Decentralized neuro-fuzzy controllers of nonlinear quadruple tank system

Abstract

This paper presented two decentralized neuro-fuzzy controllers to control the level in lower two tanks in nonlinear quadruple tank system (QTS). The controllers are designed based on adaptive neuro-fuzzy inference system technique. The relation between inputs/outputs was proved using relative gain array, and then, we can divide the quadruple tank system into two subsystems and control each of them separately. The first controller is a neuro-fuzzy inverse nonlinear (NFIN) model, which predicts the voltage required to control the level to track the referred one. So, the voltage was fed to neuro-fuzzy forward nonlinear model (NFFN) to obtain the desired level. The second is neuro-fuzzy nonlinear gain scheduling PI controller, which is designed to control the nonlinear QTS at any operating point. The results show that the NFIN controller has a more accurate tracking level and less computational time in both minimum and non-minimum phases.