Interception intention recognition strategy simulation system and method for unmanned aerial vehicle attack and defense game

Abstract

The application discloses an interception intention recognition strategy simulation system and method for unmanned aerial vehicle attack and defense game, which comprises a battle simulation data generation unit, a complex environment simulation unit and an interception intention recognition and decision unit. The battle simulation data generation unit is used for generating and storing flight performance parameters and flight data of enemy and friendly unmanned aerial vehicles. The complex environment simulation unit is used for simulating natural environment and electromagnetic interference conditions applied by the enemy when obtaining flight data of enemy unmanned aerial vehicles. The interception intention recognition and decision unit is used for predicting flight data of enemy unmanned aerial vehicles, recognizing interception intention of the enemy unmanned aerial vehicles based on the predicted flight data of the enemy unmanned aerial vehicles and current flight data of friendly unmanned aerial vehicles, and taking corresponding interception countermeasures. The application improves the efficiency of algorithm design and verification and shortens the test cycle through real-time simulation.

Description

Technical Field

[0001] This invention relates to a simulation system and method for attack and defense of unmanned aerial vehicle (UAV) swarms, and particularly to a simulation system and method for interception intent recognition strategy in UAV attack and defense game. Background Technology

[0002] Currently, with the rapid development of information-based warfare, target intent recognition, as a prerequisite and foundation for intelligent decision-making, is crucial to victory or defeat. Intent recognition refers to obtaining, processing, and analyzing information about the opponent's target status to deduce their intent. Target intent recognition can improve the accuracy of assessing potential risks, accurately identifying the degree and type of risk posed by the opponent's target, providing more reaction time for appropriate actions, and ultimately grasping the overall development trend. In multi-UAV games, detectable opponent information mainly includes the location, distance, flight speed, and trajectory of the opponent's UAV swarm. These characteristics can reflect the opponent's UAV action intent to some extent. However, due to the complex environment and rapidly changing situation of both sides, how to quickly and accurately identify the opponent's intent has become a research challenge.

[0003] Existing research methods mainly include template matching, Bayesian networks, neural networks, and deep learning networks. Template matching and Bayesian networks rely heavily on prior knowledge, making it difficult to accurately identify the other party's intent in complex situations. Neural networks have the ability to learn the latent features of data, overcoming the dependence on prior knowledge and improving the accuracy of intent recognition. They have a significant technical advantage in intent recognition problems under complex offensive and defensive situations. However, research on intent recognition based on neural networks still needs to solve three problems: (1) how to obtain learning samples; (2) how to construct an intent space and generate knowledge labels; and (3) how to construct an effective neural network model. Summary of the Invention

[0004] This invention provides a simulation system and method for identifying interception intent in UAV attack and defense game to solve the technical problems existing in the prior art.

[0005] The technical solution adopted by the present invention to solve the technical problems existing in the prior art is: a simulation system for interception intent recognition strategy in UAV attack and defense game, including a battle simulation data generation unit, a complex environment simulation unit, and an interception intent recognition and decision-making unit;

[0006] The combat simulation data generation unit is used to generate and store the flight data of enemy drones and our drone swarm, as well as the coordinate data of our base; by initializing the speed and coordinates of enemy drones and our drone swarm, and simulating the flight of both sides' drones, the simulated flight data of both sides' drones is obtained.

[0007] The complex environment simulation unit is used to simulate the natural environment and electromagnetic interference imposed by the enemy when acquiring enemy drone flight data; it adds noise to the enemy drone flight data generated by the combat simulation data generation unit.

[0008] The interception intent identification and decision-making unit is used to predict the flight data of enemy drones, identify the offensive and defensive intentions of enemy drones based on the predicted flight data, and take corresponding interception countermeasures based on the current flight data of our drone swarm.

[0009] Furthermore, the complex environment simulation unit includes a Gaussian white noise generator.

[0010] Furthermore, the interception intent recognition and decision-making unit includes: a fusion prediction module jointly constructed by a long short-term memory neural network and a backpropagation neural network, a K-nearest neighbor module, and a decision tree module;

[0011] The fusion prediction module is used to predict the flight data of enemy drones. It inputs the simulated flight data of enemy drones into a long short-term memory neural network, and uses the multiple time-series prediction results output by the long short-term memory neural network as input to the backpropagation neural network. The backpropagation neural network outputs the predicted flight data of the enemy drones. The flight data includes flight heading, flight coordinates, and flight speed. The predicted trajectory is generated from the predicted flight data.

[0012] The K-nearest neighbor module is used to process the predicted flight data output by the fusion prediction module. It calculates the Euclidean distance between the predicted flight data of the enemy UAV and the current flight data of each friendly UAV. It classifies the predicted trajectories into different priorities based on the probability magnitude and outputs the predicted trajectory of the enemy UAV with the highest priority as the final predicted trajectory.

[0013] The decision tree module is used for decision-making in the interception process. It includes a decision tree, in which decision classification nodes are set, and corresponding decisions are made based on the predicted values ​​of enemy drone flight data and the current flight data of our drone swarm.

