Online system identification based on a state-space neuro-fuzzy system

In this paper a new general recurrent statespace Neuro-Fuzzy model structure based on the combination of a modified Jordan network and an Adaptive Neuro-Fuzzy Inference System is proposed. The Neural-Fuzzy System's online training relies on a Constrained Unscented Kalman Filter, where weights, rules, membership functions and consequents are recursively updated. Results from a benchmark MIMO system demonstrate the applicability and effectiveness of the proposed framework.

I. Introduction

Data driven-based modelling has been under extensive research in the last decades, leading to a number of methodologies, and some particular nomenclatures emerging in different scientific areas (see e.g. [1]). In the control systems community, this paradigm is commonly referred to as System Identification.