foraging and anti-predation of sparrows, which has the advantages of strong optimization ability, fast convergence speed and strong stability, and can be applied to the hyperparameter optimization of a model. Based on this, the embodiment of the present application provides an unmanned aerial vehicle data link electromagnetic interference effect evaluation method, as shown in Figure 1 、 Figure 2 and Figure 4 , which comprises:

[0042] Step 101: obtaining a data set of a to-be-trained unmanned aerial vehicle data link, the data set comprising: an electromagnetic interference time-frequency spectrum, a data link performance parameter histogram and an actual interference performance grade under electromagnetic interference.

[0043] Step 102: using an SSA algorithm to optimize hyperparameters of a double-channel convolutional neural network according to the data set of the to-be-trained unmanned aerial vehicle data link to obtain a SSA-DCNN model.

[0044] Step 103: training the SSA-DCNN model according to the data set of the unmanned aerial vehicle data link to be trained to obtain an SSA-DCNN prediction model, and the SSA-DCNN prediction model is used to determine the actual interference performance level of the unmanned aerial vehicle data link under electromagnetic interference.

[0045] In practical applications, step 101 specifically includes:

[0046] IQ data, performance parameters, and actual interference performance levels under electromagnetic interference of electromagnetic signals of the unmanned aerial vehicle data link to be trained are obtained, and the performance parameters include signal gain control, signal-to-noise ratio, and bit error rate.

[0047] The time-frequency spectrum diagram of electromagnetic interference is obtained according to the IQ data of the electromagnetic signals.

[0048] The data link performance parameter histogram is obtained according to the performance parameters.

[0049] In practical applications, the equivalent injection test method can be used to obtain a large amount of IQ data and performance parameters of electromagnetic signals.

[0050] In practical applications, the double-channel convolutional neural network includes a convolutional network, an addition layer, a third full connection layer, and a regression layer connected in sequence; the convolutional network includes a first convolutional module and a second convolutional module; the first convolutional module includes a first convolutional layer, a first maximum pooling layer, a third convolutional layer, a first RELU activation function, and a first full connection layer connected in sequence; the second convolutional module includes a second convolutional layer, a first maximum pooling layer, a fourth convolutional layer, a second RELU activation function, and a second full connection layer connected in sequence; the first full connection layer and the second full connection layer are connected with the addition layer, two CNN channels are independent of each other, and then a series of dimension reduction and feature extraction are performed to obtain feature vectors with the same dimension, which are combined in the addition layer, wherein the convolutional layer and the pooling layer are core modules for realizing the feature extraction function of the convolutional neural network, the kernel size and the number of the convolutional layer and the pooling layer are set independently, the network model performs feedback adjustment on the weight parameters in the network layer by layer by minimizing the loss function, and the accuracy of the network is improved through frequent iteration training.

[0051] In practical applications, step 102 specifically includes:

[0052] The time-frequency spectrum diagram of electromagnetic interference is input into the first convolutional layer of the double-channel convolutional neural network, and the data link performance parameter histogram is input into the second convolutional layer of the double-channel convolutional neural network to obtain a test interference performance level.

[0053] The SSA algorithm is adopted to optimize the hyperparameters of the double-channel convolutional neural network to obtain an SSA-DCNN model, with the minimum root mean square error of the test interference performance level and the actual interference performance level under electromagnetic interference as the target, specifically: on the basis of the DCNN network structure, the combination of hyperparameters is biomimeticized as a sparrow population. In the process of finding the optimal hyperparameter combination, model training and evaluation are required each time. The root mean square error is taken as the objective function to evaluate the network performance, and the hyperparameters are iteratively trained multiple times within a certain optimization space, the root mean square error between the test interference performance level and the actual interference performance level under different hyperparameter combinations is calculated, and the hyperparameter combination that minimizes the root mean square error is obtained.

[0054] The embodiment of the application also provides an unmanned aerial vehicle data link electromagnetic interference effect evaluation system corresponding to the above method, comprising:

[0055] The acquisition module is configured to acquire a data set of a to-be-trained unmanned aerial vehicle data link, the data set comprising: an electromagnetic interference time-frequency spectrum, a data link performance parameter histogram, and an actual interference performance level under electromagnetic interference.