[0014] Furthermore, the fusion prediction module includes a Long Short-Term Memory (LSTM) neural network module and a backpropagation neural network module connected in sequence; the LSM neural network module includes n LSM neural network units connected in sequence; let the current time step be i, and the previous m time steps be im, where: m = n-1, i ≥ m+1; let the flight data of the enemy UAV at the current time step be X. i Let Y be the enemy drone flight data predicted by the Long Short-Term Memory neural network module at the current time step. i Let the coordinates of our base at the current time step be Z. iLet the flight data of the enemy drone from the previous time step to the previous m time steps be: X i-1 X i-2 , ..., X i-(m-2) X i-(m-1) X i-m Let the enemy drone flight data predicted by the Long Short-Term Memory Neural Network module from the previous time step to the previous m time steps be as follows: Y i-1 Y i-2 , ..., Y i-(m-2) Y i-(m-1) Y i-m The inputs to the 1st to nth Long Short-Term Memory (LSTM) neural network units are respectively: X i-m X i-(m-1) X i-(m-2) , ..., X i-2 X i-1 X i The outputs of the 2nd to nth Long Short-Term Memory (LSTM) neural network units correspond to: Y i-(m-1) Y i-(m-2) , ..., Y i-2 Y i-1 Y i The memory cell states and hidden states of the first to n-1th Long Short-Term Memory (LSTM) neural network units are output to the next LSM neural network unit; the backpropagation neural network module has the following inputs: Z i X i-1 X i-2 , ..., X i-(m-2) Y i-1 Y i-2 , ..., Y i-(m-1) Its output includes predicted flight data of enemy drones.

[0015] This invention also provides a simulation method for interception intent recognition strategy in UAV attack and defense game, which sets up a battle simulation data generation unit, a complex environment simulation unit, and an interception intent recognition and decision-making unit;

[0016] The combat simulation data generation unit is used to generate and store flight data of enemy drones and our drone swarm, as well as our base coordinate data; by initializing the speed and coordinates of enemy drones and our drone swarm, and simulating the flight of both sides' drones, the simulated flight data of both sides' drones is obtained.

[0017] The complex environment simulation unit is used to simulate the natural environment and electromagnetic interference imposed by the enemy when acquiring enemy drone flight data; it adds noise to the enemy drone flight data generated by the combat simulation data generation unit.

[0018] The interception intent identification and decision-making unit is used to predict the flight data of enemy drones, identify the offensive and defensive intentions of enemy drones based on the predicted flight data, and take corresponding interception countermeasures based on the current flight data of our drone swarm.

[0019] Furthermore, the interception intent recognition and decision-making unit is set up as follows: a fusion prediction module, a K-nearest neighbor module, and a decision tree module jointly constructed by a long short-term memory neural network and a backpropagation neural network;

[0020] The fusion prediction module is used to predict the flight data of enemy drones. It inputs the simulated flight data of enemy drones into a long short-term memory neural network, and uses the multiple time-series prediction results output by the long short-term memory neural network as input to the backpropagation neural network. The backpropagation neural network outputs the predicted flight data of the enemy drones. The flight data includes flight heading, flight coordinates, and flight speed. The predicted trajectory is generated from the predicted flight data.

[0021] The K-nearest neighbor module is used to process the predicted flight data output by the fusion prediction module. It calculates the Euclidean distance between the predicted flight data of the enemy UAV and the current flight data of each friendly UAV. It classifies the predicted trajectories into different priorities based on the probability magnitude and outputs the predicted trajectory of the enemy UAV with the highest priority as the final predicted trajectory.

[0022] The decision tree module is used for decision-making in the interception process. It includes a decision tree, in which decision classification nodes are set, and corresponding decisions are made based on the predicted values ​​of enemy drone flight data and the current flight data of our drone swarm.

[0023] Furthermore, the fusion prediction module is configured with a Long Short-Term Memory (LSTM) neural network module and a backpropagation neural network module connected in sequence; the LSM neural network module is configured with n LSM neural network units connected in sequence; let the current time step be i, and the previous m time steps be im, where: m = n-1, i ≥ m+1; let the flight data of the enemy UAV at the current time step be X. i Let Y be the enemy drone flight data predicted by the Long Short-Term Memory neural network module at the current time step. i Let the coordinates of our base at the current time step be Z. i Let the flight data of the enemy drone from the previous time step to the previous m time steps be: X i-1 X i-2 , ..., X i-(m-2) X i-(m-1) X i-m Let the enemy drone flight data predicted by the Long Short-Term Memory Neural Network module from the previous time step to the previous m time steps be as follows: Y i-1 Y i-2, ..., Y i-(m-2) Y i-(m-1) Y i-m ; X i-m X i-(m-1) X i-(m-2) , ..., X i-2 X i-1 X i The inputs are sequentially fed into the first to nth Long Short-Term Memory (LSTM) neural network units; the outputs are sequentially fed into the second to nth LSTM neural network units: Y i-(m-1) Y i-(m-2) , ..., Y i-2 Y i-1 Y i The memory cell states and hidden states of the first to (n-1)th long short-term memory neural network units are all output to the next long short-term memory neural network unit, and Z is... i X i-1 X i-2 , ..., X i-(m-2) Y i-1 Y i-2 , ..., Y i-(m-1) The data is input to the backpropagation neural network module, which then outputs the predicted flight data of the enemy drone.

[0024] Furthermore, the method includes the following steps:

[0025] Step 1: The combat simulation data generation unit generates and stores the flight data of the enemy drones and our drone swarm, as well as the coordinate data of our base; the complex environment simulation unit adds noise to the generated enemy drone flight data.

[0026] Step 2: Select flight heading, flight coordinates, and flight speed as feature labels, compile the data generated by the combat simulation data generation unit into a prediction dataset, and establish a decision tree;

[0027] Step 3: Input the current and historical flight data of each enemy drone into a long short-term memory neural network for multi-step time series prediction, and input the multi-step time series prediction results into a backpropagation neural network to predict the next flight data of the enemy drone.

