Research on Decision-making Method for Territorial Defense Based on Fuzzy Reinforcement Learnin

摘要

Protecting important targets, which also called territorial defense, will be an important application of the unmanned aerial vehicles (UAVs) in the future. This paper designs a method to generate interception strategy by learning, which can deal with invaders launched from different directions and different velocity. Firstly, we analyze the influence of initial states on the game results, and explored the initial condition boundary inside which the invader could be intercepted. Secondly, territorial defense game is a complex multi-steps decision-making problem which has continuous action and state spaces. To address this problem, conventional decision methods such as dynamic programming, moving horizon optimization method and Q-learning will cause dimension explosion issue. In this paper, we introduce a fuzzy logic into the actor-critic algorithm and reduce the amount of computation effectively. We consider invaders with different directions and speeds, can offer a more realistic result. Experiments showed that the algorithm can balance the exploration and utilization behavior well and the defender can learn to intercept the invader without prior knowledge.