[0056] The hyperparameter adjustment module is configured to optimize hyperparameters of a double-channel convolutional neural network to obtain an SSA-DCNN model by using an SSA algorithm according to the data set of the to-be-trained unmanned aerial vehicle data link.

[0057] The training module is configured to train the SSA-DCNN model to obtain an SSA-DCNN prediction model according to the data set of the to-be-trained unmanned aerial vehicle data link, and the SSA-DCNN prediction model is configured to determine an actual interference performance level of the unmanned aerial vehicle data link under electromagnetic interference.

[0058] As an optional implementation, the acquisition module specifically comprises:

[0059] The acquisition unit is configured to acquire IQ data, performance parameters, and an actual interference performance level under electromagnetic interference of an electromagnetic signal of a to-be-trained unmanned aerial vehicle data link, and the performance parameters comprise signal gain control, signal-to-noise ratio, and bit error rate.

[0060] The time-frequency spectrum determination unit is configured to obtain an electromagnetic interference time-frequency spectrum according to the IQ data of the electromagnetic signal.

[0061] The histogram determination unit is configured to obtain a data link performance parameter histogram according to the performance parameters.

[0062] As an optional implementation, the double-channel convolutional neural network comprises a convolutional network, an addition layer, a third full connection layer and a regression layer connected in sequence; the convolutional network comprises a first convolutional module and a second convolutional module; the first convolutional module comprises a first convolutional layer, a first max-pooling layer, a third convolutional layer, a first RELU activation function and a first full connection layer connected in sequence; the second convolutional module comprises a second convolutional layer, a first max-pooling layer, a fourth convolutional layer, a second RELU activation function and a second full connection layer connected in sequence; the first full connection layer and the second full connection layer are connected with the addition layer.

[0063] As an optional implementation, the hyperparameter adjustment module specifically comprises:

[0064] The initialization unit is configured to input the electromagnetic interference time-frequency spectrogram into a first convolutional layer of the double-channel convolutional neural network and input the data link performance parameter histogram into a second convolutional layer of the double-channel convolutional neural network to obtain a test interference performance level.

[0065] The hyperparameter adjustment unit is configured to adopt an SSA algorithm to optimize hyperparameters of the double-channel convolutional neural network to obtain an SSA-DCNN model, with the root mean square error of the test interference performance level and an actual interference performance level under electromagnetic interference being minimized as a target.

[0066] The application further provides an embodiment of applying the above method in practice:

[0067] Step 1: Obtain an electromagnetic interference time-frequency spectrogram and a data link performance parameter histogram sample set, pre-process the sample set, and divide the pre-processed sample set into a training set and a test set.

[0068] Step 2: Construct an SSA-DCNN model. First, build a double-channel convolutional neural network (DCNN), and then optimize hyperparameters by using an SSA method to obtain an SSA-DCNN model (SSA-DCNN is a double-channel convolutional neural network with optimized hyperparameters by using an SSA method. The data link is divided into four performance boundary levels according to the interference condition, which respectively represent the performance decline process of the data link from normal operation to communication interruption. The model adopts a double-channel convolutional neural network to fuse the electromagnetic interference time-frequency spectrogram and the data link performance parameter histogram, and predict the performance level of the data link under interference. The input of the model is the electromagnetic interference time-frequency spectrogram and the data link performance parameter histogram, and the output is the interference level of the data link under electromagnetic interference).

[0069] Step 3: Input the training set into the SSA-DCNN model for training to obtain an SSA-DCNN prediction model.

[0070] Step 4: input the test set into the SSA-DCNN prediction model, predict the data link interference level in the test set to obtain the test interference performance level, calculate the prediction error according to the test interference performance level and the actual interference performance level, and determine the optimal SSA-DCNN prediction model according to the prediction error.

[0071] Step 1 is specifically: injecting electromagnetic interference test on the unmanned aerial vehicle data link, as shown in Figure 3 , through the zero intermediate frequency sampling of Figure 3 , the IQ data of the electromagnetic signal is obtained. In order to facilitate the extraction of the characteristics of different interference sources, the IQ data of the electromagnetic space signal is converted into a time-frequency spectrum, and the time-frequency spectrum is used as a visual representation of the power of the observation signal changing with frequency and time.

