A genetic algorithm for learning fuzzy controllers

Publisher Website
Google Scholar
Abstract
Fuzzy rules and inferences provide a powerful framework for controlling complex processes. Like symbolic AI systems, they offer a comprehensible representation, which facilitates transfer of knowledge between the system and human experts and users. Moreover, unlike symbolic systems, the responses are more gradual. When human expertise is used to build a fuzzy conttoiler, it must often be tuned to fit particular objecrives or simply be corrected when it is incomplete or otherwise incorrect. When such expertise does not exist or is difficult to acquire and model in the controller, the knowledge must be generated. In either case, it is desirable to perform these tasks automatically. This can be accomplished by searching the space of all controllers, while sampling their performance to provide heuristic quality measures. This requires a robust search mechanism. Moreover, the search should be conducted simultaneously at all representation levels to avoid any assumptions about the knowledge. Generic algorithms provide the necessary robustness. In this paper~ ~ve investigate their applications to simultaneous manipulations of all representation levels. A number of experiments is conducted and reported, which illustrate both the tuning and learning capabilities of the approach. Preliminaries Advancing microprocessor technologies have brought automation capabilities to new levels of applications. Process control is a very important industrial application area which can greatly benefit from such advancements. However, development and deployment of applications are often difficult because of the complex dynamics of actual processes. Conventional control theory is based on mathematical models that describe the dynamic behavior of controlled systems. It is based on the deterministic nature of systems, but its applicability is reduced by computational complexity, parametric and structural uncertainties, and the presence of non-l.inearities. These charaeterisrics often make the controller design complicated and unreliable. A potential solution is offered by shifting the attention from modelling of the process to extracting the control knowledge of human experts. This artificial intelligence approach, implemented in Intelligent Control Systems (ICSs), utilizes the fact that human operators normally do not handle the system control problem with a detailed mathematical modal but rather with a qualitative, or symbol.ie, description of the controlled system. ICSs have two unique features: ability to make decisions and learning from data or experience. Decision making capabilities provide for the controllers to operate in real-time process-control environments, at both micro and macro levels of system operations. Learning capabiliries make it possible for the controller to adapt its knowledge to given performance criteria, to reason about potential dynamics of the environment, to predict advantageous features, or even to acquire the needed knowledge. ICSs have the ability to react to a

This publication has 0 references indexed in Scilit: