Research On Scheduling Of Fly Target Acquisition Based On Deep Reinforcement Learning

With the development and progress of aircraft technology,the observation demand in the process of aircraft flight is also increasing day by day,and the observation process includes the problem of decisions in multi-time and multi-station circumstances.The problem is about the situation of uneven resource allocation and low efficiency,which is caused by the complex relationship and the high real-time requirements of the scheduling of on-line sensor resource.Therefore,the research on online scheduling of aircraft sensor resources is helpful to develop the response ability of the observation system and improve the efficiency of application,which is of importance in theory and application.The scheduling of aircraft sensor resource is a dynamic optimization decision problem,which has characteristic of the complex spatial state,the high real-time requirements and the acount of constraints.The complexity of the scheduling background and the high demand of the performance time make it difficult for traditional algorithms to solve this problem.At present,especially deep reinforcement learning,the new generation of artificial intelligence methods have made subversive breakthroughs in man-machine games and etc,which have solved some difficult problems that could not be solved before.It shows great advantages in the field of dynamic decision-making under complex conditions.Therefore,the method of aircraft sensor resource scheduling that is based on deep reinforcement learning is proposed by the paper,of which the main research includes the following aspects:Firstly,the operation mechanism of the aircraft sensor resource scheduling scene and process link is comprehensively analyzed,and the key variables involved in the sensor resource scheduling process are extracted,and the evaluation method to evaluate the advantages and disadvantages of the scheduling process is discussed.Then the aircraft resource scheduling is mathematically abstracted by the Markov decision process description method,and the state equation,action space and reward function of the scheduling model are designed respectively.And the logical relationship design is to be completed in the process of parameter change.Secondly,on the basis of Markov decision model,the resource scheduling framework of aircraft observation sensor based on deep reinforcement learning is constructed,and the reinforcement learning method based on strategy and Value-basedd method is analyzed.The Actor-Critic method which combines the advantages of the two methods is used as the basic algorithm to realize resource scheduling.Neural network is introduced as the strategy function in reinforcement learning.At the meaning time,the paper combines with the current mainstream training methods of deep learning to increase the convergence and applicability of the algorithm.Finally,in order to verify the correctness and availability of the aircraft observation sensor resource scheduling method,the basic framework of the aircraft observation sensor scheduling system is constructed by combining the algorithm function and visualization function,and the aircraft resource scheduling process simulation platform is established to estore the scene of simulating the common observation task resource scheduling process.And the scheduling algorithm is applied to schedule the simulated flight task in the simulation environment.And then the scheduling results generated by the algorithm are evaluated to compare and analyze the advantages and disadvantages of the scheduling algorithm based on deep reinforcement learning.