[0072] The signal gain control AGC, the signal-to-noise ratio SNR and the bit error rate BER are three parameters respectively representing the state of the received signal in the amplification filtering, capture tracking and demodulation decoding processes. When the interference signal and the working signal enter the receiver together, the AGC module at the radio frequency front end will cause the gain compression of the useful signal, which is reflected in the received AGC voltage value. Then, after the signal is demodulated at the receiving end, the synchronization information is obtained through real-time capture and tracking, and the SNR is calculated at this time. Finally, after data demodulation and decoding, telemetry information is obtained, and the BER of the data link is also obtained. Joint three performance parameters can be used to evaluate the interference degree of the data link. It is stipulated that the loss of lock is the upper limit of the interference state of the data link, and according to the data characteristics of SNR, BER and AGC, they have obvious and different distribution boundaries, which are: SNR∈[0, 20], BER∈[0, 3000], AGC∈[0, 255]. Therefore, the data is normalized to obtain the normalized values of the three types of parameters:

[0073]

[0074] wherein x normalization is the normalized value, x is the value before normalization, x max , x min are the upper and lower limits of the distribution boundary of the parameter respectively. In order to unify the representation method of the three parameters and facilitate modeling, the normalized histogram of the performance parameters of the data link changing dynamically with the interference state is drawn.

[0075] In order to predict the electromagnetic interference level of the data link, the interference is divided into four levels according to the loss of lock, which are signal power distance data link loss of lock 0dB, -1dB, -3dB and -6dB. According to the electromagnetic interference level, the electromagnetic interference time-frequency spectrum and the data link performance parameter histogram are labeled to obtain the data set of the model.

[0076] Step 2 is specifically:

[0077] Step 2.1: Establish a dual-channel convolutional neural network (DCNN) to extract features of electromagnetic interference time-frequency spectrograms and data link performance parameter normalized histograms through two channels, respectively, and fully integrate the characteristics of the two kinds of data.

[0078] Step 2.2: Determine the optimization space S of the DCNN network hyperparameters, as shown in Table 1.

[0079] Table 1 Hyperparameter optimization range settings

[0080] Layer parameters Range Convolutional layer 1 / 2 size [1,10] Convolutional layer 1 / 2 number [1,90] Max pooling layer 1 / 2 size [1,20] Max pooling layer 1 / 2 number [1,50] Convolutional layer 3 / 4 size [1,10] Convolutional layer 3 / 4 number [1,90] Fully connected layer 1 / 2 size [1,20] Learning rate [[10 -7 ,10 -5 ]]]> Training times [5,10] Mini batch [1,10]

[0081] Step 2.3: Use the Sparrow Search Algorithm (SSA) to optimize the hyperparameters, and biomimetic the combination of hyperparameters as a sparrow population. Take the root mean square error f RMSE as the objective function to evaluate network performance, and perform multiple iterations on the hyperparameters x within a certain optimization space S to calculate the root mean square error between the test interference performance level y i and the actual interference performance level t i , as follows, N train is the total number of training model samples, and i represents the number of training model samples.

[0082]

[0083] Step 2.4: Obtain the hyperparameter combination that minimizes the root mean square error, which is the optimized hyperparameter of the network, as shown in Table 2.

[0084] Table 2 DCNN hyperparameters optimized by SSA

[0085]

[0086] Step 3 is as follows:

[0087] The input data of the two channels each has 484 samples, and the samples of each channel are divided into training set and test set according to the ratio of 7:3. The prediction target is the interference level label {0, 1, 3, 6}. We use SGDM, RMSprop and Adam solvers to optimize the DCNN and SSA-DCNN models, respectively, to train the input data. In order to compare the influence of different solvers on the model, only the solver type is changed without changing the model structure and parameters, and then the convergence speed and training results of the models of different solvers are compared, as shown in Figure 5 and Figure 6 , where Figure 5 (a) is the loss value curve of DCNN optimized by SGDM, RMSprop and Adam solvers, respectively,Figure 5 (b) Loss value curve of SSA-DCNN optimized by SGDM, RMSprop and Adam solver respectively, Figure 6 (a) Root mean square error value curve of DCNN optimized by SGDM, RMSprop and Adam solver respectively, Figure 6 (b) Root mean square error value curve of SSA-DCNN optimized by SGDM, RMSprop and Adam solver respectively. By Figure 5 and Figure 6 It can be concluded that the DCNN model optimized by SSA has higher stability, faster convergence speed, smaller loss function value and root mean square error value after convergence, and obvious performance improvement.

