Unmanned aerial vehicle cluster game confrontation decision method based on MASAC

Abstract

The invention particularly relates to an unmanned aerial vehicle cluster collaborative confrontation decision method based on MASAC, and the method comprises the steps: constructing a motion model of an unmanned aerial vehicle, and simulating and predicting the motion track of the unmanned aerial vehicle; setting constraint conditions of unmanned aerial vehicle attacks; designing an unmanned aerial vehicle confrontation task based on the motion model and the constraint condition, modeling the confrontation task as a partially observable Markov decision process, and constructing an unmanned aerial vehicle strategy; based on an improved MASAC algorithm, a centralized training-distributed execution framework is adopted, and an unmanned aerial vehicle strategy is optimized through maximum entropy reinforcement learning. According to the method, challenges such as complex reward function design and limited decision timeliness in a complex scene are solved, and the decision performance is improved, so that multiple agents can better complete tasks.

Description

Technical Field

[0001] The present invention relates to the field of UAV cluster control, and in particular to a MASAC-based UAV cluster collaborative confrontation decision-making method. Background Art

[0002] The rapid development of drone technology has attracted widespread attention, particularly its potential applications in drone swarm warfare. As an emerging combat platform, drones, with multiple capabilities including reconnaissance, surveillance, aerial communications, and ground attack, are playing an increasingly important role in modern warfare.

[0003] However, despite the many impressive performances of individual drones, their limitations cannot be ignored. They have limited coverage, relatively limited mission capabilities, and their information collection and processing capabilities are insufficient when dealing with complex scenarios involving multiple targets and multiple missions. In particular, single drones often struggle to adapt flexibly to rapidly changing battlefield environments.

[0004] In contrast, drone swarms, composed of multiple drones, demonstrate tremendous potential. Through the collaborative efforts of multiple nodes, drone swarms enable information sharing, adaptive task scheduling, and highly coordinated combat capabilities. This enables drone swarms to better adapt to the demands of modern warfare, demonstrating unique advantages in tasks such as electronic warfare, complex terrain reconnaissance, and simultaneous multi-target attacks. These advantages make drone swarms a key force in future military confrontations, better suited to the variability and complexity of actual warfare.

[0005] Despite the obvious advantages of drone swarms, their practical application faces significant challenges. In particular, the complexity of swarms, often presented by incomplete environmental information and continuous action spaces, poses significant challenges to control and decision-making. Traditional control methods often struggle to handle such high-dimensional and dynamically changing application scenarios, hindering the full potential of drone swarms. Faced with these challenges, researchers have gradually turned their attention to advanced technologies such as deep reinforcement learning (DRL).

[0006] Deep reinforcement learning, with its remarkable autonomous learning capabilities and adaptability, has achieved remarkable results in solving complex, high-dimensional problems. It not only enables effective autonomous decision-making within drone swarms, but also maintains swarm coordination and stability in the face of enemy interference and complex environmental changes. Furthermore, reinforcement learning's ability to facilitate both collaboration and confrontation among multi-agent systems holds great promise for its application in drone swarm confrontation scenarios.

[0007] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present invention, and therefore may include information that does not constitute prior art known to ordinary technicians in this field. Summary of the Invention

[0008] This paper provides a MASAC-based drone swarm collaborative confrontation decision-making method to address the intelligent decision-making challenges in drone swarm confrontation environments, particularly for asymmetric confrontations between a small number of drones and a large number of drones. Traditional multi-agent reinforcement learning algorithms suffer from slow convergence and poor strategy stability when dealing with such complex environments with high dimensions, nonlinearity, incomplete information, and continuous action spaces.

[0009] Other features and advantages of the present invention will become apparent from the following detailed description, or may be learned in part by practice of the present invention.

[0010] According to a first aspect of the present invention, a drone swarm collaborative confrontation decision-making method based on MASAC is provided, the method comprising:

[0011] Build a UAV motion model, simulate and predict the UAV's trajectory; set the constraints for UAV attacks;

[0012] Design UAV confrontation tasks based on motion models and constraints, model the confrontation tasks as partially observable Markov decision processes, and construct UAV strategies;

[0013] Based on the improved MASAC algorithm, a centralized training-distributed execution framework is adopted to optimize the UAV strategy through maximum entropy reinforcement learning; the improved MASAC algorithm references an adaptive temperature parameter update mechanism.

