Research On Algorithm Security For Intelligent Obstacle Avoidance Of UAV Swarm

With the development of unmanned aerial vehicle(UAV)technology,the application of UAVs in the military and civilian fields has become more and more in-depth,and it is developing in the trend of clustering to meet the needs of complex tasks such as monitoring,search and rescue in dangerous environments.As a core algorithm for UAV swarms,obstacle avoidance algorithm has always been the focus of research and development in industry and academia.Many scholars have carried out optimization and improvement research on the basis of swarm intelligent Flocking algorithm.However,the current algorithm only considers homogeneous UAV swarms,ignoring the heterogeneity of swarms and the security of the algorithm.Therefore,for the scenario where the obstacle avoidance algorithm is applied to heterogeneous swarms,it is of great significance to deeply study the parameter sensitivity and security problems of the obstacle avoidance algorithm,which can reduce the safety loss and improve the task application scope of the UAV swarm.This paper mainly conducts research on the algorithm security of UAV swarm intelligent obstacle avoidance from the following two aspects:First,parameter sensitivity analysis of UAV swarm obstacle avoidance algorithm.The stability of the currently widely used Locking obstacle avoidance algorithm is analyzed.In view of its poor stability in some cases,the Hybrid Flocking algorithm,an optimized cluster obstacle avoidance algorithm based on the Flocking algorithm,is proposed.The Hybrid Flocking algorithm considers both the Group Collision Avoidance(GCA)strategy and the Individual Collision Avoidance(ICA)strategy,and designs a switching algorithm for the two strategies.Through the reinforcement learning Actor-Critic(AC)algorithm,the parameter sensitivity of the obstacle avoidance algorithm is deeply analyzed,and the automatic adjustment mechanism of the weight parameters of the obstacle avoidance algorithm is realized to improve the stability of the UAV swarm obstacle avoidance algorithm.Second,the arrival state data poisoning attack model based on Rushing attack is proposed,which aims to use the vulnerability of the mainstream AODV protocol in the UAV swarm communication network to cause the time delay of the heterogeneous UAV swarm flight mission,which is safe for the algorithm.Reinforce the foundation.The research on attack protection of existing UAV swarm obstacle avoidance algorithms is still in its infancy,especially the lack of effective protection against UAV swarm task delay caused by data poisoning attacks.Aiming at the above problems,this paper introduces the Null Space Based(NSB)method to realize the security reinforcement of the obstacle avoidance algorithm,and proposes an obstacle avoidance constraint model for UAV swarms,which enhances the robustness of the model.For the above proposed method,this thesis designs an optimized cluster obstacle avoidance algorithm Hybrid Flocking algorithm based on the swarm intelligence algorithm Flocking algorithm in the heterogeneous UAV cluster scenario.In simulation experiments,compared with the Flocking algorithm,the Hybrid Flocking algorithm makes the swarm a safe distance between UAVs has less fluctuation and better retention,and the probability of collision and mission failure is lower.Through the reinforcement learning AC algorithm,the weight parameters of the obstacle avoidance algorithm are automatically adjusted to improve the stability of the obstacle avoidance algorithm for heterogeneous UAV clusters.In the experiment,the task time was reduced by up to9.68%,and the average task time was reduced by about 8%.This paper designs a data poisoning attack against the cluster obstacle avoidance algorithm in the heterogeneous UAV swarm scenario.The experiment analyzes the average task delay caused by the attack under different cluster sizes,different UAV speeds,and different arrival judgment thresholds.The cluster task delay caused by the attack can reach up to 47.84%.This paper introduces the NSB method to reinforce the obstacle avoidance algorithm of heterogeneous clusters.After using the reinforcement algorithm,the delay caused by the attack is reduced by up to 41.39%,which verifies the effectiveness of the reinforcement algorithm to a certain extent.The security protection provides a certain basis.