[0028] Step 4: Input the predicted flight data of the enemy drone and the flight data of each drone in the current drone swarm into the K-nearest neighbor module to obtain the predicted flight data of the enemy drone with the highest probability.

[0029] Step 5: Input the predicted flight data of the enemy drone and the flight data of each drone in the current drone swarm into the decision tree to make a decision on whether the current iteration step is successful and whether to continue interception.

[0030] Step 6: Repeat steps 3-5 until the iteration termination condition is met.

[0031] Furthermore, when the complex environment simulation unit adds noise to the generated enemy UAV flight data, it adds Gaussian white noise with the same signal-to-noise ratio to all data based on different assignments of different characteristic signals to simulate the noise of the acquired signals under actual conditions; then, it changes different signal-to-noise ratios to test the difference in prediction accuracy under noise interference of different intensities.

[0032] The advantages and positive effects of this invention are as follows: This invention employs a combat simulation data generation unit to generate and store the flight performance parameters and flight data of both enemy and friendly UAVs; by constructing a UAV dynamics simulation model and initializing the speeds and coordinates of both enemy and friendly UAV swarms, and simulating the flight of both sides' UAVs, simulated flight data of both sides' UAVs is obtained, and this simulated flight data is used to simulate real-world combat scenarios. A complex environment simulation unit is used to simulate the natural environment and electromagnetic interference applied by the enemy when acquiring enemy UAV flight data; it adds noise to the enemy UAV flight data generated by the combat simulation data generation unit. An interception intent recognition and decision-making unit predicts the enemy UAV flight data and determines corresponding interception countermeasures based on the flight data of both enemy and friendly UAVs. Real-time simulation improves the efficiency of algorithm design and verification and shortens the experimental cycle.

[0033] This invention constructs a drone attack and defense simulation scenario through a pre-defined scenario; it provides decision information and corresponding decisions during the attack and defense interception process through decision tree construction, and makes decisions on the interception process. The drone confrontation process is carried out by the fusion prediction module jointly constructed by long short-term memory neural network and backpropagation neural network, K-nearest neighbor module and decision tree module. For the complex prediction scenario of drone swarm, the joint prediction module predicts the flight trajectory of enemy drones individually, the K-nearest neighbor module processes the predicted flight data of enemy drones, and the decision tree module makes the corresponding decisions. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of the structure of a simulation system for an interception intent recognition strategy in a game of attack and defense between unmanned aerial vehicles (UAVs) according to the present invention.

[0035] Figure 2 This is a schematic diagram of applying signals with different signal-to-noise ratios to a complex environment simulation unit.

[0036] Figure 3 This is a schematic diagram of the decision tree structure in the interception intent recognition and decision-making unit.

[0037] Figure 4 This is a schematic diagram simulating an attack and defense scenario between enemy and friendly drones.

[0038] Figure 5 This is a flowchart of the K-nearest neighbor module in the interception intent recognition and decision-making unit.

[0039] Figure 6 This is a schematic diagram of the fusion module structure jointly constructed by the long short-term memory neural network and the backpropagation neural network in the interception intent recognition and decision-making unit.

[0040] Figure 4 In the text: r represents the effective interception radius.

[0041] Figure 6 Chinese: X i-(m-2) X i-(m-1) X i-m X i These correspond sequentially to the flight data of the enemy drone at the previous m-2 time steps, the previous m-1 time steps, the previous m time steps, and the current time step; Y i-(m-2) Y i-(m-1) Y i The data presented are, in order, enemy drone flight data predicted by the Long Short-Term Memory neural network module for the first m-2 time steps, the first m-1 time steps, and the current time step; Z i This contains the coordinates of our base at the current time step; LSTM1, LSTM2, ..., LSTM n These correspond sequentially to the 1st to nth Long Short-Term Memory (LSTM) neural network units; c1, c2, c3, ... c n-1 , corresponding to the memory cell states of the 1st to n-1th Long Short-Term Memory (LSTM) neural network units; h1, h2, h3, ... h n-1 This corresponds to the hidden state of the memory cell in the 1st to n-1th long short-term memory neural network unit. Detailed Implementation

[0042] To further understand the invention's content, features, and effects, the following embodiments are provided, along with detailed descriptions in conjunction with the accompanying drawings:

[0043] Please see Figures 1 to 6 A simulation system for interception intent recognition strategy in UAV attack and defense game includes a battle simulation data generation unit, a complex environment simulation unit, and an interception intent recognition and decision-making unit.

[0044] The combat simulation data generation unit is used to generate and store flight data of enemy drones and our drone swarm, as well as our base coordinate data; by initializing the speed and coordinates of enemy drones and our drone swarm, and simulating the flight of both sides' drones, the simulated flight data of both sides' drones is obtained.

[0045] The complex environment simulation unit is used to simulate the natural environment and electromagnetic interference imposed by the enemy when acquiring enemy drone flight data; it adds noise to the enemy drone flight data generated by the combat simulation data generation unit.

[0046] The interception intent identification and decision-making unit is used to predict the flight data of enemy drones, identify the offensive and defensive intentions of enemy drones based on the predicted flight data, and take corresponding interception countermeasures based on the current flight data of our drone swarm.

