一种基于经验的德州扑克博弈系统架构  

System architecture of Texas Hold’em based on experience

作　　者：高强[1] 徐心和[2] 王昊[3] 白国力 曹瑞珉 GAO Qiang;XU Xinhe;WANG Hao;BAI Guoli;CAO Ruimin(Key Laboratory of Manufacturing Industrial Integrated Automation,Shenyang University,Shenyang 110044,China;College of Information Science and Engineering,Northeastern University,Shenyang 110819,China;School of Mechanical Engineering and Automation,Northeastern University,Shenyang 110819,China)

机构地区：[1]沈阳大学辽宁省装备制造综合自动化重点实验室,辽宁沈阳110044 [2]东北大学信息科学与工程学院,辽宁沈阳110819 [3]东北大学机械工程与自动化学院,辽宁沈阳110819

出　　处：《智能系统学报》2020年第3期468-474,共7页CAAI Transactions on Intelligent Systems

基　　金：辽宁省自然科学基金项目(20180550146,20170520386).

摘　　要：为了利用历史经验知识提高德州扑克博弈水平,提出一种二人赌注无上限的德州扑克博弈系统架构:对于知识库模块,利用海量历史牌局训练得到基于CNN的深度学习网络模型并构建了一个专家经验库;在系统的搜索模块中,构建了一种分阶段的德州扑克博弈树,利用专家经验和历史经验引导德州扑克博弈树的展开;对于系统的估值核心模块,构建了一种基于哈希技术的牌型对照表,以提高系统判定胜负的效率。实验结果表明本文提出的博弈系统架构具有更高的对弈水平。To improve the level of Texas Hold’em through historical experience,this paper proposes a system architecture of heads-up no-limit Texas Hold’em for the knowledge base module.Mass historic games are used to train the deep learning network based on convolutional neural network,and an expert database is constructed for the search module of the system.Texas Hold’em structured game tree is developed and extended,and it is applied in terms of the expertise and historical experience to the core module for evaluation.A hand-ranking hash-based table is built to reduce the time required to evaluate hand rankings.The experimental result shows a higher playing level for the proposed system architecture.

关 键 词：二人赌注无上限德州扑克 计算机博弈 非完全信息动态博弈 博弈树 深度学习 专家库 哈希表 博弈策略 

分 类 号：TP301.5[自动化与计算机技术—计算机系统结构]