牌型预测与蒙特卡洛模拟结合的麻将博弈策略  

Study on Mahjong game strategy based on tile prediction and Monte Carlo simulation

作　　者：李淑琴[1,2] 冯浩东 LI Shuqin;FENG Haodong(Computer School,Beijing Information Science and Technology University,Beijing 100101,China;Perception and Computational Intelligence Joint Lab,Beijing 100101,China)

机构地区：[1]北京信息科技大学计算机学院,北京100101 [2]感知与计算智能联合实验室,北京100101

出　　处：《重庆理工大学学报（自然科学）》2022年第12期148-154,共7页Journal of Chongqing University of Technology：Natural Science

基　　金：北京信息科技大学科技项目(5212010937,KM201911232002)。

摘　　要：针对麻将博弈中状态空间巨大和隐藏信息过多等难点,提出利用局面信息缩减未知状态空间,并通过动态划分游戏状态提升牌型预测准确率的方法。根据麻将领域知识将对手玩家的弃牌信息转换为可利用信息,预测对手玩家持有某一牌型的概率,使用这一概率约束蒙特卡洛模拟的范围,得到对手手牌及需求牌的概率分布。通过对比试验表明:局面信息利用方法和动态游戏划分方法是有效的,不仅可以降低出牌的点炮风险,还可以获得更高的得分,实现了通过缩减未知状态空间提升麻将博弈水平的目的。To address the difficulties of huge state space and too much hidden information in mahjong,this paper proposes to use situation information to reduce unknown state space and improve the accuracy of tile prediction by dynamically dividing game stage.Based on the domain knowledge of mahjong,this paper converts the opponent player’s discard information into available information,predicts the probability of the opponent player holding a certain tile type,and uses this probability to constrain the range of Monte Carlo simulation to obtain the probability distribution of the opponent’s hand and demand tiles.Finally,the comparison tests show that the situation information utilization method and the dynamic game division method are effective,not only to reduce the risk of discarding winning tiles for opponents,but also to obtain higher scores,achieving the purpose of improving the level of mahjong by reducing unknown state space.

关 键 词：麻将 计算机博弈 非完全信息博弈 领域知识 蒙特卡洛模拟 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]