Multiple local ARX modeling for system identification using the self-organizing map

Abstract

In this paper we build global NARX (Nonlinear Auto- Regressive with eXogenous variables) models from multiple local linear ARX models whose state spaces have been partitioned through Kohonen’s Self-Organizing Map. The studied models are evaluated in the task of identifying the inverse dynamics of a flexible robotic arm. Simulation results demonstrate that SOM-based multiple local ARX models perform better than a single ARX model and an MLP-based global NARX models.