摘要
针对机动作战仿真背景,运用智能体理论研究战术机动智能体的最优机动决策问题。对传统的马尔科夫决策模型进行了扩展,通过定义攻击威胁下机动智能体的模糊状态空间、模糊状态转移规律和决策收益,建立了模糊战术机动决策模型,较好地描述了实际作战决策中的模糊认知、分析、判断等信息处理过程。通过引入强化学习手段,提出融合指挥员先验信息的Q学习算法和状态动态分类识别算法,对状态转移规律不易确定时模型的求解进行了研究;仿真实验验证了模型和算法的有效性。
For the background of maneuver operational simulation, the problem of optimal maneuver
decision based on agent theory is studied. The conventional Markov decision model is extended. A fuzzy tactics ma-neuver decision-making model is proposed. The model defines the fuzzy state space, the fuzzy state transfer regularity and decision lucre, which preferably describes the information processing of fuzzy cognition, analysis and judgement under attack threaten circumstance in realistic operations. Further, reinforcement learning is introduced to solve the model when the state transfer regularity cannot make sure. Dynamic classify of state space, identification of observation state and Q-learning with prior knowledge are researched in reinforcement learning process. The simulation experiment is carried out and the result testifies the validity of the model and algorithm.
出处
《系统工程与电子技术》
EI
CSCD
北大核心
2008年第3期511-514,共4页
Systems Engineering and Electronics
关键词
战术机动决策
智能体
建模
模糊理论
马尔科夫决策理论
强化学习
tactics maneuver decision
agent
modeling
fuzzy theory
Markov decision theory
reinforcement learning
