Development of a controller based on neural networks for a multivariable level and flow system

Abstract

This paper presents the development of a neural control based on the inverse model for multivariable hydraulics system of level and flow of Surcolombiana University. The degree of coupling between the control variables is evaluated through the method of relative gain matrix. The system modeling is done and the drivers are implemented in MatLab using the Toolbox and Neural Network Simulink as interface monitoring and control. The control performance is evaluated through simulations and real-time testing. The comparison of performance between the neuronal control and the fuzzy control is done, evaluating three parameters: overshoot, steady-state error and settling time. The results show a better performance against the previously developed fuzzy controller for the same system. It is evident that the steady-state error decreases significantly as the percentage of maximum error is 0.01% and 0.003% by neural networks and 3% and 1.75% by fuzzy control for level and flow respectively. As to overshoot although the fuzzy control is minimal, the control neural networks remove it entirely. For the settling time it is observed that the neuronal control also improves considerably for the two variables