Research on turn-based war chess game based on reinforcement learning

The search algorithm for turn-based war chess game is still using the traditional algorithm, and the whole war chess game needs a huge computational power, resulting in the research of turn-based war chess game in artificial intelligence is relatively backward. In this paper, we design a turn-based war chess game algorithm based on reinforcement learning, which is composed of Monte Carlo tree search (MCTS) and deep neural network. Monte Carlo tree search relies on strategy network to simulate chess several times, to choose better chess quality; The deep neural network reversely trains the strategy network through the final chess outcome. The experimental results show that there is no defeat in 10Kvs5K and 10Kvs3K game. This algorithm can improve the chess skill through self-game when there is no human knowledge and learning from scratch in turn-based war games.