[0047] The combat simulation data generation unit can simulate air combat using UAV dynamics models. By constructing UAV dynamics models and initializing the speeds and coordinates of enemy and friendly UAV swarms, various complex real-world UAV swarm attack and defense scenarios can be simulated through programming and parameter settings. Initial speed values ​​for both sides are set with lower limits, and an interception is considered successful when the distance between the two UAVs is less than the effective interception range. During the simulated flight of the dynamics simulation model, the flight coordinates and headings of the enemy UAVs, as well as the flight speeds and headings of each UAV in the friendly UAV swarm, are continuously changed to simulate the complex and ever-changing situations of real air combat. After running the UAV dynamics model, the simulated flight speeds, headings, and flight frequencies of both sides' UAVs are recorded. The data sampling frequency is 100Hz; this sufficiently high sampling frequency and multi-dimensional combination provide a diverse and sufficient dataset foundation for subsequent decision-making.

[0048] Simplifying the UAV dynamics model to a point mass model, our UAV dynamics model is as follows:

[0049]

[0050] Where t represents any moment after the start of the interception activity; M represents the number of drones in our drone swarm; p j The derivative of (t); p j (t) represents the position of the j-th drone in our drone swarm at time t; Indicates v j The derivative of (t); v j (t) represents the flight speed of the j-th drone in our drone swarm at time t; a j (t) represents the acceleration control quantity of the j-th UAV in our UAV swarm at time t; μ j This represents the drag constant of the j-th drone in our drone swarm;

[0051] The dynamics model of the UAV is simplified to a point mass model. The dynamics model of the enemy UAV is as follows:

[0052]

[0053] Where t represents any time after the start of the interception activity; p e The derivative of (t); p e (t) represents the position of the enemy drone at time t; v e (t) represents the flight speed of the enemy drone at time t; a e (t) represents the acceleration control quantity of the enemy drone at time t; μ e This represents the drag constant of the enemy drone.

[0054] In a real-world environment, the flight data of each drone in our swarm can be detected by its own sensors, and this data is unaffected by interference from enemy drones or other environmental signals. In contrast, the flight data of enemy drones requires radar detection, making them more susceptible to interference from enemy drones and other environmental signals.

[0055] Preferably, the combat simulation data generation unit may include a flight database; the flight database unit is used to store flight data of enemy drones and friendly drone swarms.

[0056] Preferably, the complex environment simulation unit may include a Gaussian white noise generator. The complex environment simulation unit can simulate the natural environment when obtaining signals in actual combat situations, as well as noise interference such as enemy electromagnetic signals, and add Gaussian white noise to the flight data generated by the UAV dynamics model simulation flight.

[0057] Preferably, the interception intent recognition and decision-making unit may include: a fusion prediction module, a K-nearest neighbor module, and a decision tree module; the fusion prediction module is jointly constructed by a long short-term memory neural network and a backpropagation neural network.

[0058] The fusion prediction module, jointly constructed by a long short-term memory neural network and a backpropagation neural network, can be used to predict the flight data of enemy drones. The flight data of enemy drones simulated by the combat simulation data generation unit is input into the long short-term memory neural network. Multiple time-series prediction results output by the long short-term memory neural network are used as input to the backpropagation neural network. The backpropagation neural network outputs the predicted flight data of the enemy drone. The flight data includes flight heading, flight coordinates, and flight speed. The predicted trajectory is generated from the predicted flight data.

[0059] Preferably, the fusion prediction module structure can be as follows: Figure 6As shown, the fusion prediction module may include a Long Short-Term Memory (LSTM) neural network module and a backpropagation neural network module connected in sequence; the LSTM neural network module may include n LSTM neural network units connected in sequence; the current time step can be set as i, and the previous m time steps as im, where: m = n-1, i ≥ m+1; the flight data of the enemy UAV at the current time step can be set as X. i Let Y be the enemy drone flight data predicted by the Long Short-Term Memory neural network module at the current time step. i The coordinates of our base at the current time step can be set to Z. i The flight data of the enemy drone from the previous time step to the previous m time steps can be set as follows: X i-1 X i-2 , ..., X i-(m-2) X i-(m-1) X i-m Let the enemy drone flight data predicted by the Long Short-Term Memory Neural Network module from the previous 1st to the previous mth time steps be: Y i-1 Y i-2 , ..., Y i-(m-2) Y i-(m-1) Y i-m The inputs to the 1st to nth Long Short-Term Memory (LSTM) neural network units can be sequentially represented as: X i-m X i-(m-1) X i-(m-2) , ..., X i-2 X i-1 X i The outputs of the 2nd to nth Long Short-Term Memory (LSTM) neural network units can be sequentially represented as: Y i-(m-1) Y i-(m-2) , ..., Y i-2 Y i-1 Y i The memory cell states and hidden states of the first to (n-1)th long short-term memory neural network units can all be output to the next long short-term memory neural network unit; the backpropagation neural network module can include the following inputs: Z i X i-1 X i-2 , ..., X i-(m-2) Y i-1 Y i-2 , ..., Y i-(m-1) Its output can include predicted flight data of enemy drones.

[0060] The simulated flight data of the enemy drone at multiple time steps can be input into the Long Short-Term Memory Neural Network (LSTM) module. The output of the LSTM module and the coordinate data of our base are then input into the Backpropagation Neural Network (BPNN) module, which outputs the predicted flight data of the enemy drone.

[0061] The Long Short-Term Memory (LSTM) neural network unit consists of an input layer, an LSTM layer, a fully connected layer, a softmax layer, and a classification layer. The LSTM layer has 100 hidden units, and each unit's cell structure mainly comprises a forget gate, an input gate, and an output gate. The LSTM neural network module is used to process and predict important events at intervals in a time series, predicting the next flight path of an enemy drone based on existing current and historical flight data.

[0062] The backpropagation neural network module is trained using the error backpropagation algorithm. Input data is processed step-by-step through the hidden layers until the output is reached. Based on the output and the expected error, the parameters and thresholds of each neuron are adjusted backwards to make the output increasingly closer to the expected value. The training steps include network initialization, forward propagation, backward propagation, weight modification, and network generation. The backpropagation neural network module has 30 hidden layers, uses a Bayesian algorithm with a learning rate of 0.02, and the activation function of the hidden layers is the hyperbolic tangent function.

[0063]

[0064] In the formula, x∈(-∞,+∞), tanhx∈(-1,+1).

[0065] The output layer activation function uses the sigmoid function:

[0066]

[0067] In the formula, x∈(-∞,+∞) and S(x)∈(0,1).

[0068] The backpropagation neural network module can be used to correct and optimize the calculation process of the long short-term memory neural network module in predicting the trajectory of the intercepted target. It returns the errors between the enemy UAV's flight data at multiple time steps and the predicted flight data in the simulation environment, and feeds these errors into the BP network to update the parameters of the weights and thresholds. In this method, the long short-term memory neural network is combined with the backpropagation neural network. The long short-term memory neural network performs preliminary flight prediction based on the multiple possibilities of the enemy UAV's flight data, and then passes the output to the backpropagation neural network, improving the prediction accuracy and stability to adapt to the current simulation scenario.

[0069] The K-nearest neighbor module can be used to process the predicted flight data output by the fusion prediction module. It calculates the Euclidean distance between the predicted flight data of the enemy UAV and the current flight data of each of our UAVs. It classifies the predicted trajectories according to different priorities based on the probability magnitude and outputs the predicted trajectory of the enemy UAV with the highest priority as the final predicted trajectory.

[0070] In the K-nearest neighbor module, a feature space labeled with flight coordinates is established. The Euclidean distance between the predicted flight coordinates of the enemy drone and the current flight coordinates of each drone in the friendly drone swarm in the feature space is calculated sequentially. The k (0 < k ≤ 20) predicted flight data values ​​of the enemy drone with the shortest Euclidean distance are selected, and the enemy drone that appears most frequently is selected.

[0071] The K-Nearest Neighbors module employs the K-Nearest Neighbors algorithm, also known as the KNN algorithm. This is a commonly used learning algorithm in data mining and machine learning. Before performing the KNN algorithm, the data is first normalized to avoid the influence of units on the distance calculation. The distance between the data to be classified and every sample in the training set is calculated. The k nearest samples to the data to be classified are found. The classification of these k samples is observed. The category with the most frequent occurrence is taken as the category of the data to be classified. The distance calculation method in the K-Nearest Neighbors module is the Euclidean distance method, with each neighbor having only one neighbor and equal distance weights. The distance in N-dimensional space is d(x,y).

[0072]

[0073] In the formula,

[0074] x i (i = 1...n) represents the i-th dimension coordinate of sample x in N-dimensional space;

[0075] y i (i=1...n) represents the i-th dimension coordinate of sample y in N-dimensional space.

[0076] The K-Nearest Neighbors module is used to calculate the Euclidean distance between each predicted value of the enemy drone's flight trajectory and the current coordinates of our drone. It further classifies the predicted trajectories into different priorities based on probability, and the K-Nearest Neighbors module outputs the predicted flight data with the highest priority.

[0077] The probability calculation formula is as follows:

[0078]

[0079] Among them, P j denoted by , k represents the probability corresponding to the j-th type of enemy drone flight data prediction value; k represents the number of selected enemy drone flight data prediction values; h represents the number of times the j-th type of enemy drone flight data prediction value appears among the selected prediction values.

[0080] The decision tree module is used for decision-making in the interception process. It can include a decision tree, in which decision classification nodes can be set. It can make corresponding decisions based on the predicted values ​​of enemy drone flight data and the current flight data of our drone swarm.

[0081] The system can select enemy / friendly base headings, enemy / friendly drone headings, enemy / friendly drone effective interception range, and enemy / friendly drone speeds as feature labels. Based on predicted flight data of enemy drones and predicted Euclidean distances between enemy and friendly drones, it can make corresponding decisions, including resource allocation schemes for the drone swarm with the highest interception probability and the next interception maneuver scheme for friendly drones. When using the correlation between enemy / friendly base headings or the headings of valuable enemy / friendly targets as a feature, a binary decision is made; if correlated, a next decision is made; otherwise, the current action is maintained. When using the correlation between enemy / friendly drone headings as a feature, a binary decision is made; if correlated, a next decision is made; otherwise, the current action is maintained. When using the effective interception range of enemy / friendly drones or the effectiveness of enemy / friendly interception ranges as a feature, a binary decision is made; if within the effective interception range of friendly drones, a next decision is made; otherwise, formation flight and combined maneuvers are performed. When using the speed of enemy / friendly drones as a feature, a binary decision is made; if a match is found, interception is executed; otherwise, a new decision is made.

[0082] The decision classification termination condition is: when an enemy drone approaches our base, within the enemy drone's maximum endurance t... max If the enemy drone is within the effective interception range of at least one of our drones, and our base remains outside the enemy drone's attack range from the start of the interception operation, then the interception mission is successful. The formula is as follows:

[0083] |p n (t end )-p e (t end )|≤r cap ,1≤n≤M;

[0084] |p base -p e (t)|≥r att ,0<t≤t end ;

[0085] 0 < t end ≤t max ;

[0086] In the formula, p n This indicates the position of the nth drone in our drone swarm that can effectively intercept enemy drones; p e Indicates the location of the enemy drone; p base Indicates the location of our base; r capIndicates the effective interception range of our drones; r att M represents the enemy drone's attack range; M represents the number of drones in our drone swarm; t represents any moment after the start of the interception operation; t end Indicates the end time of the interception activity; t max Indicates the maximum endurance of the enemy drone; p n (t end ) represents the position of the nth drone that can be effectively intercepted in our drone swarm at the end of the interception operation; p e (t) represents the position of the enemy drone at any point after the start of the interception operation.

