Online Global Learning in Direct Fuzzy Controllers

Abstract

Controlling non-linear systems in the most intelligent and efficient manner is the biggest and ongoing challenge for control engineers and scientists. Current research revolves around developing methods for control processes which can work with no plant information at all. Implementing online global learning in direct fuzzy controllers is an attempt to pursue this objective. Starting from an empty knowledge free controller, these controllers are able to learn, adapt and tune themselves with the desired results. Online tuning of rule consequents of the fuzzy controllers is discussed on the basis of plant output error i.e., adaptive control scheme was utilized. Modifying the adaptive controllers parameters by using only the qualitative information of the plant i.e., the monotony of input and output along with the adequate no of iterations required for learning was the basis of this work. Detailed error analysis on linear and non-linear plants concluded that the learning/tuning of the scheme was drastically improved with the correct combination of controller parameters. Finally, this empty knowledge free adaptive fuzzy controller was tested on a real plant in which it was successful in online controlling of water level on a real three tank system.