New methodologies for the design of evolving fuzzy systems for online intelligent control

Resumen

The work presented in this dissertation is a contribution to the areas of automatic design of fuzzy systems and intelligent control. Specifically, this dissertation focuses in the automatic design of fuzzy controllers with less prior knowledge about the plants. Two new methodologies for online self-evolving fuzzy controllers are proposed. These methods are able to design fuzzy controllers from scratch, in an online manner, based on the analysis of the input/output data obtained from the normal operation of the plant. They do not use the information regarding the differential equations governing the system, or make any assumptions about them. The evolution of the controller is achieved through a life-long learning process that combines the adaptation of the rule consequents and the addition of new membership functions and rules. Due to their adaptive nature, these controllers are robust against unexpected changes in the plants and perform well in noisy environments. Moreover, the first methodology is able to automatically select the relevant control inputs, whilst the second tackles the curse of dimensionality and improves the controller¿s intrepretability. Simulation and experimental results illustrate the capabilities of the proposed methods. In addition, we study the increasingly popular type-2 fuzzy logic systems (FLSs), which are credited to outperform traditional (type-1) FLSs in the presence of uncertainties. We propose a multi-objective evolutionary algorithm for learning the structure and parameters of type-1 and type-2 fuzzy systems. This method is applied to obtain a common framework for the comparison of type-1 and type-2 FLSs in uncertain environments. Our final aim is to investigate whether the better performance of type-2 FLSs is solely based on the use of extra parameters (as it is often criticized), or whether it is due to the use of an essentially different mechanism for uncertainty handling.