Identification Inverted Pendulum System using Multilayer and Polynomial Neural Networks

It is well known that the inverted pendulum can describe a variety of inherently unstable systems, which is a major reason to consider it as a benchmark problem in control and identification. In this paper, a comparison between two different kinds of neural networks is presented, on one hand the feed-forward multilayer network with back-propagation learning method, and in the other hand the Volterra polynomial basis function network. A Fuzzy Logic controller was implemented to stabilize the system around its operation point. Both neural networks were trained using the error between the model´s output and the plant´s actual output. The polynomial network shows better performance against the multilayer network.