A dynamic interference behavior and intent emulation and prediction method
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HARBIN INST OF TECH
- Filing Date
- 2023-07-05
- Publication Date
- 2026-07-03
Smart Images
Figure CN116840790B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of radar anti-jamming technology, and in particular to a method for simulating and predicting dynamic jamming behavior and intent. Background Technology
[0002] Radar is an indispensable and crucial component of the modern battlefield environment. Jammers attempt to disrupt radar's normal detection capabilities to gain the initiative in battlefield information. In the game between radar and jamming, jammers perceive changes in the radar's operational status and electromagnetic environment, then target and interfere with the radar to reduce its effectiveness in acquiring battlefield information, obscure their own operational intentions, and disrupt its normal operation. In jamming countermeasures, to achieve a specific jamming objective, jammers need to respond to the electromagnetic environment by allocating internal resources and radiating external signals—this is jamming behavior, specifically manifested as emitting jamming signals of different patterns and modulations.
[0003] Analyzing and studying information about the jammer during the jamming process, such as the jamming behavior and intent, requires the support of real data. However, since the jammer's behavior and intent are implicit information that is not visible to the other party, it is currently impossible to obtain data on the jamming behavior and intent. Therefore, it is currently impossible to predict the jamming intent, resulting in the radar lacking initiative and accuracy in countering jamming. Summary of the Invention
[0004] The purpose of this invention is to solve the problem that current radar anti-jamming methods cannot obtain data on jamming behavior and intent, which leads to the inability to predict jamming intent and consequently results in a lack of initiative and low accuracy in radar anti-jamming. Therefore, this invention proposes a dynamic jamming behavior and intent simulation and prediction method.
[0005] The specific process of a dynamic interference behavior and intent simulation and prediction method is as follows: acquire historical moment variable data, input the historical moment variable data into the interference intent prediction model, and obtain the interference intent at the next moment;
[0006] The historical moment variable data includes: time information, the interference intent of the historical moment, and the interference behavior of the historical moment;
[0007] The interference intent prediction model is obtained through the following method:
[0008] Step 1: Establish the jamming scenario and set the parameters of the radar, jammer, and target in the jamming scenario;
[0009] Step 2: Based on the interference scenario established in Step 1, simulate the dynamic confrontation process between the radar and the jammer to generate the jamming intent and the corresponding jamming behavior.
[0010] Step 3: Obtain a interference intent prediction model by utilizing the interference intent and the corresponding interference behavior.
[0011] Furthermore, in step two, based on the interference scenario established in step one, the dynamic detection process of the radar on the target airspace is simulated to generate interference intent and the corresponding interference behavior, specifically as follows:
[0012] Step 2: Based on the interference scenario established in Step 1, simulate the dynamic detection process of the radar in the target airspace, and generate a radar operating mode sequence using the Markov property of the multi-function radar operating mode.
[0013] Step 22: Generate jamming intent using radar operating mode sequences, input the jamming intent into the jamming behavior decision model, and generate the jamming behavior corresponding to the jamming intent.
[0014] Furthermore, the step of generating jamming intent using radar operating mode sequences, inputting the jamming intent into the jamming behavior decision model, and generating jamming behavior corresponding to the jamming intent specifically involves:
[0015] Step 221: Set the recognition probability of the jammer for the radar operating mode, input the radar operating mode sequence into the jammer, obtain the jammer's recognition result of the radar operating mode, and convert the jammer's recognition result of the radar operating mode into the jamming intention according to the correspondence between the recognition result and the jamming intention.
[0016] Step 222: Set the conditional transition probabilities for the interference behavior decision model;
[0017] Step 223: Use the interference machine to obtain the target feature attributes, perform probability membership fuzzy processing on the target feature attributes to obtain the probability vector, and then input the probability vector into the interference behavior decision model after setting the transition probability to obtain the support confidence of each interference behavior, and output the interference behavior with the highest support confidence.
[0018] The target's characteristic attributes include: the target's state information and interference intent.