[0087] The present invention also provides an embodiment of a simulation method for interception intent recognition strategy in UAV attack and defense game, which includes a battle simulation data generation unit, a complex environment simulation unit, and an interception intent recognition and decision-making unit.

[0088] The combat simulation data generation unit is used to generate and store flight data of enemy drones and friendly drone swarms, as well as the coordinate data of friendly bases; by initializing the speed and coordinates of enemy drones and friendly drone swarms, and simulating the flight of drones of both sides, the simulated flight data of drones of both sides is obtained.

[0089] The complex environment simulation unit is used to simulate the natural environment and electromagnetic interference imposed by the enemy when acquiring enemy drone flight data; it adds noise to the enemy drone flight data generated by the combat simulation data generation unit.

[0090] The interception intent identification and decision-making unit is used to predict the flight data of enemy drones, identify the offensive and defensive intentions of enemy drones based on the predicted flight data, and take corresponding interception countermeasures based on the current flight data of our drone swarm.

[0091] Preferably, the combat simulation data generation unit can be equipped with a flight database; the flight database unit can be used to store the performance parameters of UAVs and flight data of both friendly and enemy UAVs. By collecting and analyzing performance data of various types of UAVs in the prior art, preset labels are set for parameters related to maneuverability, such as maximum speed, maximum acceleration, and maximum flight distance, for each UAV type, and these are stored in the flight database. The flight database provides the correspondence between UAV performance parameters and UAV types. The relationship between UAV characteristic parameters and their threat level to friendly targets is analyzed, and relevant mathematical models are constructed to represent this relationship, including enemy and friendly maneuver angles, enemy and friendly flight altitudes, enemy and friendly flight speeds, and enemy and friendly distances. This completes the mapping from target characteristics to threat level assessment and develops a real-time query function, providing a reference for further evaluation of indicators such as interception intent identification.

[0092] Preferably, the interception intent recognition and decision-making unit may be configured with: a fusion prediction module jointly constructed by a long short-term memory neural network and a backpropagation neural network, a K-nearest neighbor module, and a decision tree module.

[0093] The fusion prediction module can be used to predict the flight data of enemy drones. It can input the simulated flight data of enemy drones into a long short-term memory neural network, and use the multiple time-series prediction results output by the long short-term memory neural network as input to the backpropagation neural network. The backpropagation neural network outputs the predicted flight data of the enemy drone. The flight data includes flight heading, flight coordinates, and flight speed. The predicted trajectory is generated from the predicted flight data.

[0094] The K-nearest neighbor module can be used to process the predicted flight data output by the fusion prediction module. It can calculate the Euclidean distance between the predicted flight data of the enemy UAV and the current flight data of each of our UAVs. It classifies the predicted trajectories according to different priorities based on the probability magnitude and outputs the predicted trajectory of the enemy UAV with the highest priority as the final predicted trajectory.

[0095] The decision tree module is used for decision-making in the interception process. It can include a decision tree, in which decision classification nodes can be set. It can make corresponding decisions based on the predicted values ​​of enemy drone flight data and the current flight data of our drone swarm.

[0096] Preferably, the fusion prediction module can be configured with a Long Short-Term Memory (LSTM) neural network module and a backpropagation neural network module connected in sequence; the LSM neural network module can be configured with n LSM neural network units connected in sequence; the current time step can be set as i, and the previous m time steps can be set as im, where: m = n-1, i ≥ m+1; the flight data of the enemy UAV at the current time step can be set as X. i Let Y be the enemy drone flight data predicted by the Long Short-Term Memory neural network module at the current time step. i The coordinates of our base at the current time step can be set to Z. i The flight data of the enemy drone from the previous time step to the previous m time steps can be set as follows: X i-1 X i-2 , ..., X i-(m-2) X i-(m-1) X i-m Let the enemy drone flight data predicted by the Long Short-Term Memory Neural Network module from the previous 1st to the previous mth time steps be: Y i-1 Y i-2 , ..., Y i-(m-2) Y i-(m-1) Y i-m X can be i-m X i-(m-1) X i-(m-2) , ..., Xi-2 X i-1 X i The inputs are sequentially fed into the first to nth Long Short-Term Memory (LSTM) neural network units; the outputs are sequentially fed into the second to nth LSTM neural network units: Y i-(m-1) Y i-(m-2) , ..., Y i-2 Y i-1 Y i The memory cell states and hidden states of the first to (n-1)th long short-term memory (LSTM) neural network units can be output to the next LSM neural network unit, thus enabling Z... i X i-1 X i-2 , ..., X i-(m-2) Y i-1 Y i-2 , ..., Y i-(m-1) The data is input into the backpropagation neural network module, which can then output the predicted flight data of the enemy drone.

[0097] Preferably, the method for setting up a decision tree may include: classifying the flight data of enemy and friendly UAVs, and using the label data with greater information gain as the decision classification node.

[0098] Preferably, the method may include the following steps:

[0099] Step 1: The combat simulation data generation unit generates and stores the flight data of the enemy drones and our drone swarm, as well as the coordinate data of our base; the complex environment simulation unit adds noise to the generated enemy drone flight data.

