Deep reinforcement learning intelligent guidance for intercepting high maneuvering targets

An adaptive proportional navigation law with intelligent parameter adjustment by deep reinforcement learning is presented to address the issue of excessive miss distances and energy loss in the interception of moving targets using fixed coefficient proportional navigation law. First, a state space based on real-time flight states, an action space containing lateral and vertical gains, and a reward function model integrating different states is established. Meanwhile, a prediction-correction method is introduced to improve the accuracy of action evaluation in the model design of the reward function. Secondly, the soft actor-critic (SAC) algorithm is employed to train a network parameter and guidance parameter decision system that takes into account the miss distances and energy loss according to the relative motion states of the interceptor and the target. In comparison to traditional proportional navigation guidance, the simulation results demonstrate that the guidance technique has strong adaptability and can greatly minimize energy loss while retaining low miss distances.