[0019] Furthermore, the target's status information includes: target distance and target orientation.
[0020] Furthermore, the radar operating modes include: search, confirm, track, and identify.
[0021] Furthermore, the correspondence between the identification results and the interference intent is specifically as follows: search - reduce detection, confirmation - affect confirmation, tracking - get rid of tracking, identification - destroy identification.
[0022] Furthermore, the interference behavior decision model is a Bayesian network model.
[0023] Furthermore, the step three, which involves obtaining a interference intent prediction model using the interference intent and the corresponding interference behavior, specifically includes:
[0024] Step 3: First, digitally encode the interference intent and interference behavior obtained in Step 2. Based on the relationship between the interference intent and interference behavior obtained in Step 2, obtain an encoded dataset consisting of the interference intent encoding and the interference behavior encoding corresponding to the interference intent.
[0025] Step 32: Add the time variable to the encoded dataset to obtain the sample dataset. Normalize the sample dataset and divide the normalized sample dataset into training set and test set.
[0026] Step 3: Train the LSTM network using the training set to obtain the trained LSTM network;
[0027] Steps 3 and 4: Test the trained LSTM network using the test set. If the accuracy of the trained LSTM network is greater than or equal to the preset threshold, then use the currently trained LSTM network as the interference intention prediction model. If the accuracy of the trained LSTM network is less than the preset threshold, then repeat step 1 until the accuracy of the trained LSTM network is greater than or equal to the preset threshold.
[0028] Furthermore, the interference intent obtained in step two is specifically encoded as follows: the interference intents of reducing detection, influencing confirmation, escaping tracking, and disrupting identification are encoded as 1, 2, 3, and 4 respectively.
[0029] Furthermore, the interference behavior obtained in step two is digitally encoded, specifically as follows:
[0030] The interference behaviors of suppression, deception, dragging, suppression + deception, and deception + deception are sequentially coded as 1, 2, 3, 4, and 5.
[0031] The beneficial effects of this invention are as follows:
[0032] This invention constructs a dynamic jamming scenario, intuitively displaying the confrontation process between radar and the jammer. It employs a Bayesian network-based jamming strategy to simulate the jammer's jamming decision-making process, generating and recording jamming behaviors and intent sequences. By setting parameters for the jamming scenario, it simulates jamming behaviors and intent information under different scenarios, thereby obtaining potential jamming intent information. Based on simulation data, this invention uses LSTM to predict jamming intent. After parameter optimization, the jamming intent prediction model based on a Long Short-Term Memory Neural Network achieves an accuracy of 85% in predicting jamming intent from simulation data, demonstrating excellent prediction performance. This assists combat personnel in making rapid and accurate judgments about enemy jamming intent, thereby improving the initiative and accuracy of radar counter-jamming and enabling agile operations. Attached Figure Description
[0033] Figure 1 This is a flowchart of the simulation process of the present invention;
[0034] Figure 2 This is a schematic diagram of an interference scenario;
[0035] Figure 3 This is the target display interface for the radar.
[0036] Figure 4 A diagram illustrating interference in the decision-making process;
[0037] Figure 5(a) shows the RD spectrum of the range decoy plus the velocity decoy;
[0038] Figure 5(b) shows the radar interface affected by range decoys and velocity decoys;
[0039] Figure 6 Example of the movement trajectory of the jammer;
[0040] Figure 7 The transition probability of the radar operating mode;
[0041] Figure 8(a) shows the sequence of disturbance behaviors generated by the simulation.
[0042] Figure 8(b) shows the simulated interference sequence.
