On the Use of Participatory Genetic Fuzzy System Approach to Develop Fuzzy Models

Abstract

Genetic fuzzy systems constitute an essential approach to build fuzzy models. There is an increasing interest to develop fuzzy models in the realm of complex, large-scale, multiobjective, and high-dimensional systems. Nowadays, fast and scalable evolutionary algorithms to handle complex fuzzy modeling is a major need. Procedures to learn rule bases and tune their parameters are being shaped with the purpose to produce parsimonious and accurate models. Approaches to develop distinct types of fuzzy models such as type one and higher types, or higher order fuzzy trees, fuzzy relations, fuzzy cognitive maps, and neural fuzzy networks rare. This chapter introduces participatory evolutionary learning as a framework for data driven fuzzy modeling. The participatory evolutionary learning approach is a population-based paradigm in which the population itself defines the fitness of the individuals as evolution progress. The approach uses compatibility between population individuals during selection and recombination. A mechanism for information exchange in recombination based on selective transfer is introduced. Combination of participatory learning and selective transfer offers a new class of genetic fuzzy systems. Despite the focus on participatory learning and the selective transfer to build first order fuzzy rule-based models, the use of the genetic fuzzy systems to develop higher order fuzzy rule-based models is also discussed. An electric system maintenance data modeling problem is explored to illustrate the usefulness of the participatory genetic fuzzy systems approach in practice. The performance of participatory evolutionary learning is evaluated using the mean squared modeling error and number of fuzzy rules to measure model accuracy and complexity, respectively. The results suggest that the participatory evolutionary learning develops high quality models and is highly competitive with current state of the art approaches.