[0088] Step 4 is specifically:

[0089] In order to further analyze the prediction accuracy of the model, we introduce four evaluation indexes of the model, and then analyze the results of each model, which are the determination coefficient R 2 , mean absolute percentage error MAPE, root mean square error and accuracy.

[0090] According to the calculation of the above indexes by the model test set, table 3 is obtained. From the evaluation indexes of different optimization algorithms, in the DCNN model without using SSA optimization, the accuracy of the DCNN model optimized by SGDM is 79.17%, and other index parameters are obviously better than the other two optimization methods. The prediction effect of the DCNN model optimized by SSA is obviously better than that of the model without optimization. Among them, the prediction accuracy of the SSA-DSNN model optimized by RMSprop is the highest, which is 97.92%, and the accuracy of the SSA-DCNN model optimized by SGDM is slightly lower, which is 97.22%. But compared with the SSA-DCNN optimized by RMSprop, the root mean square error of the SSA-DCNN optimized by SGDM is smaller, which is 0.3882, the R 2 coefficient is 0.88891, which is closer to 1, and the MAPE value is about 12.5%, which is the lowest in all models.

[0091] Table 3 calculation results of model evaluation indexes

[0092]

[0093] In summary, from the analysis of the training process of the model and the evaluation index, it can be concluded that the DCNN model without hyperparameter optimization is likely to fall into a local minimum in the training process, resulting in non-convergence of the model. The model convergence speed of the SSA-optimized DCNN is faster and more stable, the prediction accuracy of the SSA-DCNN using the RMSProp solver is the highest, and the model convergence process is stable; the SSA-DCNN optimized by SGDM has the best comprehensive index, and the model converges the fastest and has the best performance.

[0094] The present application has the following technical effects:

[0095] The present application adopts an equivalent injection test method to collect a large amount of electromagnetic interference signal IQ data and data link performance parameters, converts them into time-frequency spectrograms and data link performance parameter histograms through preprocessing as model inputs, and then constructs a double-channel convolutional neural network (SSA-DCNN) based on SSA optimization to predict the data link electromagnetic interference effect level. The new method provided has good compatibility for different types of electromagnetic interference, can avoid manual extraction of electromagnetic interference data features, and improves the intelligent level of data link electromagnetic interference perception and evaluation.

[0096] The present application also compares the model performance of DCNN and SSA-DCNN under SGDM, RMSProp and Adam solvers. The results show that compared with DCNN, the SSA-DCNN model can significantly improve the convergence speed and prediction accuracy of model training, and the SSA-DCNN optimized by SGDM has the best performance.

[0097] The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other. For the system disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method part.

[0098] The principles and implementation modes of the present application are described by applying specific examples in this paper, and the above embodiment description is only used to help understand the method and core idea of the present application; at the same time, for those skilled in the art, according to the idea of the present application, the specific implementation mode and application range will be changed. In summary, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A method for evaluating the electromagnetic interference effect of a UAV data link, characterized in that, include: Obtain the dataset of the UAV data link to be trained, which includes: electromagnetic interference spectrum, data link performance parameter histogram, and actual interference performance level under electromagnetic interference; The SSA-DCNN model is obtained by optimizing the hyperparameters of the dual-channel convolutional neural network using the SSA algorithm based on the dataset of the UAV data link to be trained. The SSA-DCNN model is trained on the dataset of the UAV data link to be trained to obtain the SSA-DCNN prediction model, which is used to determine the actual interference performance level of the UAV data link under electromagnetic interference. The step of optimizing the hyperparameters of the dual-channel convolutional neural network using the SSA algorithm based on the dataset of the UAV data link to be trained to obtain the SSA-DCNN model specifically includes: The electromagnetic interference spectrum is input into the first convolutional layer of the dual-channel convolutional neural network, and the data link performance parameter histogram is input into the second convolutional layer of the dual-channel convolutional neural network to obtain the test interference performance level. With the goal of minimizing the root mean square error of the test interference performance level and the actual interference performance level under electromagnetic interference, the hyperparameters of the dual-channel convolutional neural network are tuned using the SSA algorithm to obtain the SSA-DCNN model; the SSA algorithm is the Sparrow Search algorithm.