[0043] Figure 8(c) shows the time-domain signal of the interference signal pattern generated in a certain round;
[0044] Figure 9 A schematic diagram of constructing a dataset for interfering intents;
[0045] Figure 10(a) shows the relationship between prediction accuracy and the number of iterations;
[0046] Figure 10(b) shows the relationship between the loss function and the number of iterations. Detailed Implementation
[0047] Specific Implementation Method 1: This implementation method is a dynamic interference behavior and intent simulation and prediction method, including the following steps: acquiring historical moment variable data, and inputting the historical moment variable data into the interference intent prediction model to obtain the interference intent at the next moment;
[0048] The historical moment variable data includes: time information, the interference intent of the historical moment, and the interference behavior of the historical moment;
[0049] The interference intent prediction model is obtained through the following method:
[0050] Step 1: In the simulation system, establish an interference scenario and set the parameters of the radar, jammer, and target in the interference scenario, specifically as follows:
[0051] Interference scenarios such as Figure 2 As shown, a simulation of an interference scenario is performed, in which elements include a multi-functional ground-based air defense radar and a formation of penetrating aircraft.
[0052] The multi-functional ground-based air defense radar is used to perform functions such as search, identification, tracking, and velocity detection. This multi-functional radar is equipped with a phased array antenna, and the airspace it scans can be changed by setting the azimuth and elevation angles.
[0053] The penetrating aircraft formation includes an electronic warfare aircraft that provides accompanying jamming support and a fighter jet under cover.
[0054] The simulation system includes: a fighter jet and jammer trajectory setting unit, an echo generation module, a jamming signal pattern generation and parameter setting module, a signal processing module, and a radar display module;
[0055] The fighter jet and jammer trajectory setting unit is used to set the trajectories of the fighter jet and jammer, and send the parameters of the completed trajectory setting to the echo generation module;
[0056] The trajectories of the fighter jet and the jammer include initial position, speed, direction, etc., among which the direction of movement includes three types: forward, side and backward towards the radar;
[0057] The echo generation module is used to simulate and generate fighter jet and jammer echo A after the radar antenna array element receives the set trajectory, and input the fighter jet and jammer echo A after the set trajectory into the jamming signal pattern generation and parameter setting module.
[0058] The interference signal pattern generation and parameter setting module is used to set the interference pattern signal parameters, then superimpose several interference pattern signals on echo A to obtain the echo of the superimposed interference pattern signal, and then input the echo of the superimposed interference pattern signal to the signal processing module.
[0059] The parameters of the interference pattern signal include: the interference-to-noise ratio of the suppression interference, the distance difference of the deception interference, the number of false targets, and the dragging speed of the dragging interference, etc.
[0060] The signal processing module is used to perform matched filtering on the echo after pulse accumulation, determine the detection threshold, measure the distance and angle of the target, and finally obtain the detection result, which is then input into the radar display module.
[0061] The radar display module is used to display the detection results;
[0062] Step Two, as follows Figure 1 As shown, based on the interference scenario established in step one, the dynamic detection process of radar over the target airspace is simulated, thereby generating interference intent and the corresponding interference behavior:
[0063] Step 2.1: Based on the interference scenario established in Step 1, simulate the dynamic detection process of the radar in the target airspace in the simulation system. Utilize the Markov property of the multi-function radar operating modes to generate a radar operating mode sequence, specifically:
[0064] The simulation demonstrates the dynamic detection process of radar over target airspace, displaying the real target's azimuth and detection results under radar interference in real time. The simulation time unit is one pulse position. During mission execution, the radar transmits several pulses per time unit. The simulation system then simulates the echoes received by each antenna element. After pulse accumulation, the echoes enter the signal processor, which performs matched filtering to determine the detection threshold, measures the target's range and angle, and finally updates the detection results to the radar display interface. The radar display interface is shown below. Figure 3 As shown.
[0065] The simulation system is designed to superimpose several jamming signal patterns onto the radar echo at any simulation time, and to set parameters for different jamming patterns, such as the interference-to-noise ratio for suppressing jamming, the range difference or number of false targets for deceiving jamming, and the dragging speed for dragging jamming, in order to simulate the constantly changing jamming signals of the jammer during the countermeasure process.
