Research On Mahjong Game Strategy Combining Hand tiles Optimization and Situation Search

Mahjong, being a classic imperfect information game, has a large playing space, a lot of hidden information, and complex game rules. It’s a fantastic research platform for artificial intelligence. This paper proposes a two-stage mahjong game strategy: in the early stages of the game, hand tile optimization based on the ShangTing distance and fundamental tile type is used to reduce the Hu distance and increase the Hu probability, and in the later stages of the game, a new type of Monte Carlo tree search algorithm is used to guide Hu selection and acquire more scores based on simulation findings. Finally, repeated comparison studies are used to test the strategy’ efficiency. The agent AgentMT was created using this approach, and it competed with the agent KnighTeam-MI01, which placed fourth in the 15th China computer gaming championship. At the end of the experimental part of the article, AgentMT and KnightTeam-MI01 are designed to play 3 sessions with 1000 games each. The experimental results show that, compared with KnighTeam-MI01, AgentMT is 53.3 points higher in the average score in a single session, 21.9 percent higher in the average time in a single session, and 3 points higher in the highest score in a single game.