2. The method for evaluating the electromagnetic interference effect of a UAV data link according to claim 1, characterized in that, The acquisition of the dataset for the UAV data link to be trained specifically includes: Acquire the IQ data, performance parameters, and actual interference performance level of the electromagnetic signals of the UAV data link to be trained. The performance parameters include signal gain control, signal-to-noise ratio, and bit error rate. The electromagnetic interference spectrum is obtained based on the IQ data of the electromagnetic signal. A histogram of data link performance parameters is obtained based on the aforementioned performance parameters.

3. The method for evaluating the electromagnetic interference effect of a UAV data link according to claim 1, characterized in that, The dual-channel convolutional neural network includes: a convolutional network, an additive layer, a third fully connected layer, and a regression layer connected in sequence; the convolutional network includes a first convolutional module and a second convolutional module; the first convolutional module includes a first convolutional layer, a first max pooling layer, a third convolutional layer, a first ReLU activation function, and a first fully connected layer connected in sequence; the second convolutional module includes a second convolutional layer, a first max pooling layer, a fourth convolutional layer, a second ReLU activation function, and a second fully connected layer connected in sequence; both the first fully connected layer and the second fully connected layer are connected to the additive layer.

4. A system for evaluating the electromagnetic interference effects of UAV data links, characterized in that, include: The acquisition module is used to acquire the dataset of the UAV data link to be trained. The dataset includes: electromagnetic interference spectrum, data link performance parameter histogram, and actual interference performance level under electromagnetic interference. The hyperparameter tuning module is used to tune the hyperparameters of the dual-channel convolutional neural network using the SSA algorithm based on the dataset of the UAV data link to be trained, so as to obtain the SSA-DCNN model. The training module is used to train the SSA-DCNN model based on the dataset of the UAV data link to be trained to obtain an SSA-DCNN prediction model, which is used to determine the actual interference performance level of the UAV data link under electromagnetic interference. The hyperparameter adjustment module specifically includes: An initialization unit is used to input the electromagnetic interference spectrum into the first convolutional layer of the dual-channel convolutional neural network and input the data link performance parameter histogram into the second convolutional layer of the dual-channel convolutional neural network to obtain the test interference performance level. The hyperparameter tuning unit is used to fine-tune the hyperparameters of the dual-channel convolutional neural network to obtain the SSA-DCNN model with the goal of minimizing the root mean square error of the test interference performance level and the actual interference performance level under electromagnetic interference; the SSA algorithm is the Sparrow Search algorithm.

5. The UAV data link electromagnetic interference effect assessment system according to claim 4, characterized in that, The acquisition module specifically includes: The acquisition unit is used to acquire the IQ data, performance parameters, and actual interference performance level of the electromagnetic signal of the UAV data link to be trained. The performance parameters include signal gain control, signal-to-noise ratio, and bit error rate. The time-spectrum diagram determination unit is used to obtain the electromagnetic interference time-spectrum diagram based on the IQ data of the electromagnetic signal; The histogram determination unit is used to obtain a histogram of data link performance parameters based on the performance parameters.

6. The UAV data link electromagnetic interference effect assessment system according to claim 4, characterized in that, The dual-channel convolutional neural network includes: a convolutional network, an additive layer, a third fully connected layer, and a regression layer connected in sequence; the convolutional network includes a first convolutional module and a second convolutional module; the first convolutional module includes a first convolutional layer, a first max pooling layer, a third convolutional layer, a first ReLU activation function, and a first fully connected layer connected in sequence; the second convolutional module includes a second convolutional layer, a first max pooling layer, a fourth convolutional layer, a second ReLU activation function, and a second fully connected layer connected in sequence; both the first fully connected layer and the second fully connected layer are connected to the additive layer.

Images