[0066] Step 22: The jammer formulates a targeted jamming intent based on the radar's operating mode, and then determines the jamming behavior based on the jamming intent using a jamming behavior decision model, specifically:
[0067] Step 221: Set the recognition probability of the jammer for each working mode of the radar to obtain the recognition result of the jammer for the working mode of the radar, and convert the recognition result into jamming intention according to the corresponding relationship;
[0068] The jammer's identification results of the radar's operating mode include: search, confirmation, tracking, and identification;
[0069] The jammer's identification results of the multi-functional air defense radar's operating modes and the corresponding jamming intentions are as follows: Search - Reduce detection, Confirm - Affect confirmation, Track - Escape tracking, Identify - Destroy identification;
[0070] Step 222: Set the conditional transition probabilities of the interference behavior decision-making model based on expert experience;
[0071] Steps 2 and 3: Obtain the target feature attributes after measurement or identification by the jammer. Perform probabilistic membership fuzzy processing on the target feature attributes to obtain a probability vector. Input the probability vector into the jamming behavior decision model, calculate the support confidence of each jamming behavior, and select the jamming behavior with the highest support confidence as the output, such as... Figure 4 As shown;
[0072] Among them, the interference behavior decision model is a Bayesian network model;
[0073] The characteristics and attributes measured or identified by the jammer include: the target's state information (distance, orientation) and jamming intent.
[0074] Step 3: Obtain the interference intent prediction model by associating the interference intent with the corresponding behavior. Specifically:
[0075] Step 3: 1. Divide the set of interference intentions and corresponding interference behaviors into a training set and a test set. Use the training set to train the LSTM network to obtain a trained LSTM network.
[0076] First, the interference intent obtained in step two is encoded using 1 to 4, and the interference behavior corresponding to the interference intent is encoded using 1 to 5 to obtain the encoded dataset;
[0077] Then, the time variable is added to the encoded dataset to obtain the sample dataset:
[0078] Three variables—the interference intent, interference behavior, and time information from previous moments—are selected as input variables for the Long Short-Term Memory (LSTM) network. The length T of the input variables is determined, representing the data from T rounds prior to the predicted intent. The interference intent from round T+1 is used as the label for the sample. The simulation data obtained in step two is then segmented to obtain the sample dataset, as shown below. Figure 9 As shown.
[0079] Then, to prevent gradient vanishing, the sample dataset was normalized. The normalized dataset was then randomly divided into a training set and a test set, with the test set comprising 25% of the total samples.
[0080] The input data dimension of the LSTM is equal to the input variable dimension T*3. The LSTM network includes 3 hidden layers and 1 fully connected layer, where the output data dimension of the hidden layer is 64.
[0081] The input dataset to the LSTM includes: interference intent, interference behavior, and timing information.
[0082] Step 3.2: Test the trained LSTM network using the test set. If the accuracy of the trained LSTM network is greater than or equal to the preset threshold, obtain the interference intention prediction model. If the accuracy of the trained LSTM network is less than the preset threshold, repeat Step 1 until the accuracy of the trained LSTM network is greater than or equal to the preset threshold.
[0083] Example:
[0084] The beneficial effects of the present invention are verified using the following embodiments:
[0085] First, a specific interference scenario is defined. The scenario includes an L-band radar capable of searching, confirming, tracking, and velocity identification. This multi-functional radar is equipped with a phased array using a 50×50 rectangular array, transmitting a linear frequency modulated signal with a PRF of 1500Hz, a detection range of 2–100km, a duty cycle of 0.1, a range resolution of 150m, a half-power beamwidth of 2°, an azimuth angle of -50° to 50°, and a dwell time of 0.01s.
[0086] The simulation system can superimpose jamming signals onto radar echoes at any simulation time to simulate the constantly changing jamming signals of the jammer during countermeasures. The scenario simulates five jamming behaviors, each corresponding to two jamming signal patterns. After determining the jamming behavior, the jammer randomly selects a jamming pattern from the corresponding jamming signals, as shown in Table 1.
