Adaptation and Learning with Fuzzy Clustering Algorithms: A Data Re-Initialization (DR) Scheme

Abstract

Abstract This paper presents a novel idea to avoid large transient error of hill-climbing type adaptive algorithms in abrupt changing environments. A data re-initialization (DR) scheme is proposed using multiple models approach with special initialization technique for smooth transition between different environments. The design is based on adaptive fuzzy clustering algorithm with on-line modifications; furthermore, it is equipped with learning capabilities. The fast adaptation is realized by DR and switching between different models; learning is realized by recording and retrieving the trained up models. The algorithm is conceptually simple and simulation results are appealing. Finally, conclusion and remarks are given.