融合经验知识与深度强化学习的久棋Alpha-Beta算法优化研究  

Optimization of Alpha-Beta algorithm of Jiu Chess by combining empirical knowledge and deep reinforcement learning

作　　者：张小川 杨小漫 涂飞 王鑫 严明珠 梁渝卓 ZHANG Xiaochuan;YANG Xiaoman;TU Fei;WANG Xin;YAN Mingzhu;LIANG Yuzhuo(School of Artificial Intelligence,Chongqing University of Technology,Chongqing 401135,China)

机构地区：[1]重庆理工大学两江人工智能学院,重庆401135

出　　处：《重庆理工大学学报（自然科学）》2024年第5期115-120,共6页Journal of Chongqing University of Technology：Natural Science

基　　金：国家自然科学基金项目(60443004);重庆市技术创新与应用发展专项项目(cstc2021jscx-dxwtBX0019)。

摘　　要：藏族久棋作为一种传统的棋类博弈游戏,具备高度复杂的规则体系以及变幻莫测的棋局演变。传统的博弈策略在面对不同对手和棋局时不稳定,性能差,需要新的方法提高藏族久棋AI的博弈水平。以藏族久棋为研究对象,针对布局阶段,改进传统Alpha-Beta剪枝搜索算法,并结合经验知识,融入深度强化学习算法完成棋盘布局合理性的落子选择,以此为后续阶段铺路。在行棋阶段与飞子阶段,结合经验知识使用Alpha-Beta算法,完成行棋路径。最后,将所提算法和策略集成于久棋AI程序,在中国计算机博弈锦标赛中取得了良好的成绩,验证了该方法的有效性。Zangzu Jiu Chess,as a traditional board game,has highly complex rule systems and ever-changing board configurations.Traditional gaming strategies are unstable and perform poorly when faced with different opponents and situations,necessitating the development of new approaches to enhance the gaming capabilities of Zangzu Jiu Chess AI.This paper focuses on Zangzu Jiu Chess and,during the board layout phase,improves the traditional Alpha-Beta pruning search algorithm.It integrates empirical knowledge with deep reinforcement learning algorithms to make informed choices for rational piece placement on the chessboard,thereby paving the way for subsequent stages.During the chess stage and the flying stage,the Alpha-Beta algorithm is employed in conjunction with empirical knowledge to determine movement paths.Finally,the previously mentioned algorithms and strategies are integrated into a Zangzu Jiu Chess AI program,which achieves favorable results in the China Computer Game Championship,validating the effectiveness of this approach.

关 键 词：藏族久棋 经验知识 Alpha-Beta算法 深度强化学习 计算机博弈 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]