基于Q-学习的非线性控制

Non-linear Control Based on Q-learning Algorithms

给出了一种新的激励学习(RL)方法,它能够有效地解决一个状态与动作空间为连续的非线性控制问题.在实际的应用中,离散的RL方法能把许多非线性控制问题分解成几个局部线性控制作业.提供了一种由简单的线性控制器和Q-学习构成的层次性RL算法.连续状态 动作空间被形象地分成为一些格子,并且每个格子都有它自己的局部线性控制器.状态空间的隔离是处理维数问题的一种相当简单的方法,但是经常会引起非Markov效果,而局部线性控制器能弥补这个缺陷.

A novel approach to reinforcement learning is given to solve a nonlinear control problem,in which state spaces are continuous and so action spaces are. In this method, an approach combining discrete RL methods with linear controllers is promising since there are many nonlinear control problems that can be decomposed into several local linear control tasks.