[0087] Table 1
[0088]
[0089] For example, when a jammer performs a deception-plus-deception jamming action, it selects a jamming signal pattern that creates range decoys and velocity decoys on the radar. The RD spectrum of the echo and the corresponding radar interface are shown in Figures 5(a)-(b). The jammer approaches the air defense radar at a speed of 80 m / s at a distance of 50 km. Upon detecting the radar's velocity measurement, the jammer performs Doppler modulation and delay forwarding of the velocity measurement pulse, displaying four range decoys on the radar display interface and five velocity decoys of 93 m / s on the RD spectrum, thus achieving both range deception and speed decoy effects.
[0090] The jammer's initial position is (49636, 2000, 2112), initial velocity is 100 m / s, simulation sequence length is 100, rounds 1-50 are forward flight, rounds 50-85 are lateral flight, and rounds 85-100 are flight away from the radar, forming a trajectory as shown below. Figure 6 As shown.
[0091] Set the transition probability of the radar operating mode, such as Figure 7 The radar operating mode is generated for 100 rounds, and the jammer identifies the corresponding jamming intent. The conditional transition probabilities of the jamming decision model are set as shown in Table 2.
[0092] Table 2
[0093]
[0094] This provides an example of the interference decision model making an interference behavior decision at a certain moment. First, the probability vectors of the battlefield environment attribute state values detected by the jammer after probabilistic membership degree fuzzy processing are given, as shown in Table 3.
[0095] Table 3
[0096]
[0097] The confidence level of each attribute's membership degree supporting the formation of suppression behavior at this moment can be calculated using the formula:
[0098] Bel(forming suppression) = αp(int)λ∝(0.90×0.6+0.10×0.1+0×0.3)
[0099] ×(0.85×0.8+0.11×0+0.04×0.2)
[0100] ×(0.90×0.7+0.03×0.2+0.04×0.06+0.03×0.04)
[0101] ≈0.2420
[0102] Where α is the normalization factor and λ represents the information obtained from the features;
[0103] Similarly, it can be calculated that:
[0104] Bel (forming deception) ∝ 0.0191
[0105] Bel (forming drag) ∝ 0.0047
[0106] Bel (forming suppression + deception) ∝ 0.0098
[0107] Bel(forming deception + deception) ∝ 0.0018
[0108] It is evident that the decision-making confidence in initiating suppression behavior is highest at this moment, thus the jammer chooses to initiate suppression behavior. The jamming signal pattern is randomly selected between aiming jamming and blocking jamming, and the jamming signal pulse length is 6000. Similarly, throughout the simulation process, based on the set ranging and direction-finding accuracy of the jammer and the recognition accuracy of each radar operating mode, probability vectors of each attribute are generated and input into a Bayesian network to simulate 100 rounds of jamming behavior, the true value of jamming intent, and the time information of the corresponding rounds. Each round generates a jamming signal with a pulse length of 4, as shown in Figures 8(a)-(c), and the output is saved as a 24003*100 .mat file.
[0109] In the LSTM experiment predicting interference intent, the model's learning rate was set to 0.01, the mean squared error loss function was used, and the training algorithm was stochastic gradient descent with backpropagation. The batch size was set to 32, the variable length to 10 epochs, and the number of iterations to 10. The prediction accuracy and loss function during training are shown in Figures 10(a)-(b). Figures 10(a)-(b) show that the network's prediction accuracy generally increases when the number of iterations is less than 7, then levels off and stabilizes at around 85%. The loss function decreases rapidly at the beginning of training, gradually decreasing and stabilizing as the number of iterations increases. Considering both training time and network robustness, the network can be saved for testing after 10 iterations.
[0110] Experience shows that the more information available before prediction, the more accurate the prediction. Therefore, it can be predicted that the length of the input variable will have a significant impact on the prediction accuracy. Thus, prediction experiments were conducted with different input variable lengths. After training for 10 epochs, the network's prediction accuracy for interference intent was tested using a test set. Multiple experiments were conducted for each input variable length to calculate the average accuracy. The experimental results are shown in Table 4.
[0111] Table 4
[0112]
[0113] The experimental results above show that when the interference intent, interference behavior and time information of the previous time step are selected as the input variables of the Long Short-Term Memory Network to predict the interference intent of the next time step, the LSTM can achieve an accuracy of about 85% in predicting the interference intent of the simulation data, provided that the network parameters and variable lengths are set reasonably. This shows that the prediction effect is good.
