Participatory genetic learning in fuzzy system modeling

Abstract

Genetic Fuzzy Systems have been successfully used as a modeling approach for numerous applications. There is an increasing interest on how to construct fuzzy models for different types of complex systems such as highly nonlinear, large-scale, multiobjective, and high-dimensional systems. Current state of the art indicates the use of fast and scalable evolutionary algorithms in complex fuzzy modeling tasks. Genetic fuzzy systems offer an effective approach to embed genetic database learning and fast learning of parsimonious and accurate models. This paper suggests a participatory genetic learning approach as a tool for genetic fuzzy system modeling. Participatory genetic learning is an evolutionary computation paradigm in which the population itself plays an important role to assign fitness values to individuals. The approach uses compatibility between two randomly chosen individuals and the fittest to select the mates, and selective transfer recombination mechanism to exchange information between mates. Mutation is done similarly as in the canonical genetic algorithm. The usage of participatory learning, selective transfer, and mutation translates into a new type of genetic algorithm for genetic fuzzy system modeling. This paper focuses on the application of participatory genetic learning for rule-based fuzzy modeling of regression problems. Actual data concerning an electric system maintenance problem and results reported in the literature are employed to evaluate the performance of participatory genetic learning. The mean squared error and number of rules measure modeling accuracy and complexity, respectively. The result shows that participatory genetic learning produces accurate, parsimonious models, and is fast when compared with current state of the art approaches.