论文部分内容阅读
提出了在动态环境中,多Agent的一种协作模型,适用于环境信息不完备的复杂情况.将Agent的独立强化学习与BDI模型结合起来,使多Agent系统不但拥有强化学习的高度反应性和自适应性,而且拥有BDI的推理能力,使只使用数值分析而忽略推理环节的强化学习结合了逻辑推理方法.使用了Borlzman选取随机动作,并且采用了新的奖励函数和表示方法,减少了学习空间,提高了学习速度.仿真结果表明所提方法可行,能够满足多Agent系统的要求.
A collaborative model of multi-agent in dynamic environment is proposed, which is suitable for the complex situation of incomplete environmental information. Combining the independent reinforcement learning of Agent and BDI model makes the multi-agent system not only possess the high reactivity of intensive learning and Adaptive, and has the reasoning ability of BDI, so that intensive learning using only numerical analysis while ignoring reasoning is combined with logical reasoning methods. Borlzman uses random actions and introduces new reward functions and representations to reduce learning Space and improve the learning speed.The simulation results show that the proposed method is feasible and can meet the requirements of multi-agent system.