基于进化行为树的智能体决策行为建模研究

针对传统仿真系统中建模方法存在领域知识获取困难、生成行为固定、缺乏适应性等问题,提出基于静态约束的进化行为树方法构建智能体决策行为模型.在通用进化算法基础上自动生成反映智能体决策行为逻辑的行为树拓扑结构,通过通用学习方法与领域规则的较好结合提升决策模型的生成效率和适应性,并以坦克对战军事游戏中的决策行为建模为例进行验证.结果表明:该方法能增强决策逻辑的可解释性,具备可行性和科学性.

Research on Decision-making Behavior Modeling Based on Evolutionary Behavior Tree

Aiming at the problem of acquiring domain knowledge, fixed behavior pattern and adaptability lack of behavior modeling in traditional simulation systems, this paper establish decision-making behavior model, proposes an evolutionary behavior tree method based on static constraints. Based on the general evolutionary algorithm, the behavior tree topology structure that reflects the decision-making behavior logic is automatically generated, and the generation efficiency and adaptability of the decision-making model is improved through combination of general learning methods and domain rules. Preliminary experiments in tank game validate the proposed method. The results show that the method enhances the interpretability of decision-making logic, and is feasible and scientific.