Claims
1. A method for simulating and predicting dynamic interference behavior and intent, characterized in that... The specific process of the method is as follows: acquire historical moment variable data, and input the historical moment variable data into the interference intent prediction model to obtain the interference intent at the next moment; The historical moment variable data includes: time information, the interference intent of the historical moment, and the interference behavior of the historical moment; The interference intent prediction model is obtained through the following method: Step 1: Establish the jamming scenario and set the parameters of the radar, jammer, and target in the jamming scenario; Step 2: Based on the interference scenario established in Step 1, simulate the dynamic detection process of the radar in the target airspace to generate interference intent and the corresponding interference behavior, specifically: Step 2: Based on the interference scenario established in Step 1, simulate the dynamic detection process of the radar in the target airspace, and generate a radar operating mode sequence using the Markov property of the multi-function radar operating mode. Step 22: Generate jamming intent using radar operating mode sequences, input the jamming intent into the jamming behavior decision model, and generate the jamming behavior corresponding to the jamming intent, specifically as follows: Step 221: Set the recognition probability of the jammer for the radar operating mode, input the radar operating mode sequence into the jammer, obtain the jammer's recognition result of the radar operating mode, and convert the jammer's recognition result of the radar operating mode into the jamming intention according to the correspondence between the recognition result and the jamming intention. Step 222: Set the conditional transition probabilities for the interference behavior decision model; The interference behavior decision model is a Bayesian network model. Step 223: Use the interference machine to obtain the target feature attributes, perform probability membership fuzzy processing on the target feature attributes to obtain the probability vector, and then input the probability vector into the interference behavior decision model after setting the transition probability to obtain the support confidence of each interference behavior, and output the interference behavior with the highest support confidence. The target feature attributes include: the target's state information and interference intent; Step 3: Obtain a interference intent prediction model using the interference intent and the corresponding interference behavior, specifically as follows: Step 3: First, digitally encode the interference intent and interference behavior obtained in Step 2. Based on the relationship between the interference intent and interference behavior obtained in Step 2, obtain an encoded dataset consisting of the interference intent encoding and the interference behavior encoding corresponding to the interference intent. Step 32: Add the time variable to the encoded dataset to obtain the sample dataset. Normalize the sample dataset and divide the normalized sample dataset into training set and test set. Step 3: Train the LSTM network using the training set to obtain the trained LSTM network; Steps 3 and 4: Test the trained LSTM network using the test set. If the accuracy of the trained LSTM network is greater than or equal to the preset threshold, then use the currently trained LSTM network as the interference intention prediction model. If the accuracy of the trained LSTM network is less than the preset threshold, then repeat step 1 until the accuracy of the trained LSTM network is greater than or equal to the preset threshold.
2. The method for simulating and predicting dynamic interference behavior and intent according to claim 1, characterized in that: The target's status information includes: target distance and target orientation.
3. The method for simulating and predicting dynamic interference behavior and intent according to claim 2, characterized in that: The radar operating modes include: search, confirm, track, and identify.
4. The method for simulating and predicting dynamic interference behavior and intent according to claim 3, characterized in that: The correspondence between the identification results and the interference intent is as follows: Search - Reduce detection, Confirmation - Affect confirmation, Tracking - Get rid of tracking, Identification - Destroy identification.
5. The method for simulating and predicting dynamic interference behavior and intent according to claim 4, characterized in that: The specific encoding of the interference intent obtained in step two is as follows: the interference intents of reducing detection, influencing confirmation, escaping tracking, and disrupting identification are encoded as 1, 2, 3, and 4 respectively.
6. The method for simulating and predicting dynamic interference behavior and intent according to claim 5, characterized in that: The interference behavior obtained in step two is digitally encoded as follows: The interference behaviors of suppression, deception, dragging, suppression + deception, and deception + deception are sequentially coded as 1, 2, 3, 4, and 5.