Design Of Terminal Guidance Law Based On Back-tracking Efficient Reinforcement Learning

Terminal guidance law phase has received extensive attention in control system in the current air-to-air warfare,as it is determined as the basis for judging the final intersection condition of the missile and target.The design methods based on the terminal guidance law mostly use the traditional proportional guidance law or its variants,which the effect is acceptable under ideal conditions.However,in the current terminal guidance scenario where the target maneuvering method changes frequently and the environment is noisy,the performance of guidance is poor,so the research and development for designing a new guidance laws have become a research hotspot nowadays.In view of the fact that the navigation ratio coefficient in the traditional proportional guidance law is a certain value,the missile of guidance cannot be made to adopt the corresponding navigation ratio as the current state of the missile and the target changes,that is,the ratio cannot be adjusted adaptively at each time and in each state.In order to increase the flexibility and scalability of terminal guidance,many researchers have turned to research on variable-coefficient guidance law design methods,combined with rapidly developing machine learning technology and deep learning technology,especially reinforcement learning for its ability for long-sequence interaction problems.A series of intelligent guidance law design methods have been proposed.However,due to the low convergence efficiency of the related technology,it cannot be effectively applied to the terminal guidance interception scenario and the guidance accuracy is poor.Therefore,a kind of reinforcement learning technology that efficiently and sensitively captures the value of the strategy implemented in the current state is the key innovation and breakthrough point of the current terminal guidance law design method.This article focuses on the poor performance of the terminal guidance law design method based on reinforcement learning.Through multi-dimensional and comprehensive summary and application of two high-efficiency reinforcement learning algorithms based on backtracking idea to design an adaptive terminal guidance law to achieve better guidance performance and improved guidance accuracy.This paper firstly designs the Markov decision process for the terminal guidance problem and models a simulation model that highly fits the real battlefield;secondly,this paper will use two efficient reinforcement learning algorithms,namely,efficient Q-learning adaptive terminal guidance law based on backtracking ideas method and EBU reversely update idea.We compare our method,the efficient and adaptive navigation ratio guidance law design,with existing reinforcement learning methods and classic methods in traditional fields.Through numerous experiments,we can demonstrate that the two methods used in this article can make significantly progress compared with common reinforcement learning methods and traditional methods in terms of miss distance,stability,and anti-noise ability,which reflects high efficiency.The performance can also provide some enlightenment for researchers in related fields when designing a new type of guidance law.