基于变学习率的多agent学习算法的研究  

The study of multi-agent learning algorithm based on variable learning rate

作　　者：李琳娜[1] 

机构地区：[1]长春工程学院电气与信息工程学院,长春130012

出　　处：《长春工程学院学报（自然科学版）》2009年第4期81-83,共3页Journal of Changchun Institute of Technology：Natural Sciences Edition

摘　　要：对在动态学习的环境中的IGA算法做了研究,改进了梯度方向上的步长恒定不变的不足,引入了变学习率,并介绍了调节学习率的方法——WoLF原则,加速其收敛。最后根据该方法,对Q学习算法做了改进,并通过仿真试验证明了算法的有效性。This paper studied the IGA algorithm in a dynamic learning environment,and improved the insufficiency of step constantly invariable in the gradient direction.The variable learning rate and the WoLF principle to adjust learning rate were introduced in order to accelerate its convergence.Finally the Q learning algorithm was improved based on this method and the validity of the algorithm was proved through the simulation testing.

关 键 词：多AGENT学习 变学习率 Q学习 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]