五子棋机器博弈系统评估函数的设计  

Design of evaluation-function for computer gobang game system

作　　者：张明亮[1,2] 吴俊[1] 李凡长[2] 

机构地区：[1]苏州科技学院电子与信息工程学院,江苏苏州215011 [2]苏州大学计算机科学与技术学院,江苏苏州215006

出　　处：《计算机应用》2012年第7期1969-1972,1990,共5页journal of Computer Applications

基　　金：国家自然科学基金资助项目(60775045);苏州科技学院科研基金资助项目(xky201010)

摘　　要：针对五子棋机器博弈需借助棋型评估棋局产生的速度瓶颈,提出了棋型识别粒度与搜索深度相关联的多层架构评估函数设计方法。实验证实粗颗粒度估值在深层搜索时整体效率更高;多层次判读棋型能明显加快评估速度;走法生成函数引入快速评估,提前进行博弈树自然剪枝收效显著;非关键棋型依据棋局浮动分值平衡搜索深度,对五子棋亦见成效;以及估值函数针对性地延伸评估来减少地平线效应的开销更少;从而表明估值函数动态评估更加合理。Concerning the speed bottleneck of the gobang machine game when relying on the configuration of the stones to evaluate the game states,this paper proposed a multi-layer evaluation-function method combining identification granularity of the stones＇ configuration with the search depth.The experimental results have proved： the overall efficiency of coarse-grained evaluation was higher when the deep-layer searches were processing;multi-layer judging for the configuration of the stones could obviously accelerate the evaluation;if the move-generating functions were introduced into the rapid evaluation,it was more efficient to prune certain branches from game-tree in advance;gobang game tree searches also benefited if floating the values of non-critical stone＇s configuration to balance the search depth.Therefore,dynamic evaluation of evaluation-function is more reasonable.

关 键 词：博弈树 PVS MTD(f) 评估函数 五子棋 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]