一种基于递归神经网络的自适应控制方法研究

A Novel Adaptive Control Algorithm Based on Recurrent Neural Network

本文针对快速、多变量、强非线性的复杂系统的控制问题,在强化学习方式的基础上,提出一种新的自适应控制方法。该方法在没有先验知识的条件下,基于递归神经网络并结合强化学习的自调节能力,通过自身神经网络的在线学习,有效控制不稳定的非线性系统。本文以一级倒立摆系统为实验对象,仿真实验结果表明:所提出的控制方法具有非常好的控制效果和稳定精度,抗干扰能力强。

Based on reinforcement learning, a novel adaptive control algorithm is proposed for the complex systems which have the characteristics of speediness, multiple variables, serious nonlinear. The method based on recurrent neural network needs not know the priori knowledge of system, combines the self-tune property of reinforcement learning through on-line learning of network, and at last effectively controls the unstably nonlinear system. The experimental object is a single inverted pendulum. It is shown from the simulation results that this method has good control effect, good steady accuracy and good interference rejection.