不完全信息条件下桥牌博弈算法的研究及应用

发布时间：2018-07-03 11:00

本文选题：桥牌博弈 + 机器博弈　；参考：《电子科技大学》2017年硕士论文

【摘要】：人工智能近年来受到越来越多的关注,并成为今年全国两会的热门话题。机器博弈的研究为人工智能提供了很多方法和理论,如博弈搜索等。机器博弈又分为完全信息博弈和不完全信息博弈,其中由于不完全信息博弈更贴近现实生活中的问题,如战争、股票市场及商场等,受到了越来越多研究者的重视。在现有研究工作的基础上,本文对不完全信息下桥牌博弈算法进行了研究,主要研究内容有以下几个方面:(1)提出了一种基于滑动窗口的抽样时间分配的GHA-BP叫牌学习策略。一直以来,叫牌法则的模糊性是计算机叫牌中需要解决的首要问题。现有的叫牌学习策略虽一定程度上解决了叫牌法则的模糊性,但其在抽样时间分配上不合理,并且基于ID3算法的叫牌预测准确度不高,导致学习叫牌策略能力有限。针对这些问题,本文提出了一种改进的叫牌学习策略,该策略采用滑动窗口对抽样时间进行预测分配,采用GHA-BP神经网络对模糊的叫牌法则进行分类学习。实验表明,该策略能够更合理地分配有限的时间和提高叫牌学习策略的准确率,降低与双明手叫牌结果的差异。(2)提出了一种基于启发式的蒙特卡罗打牌策略。一直以来,信息的不完全性是打牌过程中需要解决的首要问题。现有的蒙特卡罗打牌策略中通过随机抽样的方法为解决该问题提供了一种可行的思路,但常规抽样方法由于其盲目性存在抽样效率低的问题。因此,本文提出了一种改进的蒙特卡罗打牌策略,该策略将启发式的思想应用于生成样本牌局。实验验证了本策略能够在相同时间内产生更多满足叫牌约束的样本,能更准确地模拟当前状态的牌局,从而做出更为合理的打牌决策。(3)基于前面桥牌博弈算法的研究,本文设计并实现了一个计算机桥牌博弈系统。该系统包括控制系统和桥牌AI程序。控制系统实现了桥牌AI之间的通信功能,桥牌AI实现并验证了本文提出的叫牌及打牌算法。
[Abstract]:Artificial intelligence has received more and more attention in recent years, and has become a hot topic of the two sessions this year. The research of machine game provides many methods and theories for artificial intelligence, such as game search. Machine game is divided into complete information game and incomplete information game. Because incomplete information game is closer to the problems in real life, such as war, stock market and shopping mall, more and more researchers pay attention to it. Based on the existing research work, this paper studies the bridge game algorithm under incomplete information. The main research contents are as follows: (1) A GHA-BP bidding learning strategy based on sliding window sampling time allocation is proposed. All the time, the fuzziness of bidding rules is the most important problem to be solved in computer bidding. Although the existing bidding learning strategy solves the fuzziness of bidding rules to some extent, it is unreasonable in the allocation of sampling time, and the accuracy of bid prediction based on ID3 algorithm is not high, which leads to the limited ability of learning bidding strategy. In order to solve these problems, an improved bidding learning strategy is proposed, in which the sliding window is used to predict and allocate the sampling time, and the GHA-BP neural network is used to classify the fuzzy bidding rules. Experiments show that the strategy can allocate limited time more reasonably and improve the accuracy of bid learning strategy, and reduce the difference between the results of bidding and that of Shuangming hand. (2) A heuristic Monte Carlo strategy is proposed. The incompleteness of information has always been the first problem to be solved in the process of playing cards. The method of random sampling in the current strategy of Monte Carlo card playing provides a feasible way to solve this problem, but the conventional sampling method has the problem of low sampling efficiency because of its blindness. Therefore, an improved Monte Carlo card playing strategy is proposed, in which heuristic ideas are applied to generate sample cards. Experiments show that this strategy can generate more samples to meet the bidding constraints in the same time, and can more accurately simulate the current state of the card game, thus making more reasonable card playing decision. (3) based on the previous bridge game algorithm research, This paper designs and implements a computer bridge game system. The system includes a control system and a bridge AI program. The control system realizes the communication function between the bridge AI and the bridge AI, and verifies the bidding and playing algorithm proposed in this paper.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP18

【参考文献】