[0014] In some exemplary embodiments, constructing a motion model of the drone includes:

[0015] An update formula for determining the heading angle based on the heading angular velocity;

[0016] Determine the real-time motion speed of the drone based on the updated heading angle and driving force;

[0017] Determine the real-time position of the drone based on its real-time motion speed.

[0018] In some exemplary embodiments, setting constraints for drone attacks includes:

[0019] The distance between the attacker and the target is less than the attacker's flight radius;

[0020] The angle of attack is less than half of the attacker's effective attack sector.

[0021] In some exemplary embodiments, the drone confrontation task is specifically:

[0022] In each round of confrontation, the winner is determined by evaluating the survival status of both sides' drones;

[0023] If one side is completely eliminated, the other side wins;

[0024] If both teams have drones surviving, the winner will be determined by the average score of the drones, and the team with the higher score will win;

[0025] The scoring method is: a drone gets one point for each step it survives.

[0026] In some exemplary embodiments, the adversarial task is modeled as a partially observable Markov decision process, specifically:

[0027] S: state space, S=[S1,S2,…,S N ], there are N drones in the game scene, and the state of drone i is S i , i=1,2,…,N;

[0028] O i : observable state space of UAV i;

[0029] A: Joint action space, A = A1 × A2 × … × A N , this space contains all actions of all drones, where A i represents the action space of drone i;

[0030] P: is the state transition function; P(s,a,s') represents the state transition probability from one state s∈S to the next state s'∈S through a joint action a∈A,

[0031] R i : The reward obtained by drone I, which means that drone i is in state s according to strategy π i Execute action a i Rewards received;

[0032] γ: discount factor, γ∈[0,1], used to weigh the relative importance of long-term rewards and current rewards;

[0033] The goal of UAV i is to learn a i The policy π mapped to the action space distribution i : This strategy maximizes the expected discounted reward of drone i:

[0034]

[0035] Among them, r it is the reward of drone i, a Nt is the action space of drone i.

[0036] In some exemplary embodiments, the improved MASAC algorithm includes:

[0037] Using the maximum entropy reinforcement learning framework, the objective function is designed as:

[0038]

[0039] Among them, α is the adaptive temperature parameter, H(π i (·∣o i ) is the policy entropy;

[0040] Optimize the strategy by minimizing the KL divergence:

[0041]

[0042] Among them, Z i (o; μ i,j ) is the partition function.

[0043] In some exemplary embodiments, the updating mechanism of the adaptive temperature parameter α includes:

[0044] Considering entropy as a constraint, the objective function is:

[0045]

[0046] in, is the target entropy value;

[0047] Update the temperature parameter by gradient descent:

[0048]

[0049] Among them, λ α is the learning rate.

[0050] According to a second aspect of the present invention, a storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the MASAC-based drone cluster collaborative confrontation decision-making method described in the first aspect is implemented.

[0051] According to a third aspect of the present invention, a computer program product is provided, on which a computer program is stored. When the computer program is executed by a processor, the MASAC-based drone cluster collaborative confrontation decision-making method described in the first aspect is implemented.

[0052] According to a fourth aspect of the present invention, there is provided an electronic device, comprising:

[0053] processor; and

[0054] a memory for storing executable instructions of the processor;

[0055] Wherein, the processor is configured to implement the drone cluster collaborative confrontation decision-making method based on MASAC described in the first aspect above by executing the executable instructions.

[0056] The embodiments of the present invention provide a MASAC-based drone swarm collaborative adversarial decision-making method. This method constructs a drone swarm adversarial simulation environment and proposes a drone motion model. It also designs drone survival state update rules and adversarial tasks. Furthermore, a distributed drone swarm collaborative control algorithm based on multi-agent reinforcement learning is developed. The MASAC algorithm is used for policy iterative optimization, and an effective sparse reward function is designed to guide drones in adversarial games. Simulation results demonstrate that the MASAC method outperforms existing multi-agent deep reinforcement learning techniques in terms of convergence speed and stability. This method addresses challenges such as complex reward function design and limited decision-making timeliness in complex scenarios, improving decision-making performance and enabling multi-agents to better complete tasks.

