Comparitive Performance Analysis of Various Training for Controlof Inverted Pendulum Using Narma-L2 Control

Abstract—In this study, system identification of an inverted pendulum using multi-layer perceptron neural network for modeling would be demonstrated. Moreover, the effect of initial condition and size of hidden layer of multi-layer perceptron neural network in control of an inverted pendulum plant would be presented. The inverted pendulum is an interesting plant because it could be a simple model for most physical systems. The nature of this kind of systems is unstable. The ability of controlling the unstable system in a wide range bout the operation point is the advantage of nonlinear controller with respect to linear controllers. Because of the plant that is used in this article is a nonlinear and unstable one, we attempt to control it by NARMA-L2 controller. In addition, the effect of different neural network on NARMA-L2 controller is studied. It is demonstrated that this kind of controller has a satisfactory performance by different scenarios.

Keywords:

Inverteed pendulum , neural network , NARMA-L2