Research On Spatial Distribution Of Midcourse Penetration Decoys Of Ballistic Missiles Based On Deep Reinforcement Learning

Decoy technology is an important means of missile penetration,and the spatial distribution of decoy is a key link of decoy technology.In order to solve the problem of effective distribution of ballistic missile penetration decoys in airspace,the missile penetration process is modeled,and the missile and decoy motion model,interceptor seeker identification model and penetration effect evaluation model are established.Through DDPG and MADDPG deep reinforcement learning algorithms,the influence of different number of decoys on the penetration effect is simulated and analyzed,and finally the effective decoy distribution airspace is obtained.The main research contents of this paper are as follows:(1)Through the analysis of ballistic missile penetration process,the missile penetration process model is established.Firstly,the physical model of the bait is established to determine the size and quality of the bait;Analyze the placement mode and release process of decoy in the mother cabin,give five basic assumptions,and simplify the motion model of missile and decoy;The missile launch coordinate system and the relative motion coordinate system between the decoy and the missile are established.Based on the principle of relative motion and elliptical orbit operation,the motion equation of the decoy relative to the missile is deduced.Secondly,it expounds the recognition principle of infrared seeker,analyzes the working process of infrared seeker,and deduces the detection distance formula of infrared seeker in space background combined with the spectral detection rate in space background.Finally,the penetration effect of ballistic missile under the cover of decoys is studied,the main factors affecting the penetration effect are standardized and quantified,the penetration probability formula under the cover of multiple decoys is obtained,and the penetration effect evaluation model is established.(2)The problem of effective distribution of ballistic missile penetration decoys in airspace is transformed into the motion planning problem of decoys.Based on the theory of deep reinforcement learning,the decoys are regarded as agents in algorithm learning,and the space environment and the target as a whole are regarded as environments.The decoy motion environment is constructed,and the state space is designed based on the current position of the decoy,the distance from the enemy interceptor and the direction of flight speed;Taking the acceleration of the decoy as the action,the action space is defined.The distance base reward,range negative reward,collision negative reward and detected positive reward functions are constructed,and the comprehensive reward function is obtained.Using the characteristics that DDPG algorithm is suitable for dealing with continuous high-dimensional actions,DDPG algorithm and its improved MADDPG algorithm are selected for bait motion planning and design,and the network structure design is completed.(3)According to the established model,the motion of decoy and warhead is simulated,and the changes of velocity and relative missile distance after the release of decoy are obtained;the model training based on DDPG algorithm is carried out,and a certain configuration of decoy distribution airspace is obtained by simulation.The results show that the bait is distributed in the airspace of 10 km around the missile after release,forming a certain airspace configuration,and the missile has effective penetration effect.At the same time,it also verifies the feasibility of the ballistic missile penetration bait distribution airspace research model based on deep reinforcement learning.Finally,the model simulation based on MADDPG algorithm is carried out.On the basis of DDPG algorithm,the results also have faster convergence speed,higher penetration effect and more effective spatial distribution of missile penetration bait.When the number of decoys is enough,the decoy distribution airspace trained by DDPG algorithm can make the missile reach about 86% of the penetration probability,while the decoy distribution airspace trained by MADDPG algorithm can make the missile reach about 88% of the penetration probability,which improves the penetration effect compared with the former.It verifies the advantages of MADDPG algorithm compared with DDPG algorithm in multi decoy environment,and provides a reference for ballistic missile penetration decoys to form an effective distribution airspace.