[0057] Compared with the existing technology, the specific advantages are as follows:

[0058] 1. Adaptive temperature parameter update mechanism: This invention introduces an adaptive temperature parameter update mechanism, which enables the algorithm to dynamically adjust the balance between exploration and exploitation according to changes in the adversarial environment, improving the convergence speed and stability of the strategy. It is particularly suitable for scenarios where a small number of drones are fighting against a large number of drones.

[0059] 2. Maximum Entropy Reinforcement Learning Framework: By integrating the maximum entropy reinforcement learning framework into the Multi-Agent Soft Actor-Critic (MASAC) algorithm, this invention not only improves cumulative rewards but also encourages strategy diversity and flexibility, avoids strategies from falling into local optimality, and enables drones to have stronger adaptability in asymmetric confrontations.

[0060] 3. Centralized Training-Distributed Execution Architecture: This invention utilizes a centralized training-distributed execution (CTDE) architecture to enable efficient coordinated UAV operations in incomplete information environments. By sharing global information, a small number of UAVs can form a highly coordinated combat strategy, enhancing their ability to counter numerical disadvantages.

[0061] 4. Efficient processing of continuous action space: To address the continuous action space problem in drone swarm confrontation, the MASAC algorithm of the present invention optimizes the processing through a flexible strategy, enabling the algorithm to have higher decision-making efficiency and accuracy in a dynamic environment.

[0062] 5. Excellent asymmetric confrontation capability: This invention focuses on improving the confrontation capability of a small number of drones when facing a large number of drones, enabling them to achieve tactical advantages through strategy optimization and information sharing under numerical disadvantages, and provides an innovative solution for the application of drone swarm confrontation technology.

[0063] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] The accompanying drawings are incorporated into and constitute a part of this specification, illustrate embodiments consistent with the present invention, and together with the description, serve to explain the principles of the present invention. Obviously, the drawings described below are only some embodiments of the present invention, and it is clear that those skilled in the art can derive other drawings based on these drawings without inventive effort.

[0065] Figure 1 Schematic diagram of the process of the MASAC-based UAV swarm collaborative confrontation decision-making method according to an exemplary embodiment of the present invention;

[0066] Figure 2 A diagram of the attack and defense situation of two drones according to an exemplary embodiment of the present invention;

[0067] Figure 3 A training framework for reinforcement learning according to an exemplary embodiment of the present invention;

[0068] Figure 4 This is a graph showing the average reward curve of the Red Team UAV trained in an environment based on the improved MASAC algorithm according to an exemplary embodiment of the present invention;

[0069] Figure 5 Schematic diagram of the initial positions of the red and blue drones during training according to an exemplary embodiment of the present invention;

[0070] Figure 6 A schematic diagram of the locations where the red and blue drones are destroyed during training according to an exemplary embodiment of the present invention; DETAILED DESCRIPTION

[0071] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be embodied in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete and will fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0072] In addition, the accompanying drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. Identical reference numerals in the figures denote identical or similar parts, and thus repetitive descriptions thereof will be omitted. Some of the blocks shown in the accompanying drawings are functional entities that do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0073] To address the shortcomings and deficiencies of existing technologies, this example implementation provides a MASAC-based decision-making method for cooperative drone swarm confrontation. This method establishes a drone swarm confrontation environment and constructs drone motion and confrontation strategy models. The method defines the state space and action space of each drone when facing multiple enemy drones in a few-to-many coordinated drone confrontation, and designs corresponding confrontation mission planning, modeling the problem as a partially observable Markov decision process.

[0074] refer to Figure 1 As shown, the following steps may be specifically included:

[0075] Step S11: constructing a UAV motion model, simulating and predicting the UAV's motion trajectory; and setting UAV attack constraint conditions.

[0076] Step S12: designing a UAV confrontation task based on the motion model and constraints, modeling the confrontation task as a partially observable Markov decision process, and constructing a UAV strategy;

[0077] In step S13, based on the improved MASAC algorithm, a centralized training-distributed execution framework is adopted to optimize the UAV strategy through maximum entropy reinforcement learning; the improved MASAC algorithm references an adaptive temperature parameter update mechanism.