[0100] Step 2: Select flight heading, flight coordinates, and flight speed as feature labels, compile the data generated by the combat simulation data generation unit into a prediction dataset, and establish a decision tree.

[0101] Step 3: Input the current and historical flight data of each enemy drone into a long short-term memory neural network for multi-step time series prediction, and input the multi-step time series prediction results into a backpropagation neural network to predict the next flight data of the enemy drone.

[0102] Step 4: Input the predicted flight data of the enemy drone and the flight data of each drone in the current friendly drone swarm into the K-nearest neighbor module to obtain the predicted flight data of the enemy drone with the highest probability.

[0103] Step 5: Input the predicted flight data of the enemy drone and the flight data of each drone in the current drone swarm into the decision tree to make a decision on whether the current iteration step is successful and whether to continue interception.

[0104] Step 6: Repeat steps 3-5 until the iteration termination condition is met.

[0105] Preferably, when the complex environment simulation unit adds noise to the generated enemy UAV flight data, it adds Gaussian white noise with the same signal-to-noise ratio to all data based on different assignments of different characteristic signals to simulate the noise of the acquired signals under actual conditions; then, it changes different signal-to-noise ratios to test the difference in prediction accuracy under noise interference of different intensities.

[0106] The aforementioned combat simulation data generation unit, complex environment simulation unit, interception intent recognition and decision-making unit, flight database, long short-term memory neural network, backpropagation neural network module, K-nearest neighbor module, decision tree module, decision tree and other functional units, modules and components can all be functional units, modules and components in the prior art, or can be constructed using functional units, modules and components in the prior art and conventional technical means.

[0107] The embodiments described above are only used to illustrate the technical ideas and features of the present invention. Their purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly. The patent scope of the present invention should not be limited by these embodiments. That is, all equivalent changes or modifications made in accordance with the spirit disclosed in the present invention still fall within the patent scope of the present invention.

Claims

1. A simulation system for interception intent recognition strategy in UAV attack-defense game, characterized in that, It includes a combat simulation data generation unit, a complex environment simulation unit, and an interception intent recognition and decision-making unit; The combat simulation data generation unit is used to generate and store the flight data of enemy drones and our drone swarm, as well as the coordinate data of our base; by initializing the speed and coordinates of enemy drones and our drone swarm, and simulating the flight of both sides' drones, the simulated flight data of both sides' drones is obtained. The complex environment simulation unit is used to simulate the natural environment and electromagnetic interference imposed by the enemy when acquiring flight data of enemy drones. It adds noise to the enemy drone flight data generated by the combat simulation data generation unit; The interception intent identification and decision-making unit is used to predict the flight data of enemy drones, identify the offensive and defensive intentions of enemy drones based on the predicted flight data, and take corresponding interception countermeasures based on the current flight data of our drone swarm. The interception intent recognition and decision-making unit includes: a fusion prediction module jointly constructed by a long short-term memory neural network and a backpropagation neural network, a K-nearest neighbor module, and a decision tree module; The fusion prediction module is used to predict the flight data of enemy drones. It inputs the simulated flight data of enemy drones into a long short-term memory neural network, and uses the multiple time-series prediction results output by the long short-term memory neural network as input to the backpropagation neural network. The backpropagation neural network outputs the predicted flight data of the enemy drones. The flight data includes flight heading, flight coordinates, and flight speed. The predicted trajectory is generated from the predicted flight data. The K-nearest neighbor module is used to process the predicted flight data output by the fusion prediction module. It calculates the Euclidean distance between the predicted flight data of the enemy UAV and the current flight data of each friendly UAV. It classifies the predicted trajectories into different priorities based on the probability magnitude and outputs the predicted trajectory of the enemy UAV with the highest priority as the final predicted trajectory. The decision tree module is used for decision-making in the interception process. It includes a decision tree, in which decision classification nodes are set, and corresponding decisions are made based on the predicted values ​​of enemy drone flight data and the current flight data of our drone swarm.

2. The simulation system for interception intent recognition strategy in UAV attack and defense game as described in claim 1, characterized in that, The complex environment simulation unit includes a Gaussian white noise generator.

3. The simulation system for interception intent recognition strategy in UAV attack and defense game as described in claim 1, characterized in that, The fusion prediction module includes a Long Short-Term Memory (LSTM) neural network module and a backpropagation neural network module connected in sequence; the LSM neural network module includes n LSM neural network units connected in sequence; let the current time step be i, and the previous m time steps be im, where m = n - 1, i ≥ m + 1; let the flight data of the enemy UAV at the current time step be X. i Let Y be the enemy drone flight data predicted by the Long Short-Term Memory neural network module at the current time step. i Let the coordinates of our base at the current time step be Z. i Let the flight data of the enemy drone from the previous time step to the previous m time steps be: X i-1 X i-2 , ..., X i-(m-2) X i-(m-1) X i-m Let the enemy drone flight data predicted by the Long Short-Term Memory Neural Network module from the previous time step to the previous m time steps be as follows: Y i-1 Y i-2 , ..., Y i-(m-2) Y i-(m-1) Y i-m The inputs to the 1st to nth Long Short-Term Memory (LSTM) neural network units are respectively: X i-m X i-(m-1) X i-(m-2) , ..., X i-2 X i-1 X i The outputs of the 2nd to nth Long Short-Term Memory (LSTM) neural network units correspond to: Y i-(m-1) Y i-(m-2) , ..., Y i-2 Y i-1 Y i The memory cell states and hidden states of the first to n-1th Long Short-Term Memory (LSTM) neural network units are output to the next LSM neural network unit; the backpropagation neural network module has the following inputs: Z i X i-1 X i-2 , ..., X i-(m-2) Y i-1 Y i-2 , ..., Y i-(m-1) Its output includes predicted flight data of enemy drones.

