对藏棋“久”的分阶段算法研究

藏族久棋是2019年中国计算机博弈锦标赛新设棋种,在此之前,国内外对该棋种的博弈策略研究相对较少。本文基于久棋两个博弈阶段规则和目的差异性大的特点,提出一种分阶段的博弈策略:下子阶段,考虑到无明显胜负判别的因素,提出一种基于相对胜负的改进蒙特卡洛树搜索算法以获取最佳下子点;行棋阶段,考虑到过程中的行棋方式会对后续模拟局面造成一定的影响,提出一种加入过程分值的改进Alpha-Beta剪枝搜索算法以获取最优行棋方案。在上述算法模拟博弈树的过程中,通过下子阶段优先集中在中心区域,行棋阶段优先形成褡裢的估值策略,给出了一份完整的估值评估表。实验结果表明,使用上述博弈策略及估值表实现的博弈程序棋力较高。

Research on the phased algorithm of Tibetan chess "Jiu"

Tibetan Jiuqi is a new chess type in the 2019 China Computer Game Championship.Prior to this,there were relatively few researches on the game strategy of this chess type at home and abroad.Based on the large differences in rules and objectives between the two game stages of Jiuqi,this paper proposes a staged game strategy:current stage,taking into account the factors that have no obvious win or lose judgment,an improved Monte Carlo tree search algorithm is proposed based on relative wins and losses to obtain the best move point;in the chess-moving stage,considering that the chess-moving method in the process will have a certain impact on the subsequent simulation position,an improved Alpha-Beta pruning search algorithm with process scores is proposed to get the best chess plan.In the process of simulating the game tree with the above algorithm,current stage is prioritized to focus on the central area,and the chess-playing stage prioritizes the formation of a comprehensive valuation strategy,and a complete valuation evaluation table is given.The experimental results show that the game program achieved by using the abovementioned game strategy and valuation table has higher chess power.