[0078] Below, each step in this exemplary implementation will be described in more detail with reference to the accompanying drawings and embodiments.

[0079] In step S11, a motion model of the UAV is constructed to simulate and predict the motion trajectory of the UAV; and constraint conditions for the UAV attack are set.

[0080] For example, a simplified two-dimensional kinematic model is used for the motion control of a UAV. This model describes the attitude change and position update process of the UAV in the horizontal plane. The model mainly includes the following parts:

[0081] 1) Roll angle and heading angle:

[0082] Roll angle Indicates the rotation angle of the drone around its longitudinal axis. Its rate of change is determined by the roll angular velocity The following relationship is used in the model for update:

[0083] φ=φ+r φ ·dt,-30°≤φ≤30°

[0084] Heading angle Indicates the flight direction of the drone. Its rate of change is determined by the heading angular velocity Control and update according to the following formula:

[0085]

[0086] 2) Movement speed calculation:

[0087] The velocity component of the drone in the horizontal plane is determined by the heading angle and the driving force F. Calculate its velocity component v on the x-axis and y-axis respectively x and v y :

[0088]

[0089] Where m is the mass of the UAV.

[0090] 3) Location Update:

[0091] Based on the above velocity components, the model further calculates the position change of the drone in the horizontal plane. The positions x and y are updated using the following formulas:

[0092] x=x+v x ·dt

[0093] y=y+v y ·dt

[0094] Through this kinematic model, the motion trajectory of the UAV can be simulated and predicted in real time, thereby providing the necessary parameter support for the control system to achieve more accurate navigation and path planning.

[0095] The offensive and defensive situation of the two drones is as follows Figure 2 As shown, the attacker is i and the target is j. At time t, the attack and defense situation of any pair of drones can be described by Figure 2 Indicates that, is the attack angle of attacker i.

[0096]

[0097] Among them, ||v (·) ||2 is the 2-norm of the velocity vector, d (·)(·) is the Euclidean distance between attacker i and target j.

[0098] The attack capability of a drone is represented by its effective attack area. Taking the red drone i as an example, the effective attack area is defined as the area with itself as the center, the speed direction as the center line, and the radius d ac , the angle is θ ac sector area.

[0099] According to the geometric parameters, attacker i can destroy target j when the following conditions are met: (1) The distance d between the attacker and the target ij Less than the attacker's (2) Target j is in the attack zone of attacker i ( Figure 2 The above constraints can be expressed as:

[0100]

[0101] In step S12, a UAV confrontation task is designed based on the motion model and constraints, the confrontation task is modeled as a partially observable Markov decision process, and a UAV strategy is constructed.

[0102] For example, the goal of the confrontation mission is to indiscriminately destroy a greater number of enemy drones. After each round of confrontation, the winner is determined by evaluating the survival status of both sides' drones. If one side is completely destroyed, the other side wins. If both sides have drones surviving, the decision is made according to the following method: each drone scores one point for each surviving step. At the end of the round, the side with the higher average drone score wins. If the scores are the same, the game is a draw. This transforms the random game into a Markov decision problem.

[0103] In a drone swarm confrontation environment, all drones interact with the environment at the same time and learn their own strategies during the interaction. Since each drone can only observe part of the battlefield information, that is, the information is incomplete, we model this confrontation game scenario as a partially observable Markov game, which can be expressed as a multivariate group <S,O1,O2,…,O N ,A,P,R1,R2,…,R N ,γ>, the specific introduction is as follows:

[0104] S: state space, S=[S1,S2,…,S N ], there are N drones in the game scene, and the state of drone i is S i , i=1,2,…,N.

[0105] O i : The observable state space of UAV i includes part of the information obtained from the environment.

[0106] A: Joint action space, A = A1 × A2 × … × A N , this space contains all actions of all drones, where A i represents the action space of drone i.

[0107] P: is the state transition function. P(s,a,s') represents the state transition probability from one state s∈S to the next state s'∈S through the joint action a∈A,

[0108] R i : The reward obtained by drone I, which means that drone i is in state s according to strategy π i Execute action a i Rewards received.

[0109] γ: Discount factor, γ∈[0,1], used to weigh the relative importance of long-term rewards and current rewards.

