基于ID3算法的智能代理决策系统设计与研究
发布时间：2019-05-27 18:06
【摘要】：游戏系统中智能代理担负着与玩家直接进行情感交互的重任,它的决策系统作为游戏可玩性的技术核心体现着“人工智能”的技术成果,直接关系着游戏设计的成败。但是当前主流的被市场广泛认可的智能代理决策系统都是基于预制逻辑进行搭建的,它们自身的动态调整能力有限,比较容易被玩家找出破绽从而降低决策智能,进而影响用户体验。而采用传统人工智能技术的智能代理决策系统,虽然能提供自主的决策智能,但是由于比较占用系统计算能力并且使用限制较多等原因,并不能被市场广泛认可。因此一种能继承传统决策系统的传统优势,同时又能结合新型人工智能技术,根据游戏运行情况进行实时、自主地动态调整的决策系统的研发变得尤为重要。本文对当前主流和新型的智能代理决策系统:模糊逻辑、人工神经网络、遗传算法、游戏脚本、有限状态机、行为树进行了介绍和分析。针对传统智能代理决策系统存在的不能很好的在实际使用环境中保持决策合理性,无法有效适配玩家多样化操作的问题,本文提出了一种基于机器学习ID3算法与行为树的ID3行为树作为智能代理的决策系统。它通过收集玩家与NPC交互时产生的后台数据借由ID3算法的分析指导行为树进行决策,实时调节决策系统,保持决策的合理性。为验证新决策系统的有效性,基于Unity3d平台设计了一款采用此方案的NPC智能代理,最后搭建了游戏系统并对其进行了实验,实验结果验证了此方案的有效性。
[Abstract]:The intelligent agent in the game system undertakes the important task of directly interacting with the players. As the technical core of the playability of the game, its decision-making system embodies the technical achievements of "artificial intelligence", which is directly related to the success or failure of the game design. However, the current mainstream intelligent agent decision-making systems, which are widely recognized by the market, are built based on prefabricated logic, and their own dynamic adjustment ability is limited, so it is easier for players to find out the flaws and reduce the decision intelligence. And then affect the user experience. Although the intelligent agent decision system based on traditional artificial intelligence technology can provide independent decision intelligence, it can not be widely recognized by the market because it occupies the computing power of the system and has more restrictions. Therefore, it is particularly important to develop a decision system which can inherit the traditional advantages of the traditional decision system and combine the new artificial intelligence technology to carry out real time and independently dynamic adjustment according to the operation of the game. In this paper, the mainstream and new intelligent agent decision systems, such as fuzzy logic, artificial neural network, genetic algorithm, game script, finite state machine and behavior tree, are introduced and analyzed. In view of the problem that the traditional intelligent agent decision system can not maintain the rationality of decision in the actual use environment and can not effectively adapt to the diversified operation of players, In this paper, a ID3 behavior tree based on machine learning ID3 algorithm and behavior tree is proposed as an intelligent agent decision system. By collecting the background data generated by the interaction between players and NPC, it guides the behavior tree to make decisions through the analysis of ID3 algorithm, adjusts the decision system in real time, and maintains the rationality of the decision. In order to verify the effectiveness of the new decision system, a NPC intelligent agent based on Unity3d platform is designed. Finally, the game system is built and its experiments are carried out. The experimental results verify the effectiveness of the scheme.
【学位授予单位】：合肥工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP18
本文编号：2486367
[Abstract]:The intelligent agent in the game system undertakes the important task of directly interacting with the players. As the technical core of the playability of the game, its decision-making system embodies the technical achievements of "artificial intelligence", which is directly related to the success or failure of the game design. However, the current mainstream intelligent agent decision-making systems, which are widely recognized by the market, are built based on prefabricated logic, and their own dynamic adjustment ability is limited, so it is easier for players to find out the flaws and reduce the decision intelligence. And then affect the user experience. Although the intelligent agent decision system based on traditional artificial intelligence technology can provide independent decision intelligence, it can not be widely recognized by the market because it occupies the computing power of the system and has more restrictions. Therefore, it is particularly important to develop a decision system which can inherit the traditional advantages of the traditional decision system and combine the new artificial intelligence technology to carry out real time and independently dynamic adjustment according to the operation of the game. In this paper, the mainstream and new intelligent agent decision systems, such as fuzzy logic, artificial neural network, genetic algorithm, game script, finite state machine and behavior tree, are introduced and analyzed. In view of the problem that the traditional intelligent agent decision system can not maintain the rationality of decision in the actual use environment and can not effectively adapt to the diversified operation of players, In this paper, a ID3 behavior tree based on machine learning ID3 algorithm and behavior tree is proposed as an intelligent agent decision system. By collecting the background data generated by the interaction between players and NPC, it guides the behavior tree to make decisions through the analysis of ID3 algorithm, adjusts the decision system in real time, and maintains the rationality of the decision. In order to verify the effectiveness of the new decision system, a NPC intelligent agent based on Unity3d platform is designed. Finally, the game system is built and its experiments are carried out. The experimental results verify the effectiveness of the scheme.
【学位授予单位】：合肥工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP18
【参考文献】
相关期刊论文 前5条
1 刘晓伟;高春鸣;;结合行为树与Q-learning优化UT2004中agent行为决策[J];计算机工程与应用;2016年03期
2 黄宇达;王迤冉;;基于朴素贝叶斯与ID3算法的决策树分类[J];计算机工程;2012年14期
3 王永梅;胡学钢;;决策树中ID3算法的研究[J];安徽大学学报(自然科学版);2011年03期
4 但小容;陈轩恕;刘飞;柳德伟;;数据挖掘中决策树分类算法的研究与改进[J];软件导刊;2009年02期
5 李建珍;基于遗传算法的人工神经网络学习算法[J];西北师范大学学报(自然科学版);2002年02期
相关硕士学位论文 前2条
1 王振宇;计算机游戏中智能角色行为的研究与实现[D];湖南师范大学;2010年
2 王黎明;决策树学习及其剪枝算法研究[D];武汉理工大学;2007年
，本文编号：2486367
本文链接：https://www.wllwen.com/kejilunwen/zidonghuakongzhilunwen/2486367.html