4. A simulation method for interception intent recognition strategy in UAV attack-defense game, characterized in that, This method includes a combat simulation data generation unit, a complex environment simulation unit, and an interception intent recognition and decision-making unit. The combat simulation data generation unit is used to generate and store flight data of enemy drones and our drone swarm, as well as our base coordinate data; by initializing the speed and coordinates of enemy drones and our drone swarm, and simulating the flight of both sides' drones, the simulated flight data of both sides' drones is obtained. The complex environment simulation unit is used to simulate the natural environment and electromagnetic interference imposed by the enemy when acquiring flight data of enemy drones. It adds noise to the enemy drone flight data generated by the combat simulation data generation unit; The interception intent identification and decision-making unit is used to predict the flight data of enemy drones, identify the offensive and defensive intentions of enemy drones based on the predicted flight data, and take corresponding interception countermeasures based on the current flight data of our drone swarm. Interception intent recognition and decision-making unit settings: a fusion prediction module, a K-nearest neighbor module, and a decision tree module jointly constructed by a long short-term memory neural network and a backpropagation neural network; The fusion prediction module is used to predict the flight data of enemy drones. It inputs the simulated flight data of enemy drones into a long short-term memory neural network, and uses the multiple time-series prediction results output by the long short-term memory neural network as input to the backpropagation neural network. The backpropagation neural network outputs the predicted flight data of the enemy drones. The flight data includes flight heading, flight coordinates, and flight speed. The predicted trajectory is generated from the predicted flight data. The K-nearest neighbor module is used to process the predicted flight data output by the fusion prediction module. It calculates the Euclidean distance between the predicted flight data of the enemy UAV and the current flight data of each friendly UAV. It classifies the predicted trajectories into different priorities based on the probability magnitude and outputs the predicted trajectory of the enemy UAV with the highest priority as the final predicted trajectory. The decision tree module is used for decision-making in the interception process. It includes a decision tree, in which decision classification nodes are set, and corresponding decisions are made based on the predicted values ​​of enemy drone flight data and the current flight data of our drone swarm.

5. The simulation method for interception intent recognition strategy in UAV attack-defense game according to claim 4, characterized in that, The fusion prediction module is configured with a Long Short-Term Memory (LSTM) neural network module and a backpropagation neural network module connected in sequence. The LSTM neural network module consists of n LSM units connected in sequence. Let the current time step be i, and the previous m time steps be im, where m = n - 1, i ≥ m + 1. Let the flight data of the enemy drone at the current time step be X. i Let Y be the enemy drone flight data predicted by the Long Short-Term Memory neural network module at the current time step. i Let the coordinates of our base at the current time step be Z. i Let the flight data of the enemy drone from the previous time step to the previous m time steps be: X i-1 X i-2 , ..., X i-(m-2) X i-(m-1) X i-m Let the enemy drone flight data predicted by the Long Short-Term Memory Neural Network module from the previous time step to the previous m time steps be as follows: Y i-1 Y i-2 , ..., Y i-(m-2) Y i-(m-1) Y i-m ; X i-m X i-(m-1) X i-(m-2) , ..., X i-2 X i-1 X i The inputs are sequentially fed into the first to nth Long Short-Term Memory (LSTM) neural network units; the outputs are sequentially fed into the second to nth LSTM neural network units: Y i-(m-1) Y i-(m-2) , ..., Y i-2 Y i-1 Y i The memory cell states and hidden states of the first to (n-1)th long short-term memory neural network units are all output to the next long short-term memory neural network unit, and Z is... i X i-1 X i-2 , ..., X i-(m-2) Y i-1 Y i-2 , ..., Y i-(m-1) The data is input to the backpropagation neural network module, which then outputs the predicted flight data of the enemy drone.

6. The simulation method for interception intent recognition strategy in UAV attack-defense game according to claim 4, characterized in that, The method includes the following steps: Step 1: The combat simulation data generation unit generates and stores the flight data of the enemy drones and our drone swarm, as well as the coordinate data of our base; the complex environment simulation unit adds noise to the generated enemy drone flight data. Step 2: Select flight heading, flight coordinates, and flight speed as feature labels, compile the data generated by the combat simulation data generation unit into a prediction dataset, and establish a decision tree; Step 3: Input the current and historical flight data of each enemy drone into a long short-term memory neural network for multi-step time series prediction, and input the multi-step time series prediction results into a backpropagation neural network to predict the next flight data of the enemy drone. Step 4: Input the predicted flight data of the enemy drone and the flight data of each drone in the current drone swarm into the K-nearest neighbor module to obtain the predicted flight data of the enemy drone with the highest probability. Step 5: Input the predicted flight data of the enemy drone and the flight data of each drone in the current drone swarm into the decision tree to make a decision on whether the current iteration step is successful and whether to continue interception. Step 6: Repeat steps 3-5 until the iteration termination condition is met.

7. The simulation method for interception intent recognition strategy in UAV attack-defense game according to claim 4, characterized in that, When the complex environment simulation unit adds noise to the generated enemy UAV flight data, it adds Gaussian white noise with the same signal-to-noise ratio to all data based on different assignments of different characteristic signals to simulate the noise of the acquired signals under actual conditions; then it changes different signal-to-noise ratios to test the difference in prediction accuracy under noise interference of different intensities.

Images