[0110] The goal of UAV i is to learn a i The policy π mapped to the action space distribution i : This strategy maximizes the expected discounted reward of drone i:

[0111]

[0112] The above state space S and action space A are both continuous, so the drone swarm confrontation game belongs to a partially observable Markov game in an infinite domain.

[0113] In step S13, based on the improved MASAC algorithm, a centralized training-distributed execution framework is adopted to optimize the UAV strategy through maximum entropy reinforcement learning; the improved MASAC algorithm references an adaptive temperature parameter update mechanism.

[0114] Exemplarily, according to the Markov game, a reinforcement learning framework based on the improved MASAC is established, such as Figure 3 shown.

[0115] Specifically, the drone is trained; the MASAC algorithm adopts a centralized training and distributed execution architecture, and establishes two independent neural networks for training: Actor network π θ , and Critic Network Q μ Each drone has the same observation space and action space; during training, drones share the parameters of the Actor network and the Critic network, and the Critic network is only used to assist the training of the intelligent agent; after the training, each drone uses the local observation O i, the Actor network makes independent decisions without centralized observation input. The output of the Actor network is set to a multidimensional Gaussian distribution with mean and standard deviation, and the action a is obtained by sampling method i , to ensure the continuity of the UAV’s motion space;

[0116] Compared to MADDPG and MAPPO, MASAC introduces the idea of maximum entropy, which has both theoretical and practical advantages: first, it encourages the agent to explore a wider space while abandoning obviously unpromising paths; second, the strategy can capture multiple near-optimal behaviors.

[0117] The goal of MASAC is to find a set of random policy vectors to encourage drones to explore in as wide a range as possible and avoid being trapped in local optimality, so as to obtain the global maximum cumulative reward and policy entropy. Its objective function is expressed as

[0118]

[0119] Furthermore, the desired policy can be expressed as

[0120]

[0121] Among them, the temperature parameter α is used to control the relative importance of reward and entropy. When α→0, J(π i ) is restored to the standard MARL objective function form; H(π i (·∣o i )) is the entropy term,

[0122] The goal of maximum entropy reinforcement learning is to find an optimal policy that not only maximizes the expected reward but also maximizes the entropy of the policy distribution, which can be expressed as follows

[0123]

[0124] Among them, Q solf An entropy term is added to the traditional Q function.

[0125] The key idea is that maximizing the expected return of a strategy entropy can lead to a more robust and flexible strategy that is better able to cope with uncertainty and adapt to a changing environment.

[0126] In order to improve the agent's ability to explore and find the global optimum and to capture multiple near-optimal behaviors, a fitting distribution can be used to stabilize training and improve performance, that is, to minimize the distribution from the strategy to Q solf The KL (Kullback-Leibler) divergence of the distribution. So the optimal strategy can be expressed as

[0127]

[0128] here, is the partition function, which is used to achieve normalization.

[0129] The goal of the critic network is to approximate the value function of the joint state-action pair. In this algorithm, in order to prevent overestimation from bootstrapping, the Q network here is designed as a pseudo-twin network with the same structure but no shared parameters. The loss function of the pseudo-twin Q network is

[0130]

[0131]

[0132] in, is the target network parameter in the pseudo-twin Q network, which is determined by the critic network parameter μ i,j Sliding average obtained by flexible updating.

[0133] Set a parameterized policy π for each drone i (a i ∣o i θ i ), whose policy network is the actor network, can be learned by minimizing the objective function through gradient descent. The objective function is expressed as the expectation of the KL divergence:

[0134]

[0135] The desired strategy is obtained by applying gradient descent to the objective function.

[0136] Although the introduction of the maximum entropy concept can encourage drones to explore in the widest possible range and avoid being trapped in local optimality, so as to obtain the globally maximized cumulative reward and policy entropy, it is not simple to control the ratio between the reward term and the entropy term, that is, the selection of the temperature parameter α is not simple.

[0137] In the improved MASAC algorithm we proposed, entropy is considered as a constraint rather than part of the learning objective itself. In order to automatically adjust the temperature parameter α, we formulate the optimization problem of maximizing the expected return as an optimization problem with entropy constraints:

[0138]

[0139] here is the target entropy value, which specifies the entropy we want the policy to achieve.

