Adaptation and learning of a fuzzy system by nearest neighbor clustering

Abstract

Conventional hill-climbing type adaptive algorithms suffer from large transient error in abrupt changing environments. In this paper, we suggest a novel idea to address this problem. A data re-initialization (DR) scheme is proposed using multiple models approach with special initialization technique to smooth transition between different environments. The design is based on adaptive fuzzy nearest neighbor clustering algorithm with on-line enhancements. The fast adaptation is realized by the data re-initialization and switching between different models; learning is realized by recording and retrieving the trained up models. The algorithm is conceptually simple and feasible for real-time applications. The performance of the algorithm is tested with simulation studies.