多智能体系统中强化学习模型的改进及应用

多Agent系统中的强化学习是近年发展起来的一种新的人工智能方法,是以单Agent强化学习Q-learning算法为基础的一种学习模型，由于现有的强化学习模型还存在着结构信度分配困难、学习速度慢等缺陷,这些缺陷大大限制了多Agent强化学习模型的应用范围，本文对多Agent强化学习模型进行了系统的研究,分析了多Agent理论中强化学习面临的任务,指出在多Agent系统顺序型任务中遇到的时间信度分配难题及多Agent系统Agent间＂状态-动作＂对Q值估计的互通问题,对此问题提出了初步的解决办法,并在此基础建立了一个改进的多Agent强化学习模型,而且把该模型应用于电磁辐射源识别工作中。

Improvement and Application of Reinforcement Learning Model in the Multi-intelligent System

Reinforcement learning in the multi-agent system is a newly developed AI method, based on Qlearning algorithm of single agent system. Because the existing reinforcement learning models have some limitations such as distribution structure reliability problems and slow learning deficiency, these deficiencies substantially limit the scope and the application of multi-agent reinforcement learning. The model of reinforcement learning is systematically studied and the current reinforcement learning tasks in Multi-Agent system is analyzed. It focuses on the problems of intercommunication between action and Q-value assessment and allocated reliability problems. Some preliminary solutions are provided, and an advanced reinforcement learning model is built, and this model is applied into electromagnetic radiation source recognition.