[0140] Convert it into a dual optimization problem

[0141]

[0142] Considering the Lagrangian function of the dual optimization problem, the objective function can be written as:

[0143]

[0144] The temperature parameter α is updated using the gradient descent method to achieve automatic adjustment.

[0145]

[0146] Among them, λ α is the learning rate.

[0147] Other embodiments of the present invention will readily occur to those skilled in the art after considering the specification and practicing the invention herein. This application is intended to cover any variations, uses, or adaptations of the present invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered as exemplary only, with the true scope and spirit of the invention being indicated by the claims.

[0148] It should be understood that the present invention is not limited to the exact construction described above and shown in the drawings and that various modifications and variations can be made without departing from the scope thereof, which is limited only by the appended claims.

Claims

1. A drone swarm collaborative confrontation decision-making method based on MASAC, characterized by: The method comprises: Build a UAV motion model, simulate and predict the UAV's trajectory; set the constraints for UAV attacks; Design UAV confrontation tasks based on motion models and constraints, model the confrontation tasks as partially observable Markov decision processes, and construct UAV strategies; Based on the improved MASAC algorithm, a centralized training-distributed execution framework is adopted to optimize the UAV strategy through maximum entropy reinforcement learning; the improved MASAC algorithm references an adaptive temperature parameter update mechanism.

2. The method according to claim 1, characterized in that The step of constructing the UAV motion model includes: An update formula for determining the heading angle based on the heading angular velocity; Determine the real-time motion speed of the drone based on the updated heading angle and driving force; Determine the real-time position of the drone based on its real-time motion speed.

3. The method according to claim 2, characterized in that The constraints for setting drone attacks include: The distance between the attacker and the target is less than the attacker's flight radius; The angle of attack is less than half of the attacker's effective attack sector.

4. The method according to claim 1, wherein The drone confrontation mission is specifically as follows: In each round of confrontation, the winner is determined by evaluating the survival status of both sides' drones; If one side is completely eliminated, the other side wins; If both teams have drones surviving, the winner will be determined by the average score of the drones, and the team with the higher score will win; The scoring method is: a drone gets one point for each step it survives.

5. The method according to claim 1, wherein The adversarial task is modeled as a partially observable Markov decision process, specifically: S: state space, S=[S1,S2,…,S N ], there are N drones in the game scene, and the state of drone i is S i , i=1,2,…,N; O i : observable state space of UAV i; A: Joint action space, A = A1 × A2 × … × A N , this space contains all actions of all drones, where A i represents the action space of drone i; P: is the state transition function; P(s,a,s') represents the state transition probability from one state s∈S to the next state s'∈S through a joint action a∈A, R i : The reward obtained by drone I, which means that drone i is in state s according to strategy π i Execute action a i Rewards received; γ: discount factor, γ∈[0,1], used to weigh the relative importance of long-term rewards and current rewards; The goal of UAV i is to learn a i The policy π mapped to the action space distribution i : This strategy maximizes the expected discounted reward of drone i: Among them, r it is the reward of drone i, a Nt is the action space of drone i.

6. The method according to claim 1, characterized in that The improved MASAC algorithm includes: Using the maximum entropy reinforcement learning framework, the objective function is designed as: Among them, α is the adaptive temperature parameter, H(π i (·∣o i ) is the policy entropy; Optimize the strategy by minimizing the KL divergence: Among them, Z i (o; μ i,j ) is the partition function.

7. The method according to claim 6, characterized in that The updating mechanism of the adaptive temperature parameter α includes: Considering entropy as a constraint, the objective function is: in, is the target entropy value; Update the temperature parameter by gradient descent: Among them, λ α is the learning rate.

8. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the MASAC-based drone cluster collaborative confrontation decision-making method according to any one of claims 1 to 7 is implemented.

9. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the MASAC-based drone cluster collaborative confrontation decision-making method according to any one of claims 1 to 7 is implemented.

10. An electronic device, characterized in that: include: processor; as well as a memory for storing executable instructions of the processor; Wherein, the processor is configured to execute the MASAC-based drone cluster collaborative confrontation decision-making method according to any one of claims 1 to 7 by executing the executable instructions.

Images