基于深度神经网络的桥牌叫牌策略研究  

Bidding model research on contract bridge game based on deep neural networks

作　　者：王璐瑶 吴蕾[1] WANG Luyao;WU Lei(College of Computer Science and Technology,Anhui University,Anhui 230601,China)

机构地区：[1]安徽大学计算机科学与技术学院,安徽合肥230601

出　　处：《应用科技》2025年第1期198-204,共7页Applied Science and Technology

摘　　要：桥牌是棋牌类游戏中最为复杂的游戏之一,由于其拥有着很多的隐藏信息,包含玩家之间的合作和竞争,同时也是不完全信息博弈的典型代表,具有重要的研究价值。定约桥牌包括2个部分:叫牌和打牌,而其中最具挑战性的任务是叫牌部分,它不仅需要队友之间的合作,还需要干扰对手之间的合作。文章以桥牌叫牌为研究对象,提出了一种基于深度神经网络的叫牌模型,用于给出下一步的叫牌决策。由于叫牌过程中的每一步都是密不可分的,当前的叫牌决策要受到之前的叫牌动作影响,所以文章采用了门控循环单元网络进行设计模型,并通过真实数据集的综合实验,验证了该模型的可行性以及相对于其他模型而言该模型对叫牌序列间关系更高的捕捉能力。Contract bridge is one of the most complex games in chess and card games.Because it has greater hidden information such as cooperation and competition between players,it is also a typical representative of incomplete information game,which has important value of research.Contract bridge consists of two parts:bidding and playing.Bidding is the most challenging task,which demands for effective cooperation between teammates as well as disrupting the cooperation of opponents.Focusing on the bidding phase of contract bridge,this paper proposes a bidding model based on deep neural network to facilitate decision making in the subsequent bidding actions.Given the inseparable nature of the bidding process,where current bidding decisions are intricately influenced by preceding bidding actions,a gated recurrent unit(GRU)network is employed to design the model.The feasibility of this model and its higher ability to capture the relationship between the bidding sequences compared with other models through comprehensive experiments on real data sets.

关 键 词：定约桥牌 机器博弈 不完全信息 叫牌 合作与对抗 深度学习 神经网络 门控循环单元 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]