基于强化学习的作战辅助决策建模研究

Research on Modeling of Operational Assistant Decision Based on Reinforcement Learn

传统的辅助决策方法中,专家系统通常由固定知识表达,完全依赖领域专家知识,不具备学习能力。为了弥补专家系统的局限性,利用对智能体进行训练的方法自动生成决策方案。分析了自动机和行为树等过程性建模方法的优缺点,提出了基于Petri网和Q学习的CGF行为与决策建模方法。介绍了模型的基本结构和强化学习机制。在虚拟场景中,设计了决策模型实现方法和学习规则,通过对比实验,证明了此方法的可行性。

In the traditional assistant decision-making methods,the expert system is usually expressed by fixed knowledge,which completely depends on the domain expert knowledge and has no self-learning ability.In order to make up for the limitation of expert system,the method of training agent is used to generate decision scheme automatically.After analyzing the advantages and disadvantages of process modeling methods such as automata and behavior tree,a CGF behavior and decision modeling method based on Petri net and Q-learning is proposed.The basic structure of the model and reinforcement learning mechanism are intro⁃duced.In the virtual scene,the implementation method and learning rules of decision model are designed.The feasibility and good performance of this method are proved by comparative experiments.