Generating T-S Fuzzy System By Clustering Algorithm From Input-output Data

Estimating an unknown function from a set of input-output data pairs has been and is still a key issue in a variety of scientific and engineering fields. While the theory of traditional equation-based approaches is well developed and successful in practice, accurate mathematical models do not always exist nor can they be derived for all complex environments. So there has been a great deal of interest in applying model-free methods such as fuzzy systems for nonlinear function approximation. With the fuzzy control technology succeeding in applications, generating fuzzy systems from input-output data pairs has been a key subject in fuzzy control. This paper points out the drawbacks of the study in this field based on study results presented before. According to proposed problems, this paper proposes modified ISODATA algorithm based on linear prototypes to generate T-S fuzzy systems. Computer simulation experiments prove the efficiency of the proposed method. Fuzzy C-means algorithm also is modified based on linear prototypes, this paper deduces the modified FCM algorithm and uses it to generate T-S fuzzy systems, and experiment shows the feasibility. The proposed algorithms perform two steps. Step 1 uses modified ISODATA algorithm or modified fuzzy C-means algorithm to generate initial T-S fuzzy system. Step 2 optimizes the parameters of the system by particle swarm optimization algorithm. Finally, a summary is given and related subjects to be studied are presented.