Design And Simulation Of Bionic Intelligent Algorithm For Group Confrontation

The swarm intelligence algorithm originated from the study of swarm behavior in animals,such as ants,fish schools,bees,and other social animals or insects.This type of algorithm imitates a class of collective behaviors exhibited by biological communities and represents them with certain rules and strategies,thus it can be uesd to solve optimization problems in real world.Since the swarm intelligence algorithm involves the mutual cooperation of individuals within a population to achieve higher collective intelligence,it can be seen as a multi-agent system.However,traditional swarm intelligence algorithms lack the concept of adversarial thinking and cannot be applied to actual environments involving group adversarial scenarios,such as controlling a swarm of unmanned aerial vehicles(UAVs)on a battlefield to efficiently strike targets.Therefore,to address practical needs,it is necessary to research bionic intelligence algorithms oriented to group adversarial environments.Moreover,existing studies have shown promising results in applying reinforcement learning to multi-agent optimal control.If reinforcement learning can be combined with bionic algorithms in group adversarial scenarios,it has the potential to further enhance the performance of the algorithms.Based on the aforementioned background,this thesis conducts research and simulation on bionic intelligence algorithms for group adversarial scenarios.Firstly,improvements are made to the grey wolf optimizer and artificial fish swarm algorithm,and rule-based improved grey wolf optimizer and artificial fish swarm algorithm are proposed for group adversarial tasks.In the improved wolf algorithm,five group adversarial actions are proposed: search action,surround action,aggregation attack,dispersion attack,and allocation target attack.In the improved fish swarm algorithm,six group antagonism actions are proposed: random walk,crowding,tail chasing,foraging,teaching and differential communication.Based on the definition of the above adversarial actions,the grey wolf optimizer and artificial fish swarm algorithm processes are redesigned.In order to further improve the success rate of algorithms in complex adversarial tasks,this thesis combines two types of algorithms with reinforcement learning and proposes improved grey wolf algorithm and improved fish swarm algorithm based on reinforcement learning.Reinforcement learning is used instead of strategy for action selection in group adversarial algorithms.And a pre-training method is designed to ensure the effectiveness of reinforcement learning training.On the basis of completing the pre-training,reinforcement learning training continues to improve the possibility of selecting optimal actions.In order to test the effectiveness of the proposed algorithm,this thesis constructs an attack and defense simulation scenario that can be used for group confrontation in Starcraft II,deploys the designed algorithm to the attacker and tests the effectiveness of each algorithm in the simulation scenario.The experimental results show that the two improved rule-based adversarial group bionic algorithms can be used in adversarial task scenarios,and the efficiency of the algorithms can be further improved after they are combined with reinforcement learning.