A comparative study of neural and fuzzy-neural networks to identify a real system

Abstract

[EN]This paper treats a comparative study of neural networks and fuzzy neural networks used to model a complex biotechnological process: an activated sludge process taken from a real wastewater treatment plant. The neural networks used in this work are a multiplayer perceptron network and two recurrent neural networks: the Elman one and a neural network that represents the state space model. And other two fuzzy neural networks: the ANFIS network that calculates a Takagy-Sugeno type fuzzy logic system and a neurofuzzy system called FasArt, which is based on the Adaptive Resonance Theory (ART) but it also introduces fonnalisms from the fuzzy set theory. A comparative study of the five networks is carried out using real data collected fonn the plant in order to identify the dynamic behaviour ofthe sludge process in the wastewater plant.