一种博弈树静态估值算法——ΔFeature状态估值  

ΔFeature evaluation: static evaluation algorithm in game-tree searching

作　　者：叶品星[1] 

机构地区：[1]中山大学计算机系,广东广州510275

出　　处：《计算机工程与设计》2004年第7期1214-1217,共4页Computer Engineering and Design

摘　　要：在考虑下棋操作对棋盘影响的局部性后,提出了棋博弈的Δfeature状态估值算法,通过计算博弈树中相邻结点的特征变化来避免在叶结点上扫描整个棋盘,有效地减少了静态估值的时间开销。若棋子影响的局部范围足够小,还可以考虑将局部范围的所有情况列成表,以查表代替棋形匹配。ΔFeature状态估值算法也可以与其它优化博弈树搜索的方法一同使用,达到更好的效果。In many games, locality exists when considering the influence of operations on game state. Basing on such locality, a new static evaluating algorithm called Δfeature evaluation is described, which substitutes global scans of game state on leaf nodes for local scans on edges in game searching tree, and which results a considerable reduction of time consumption. In case of intense locality, state scans can be replaced by table searching, leading to a faster searching speed. ΔFeature evaluation can also co-operates with other game tree searching enhancements to reach an even better performance.

关 键 词：博弈树 静态评估函数 △Feature状态估值 空间搜索 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]