文摘
This dissertation introduces a novel systematic approach for control-oriented modeling of dynamic systems in fuzzy logic frameworks. The concepts of entropy and mutual information are employed to develop a performance measure in the modeling process. The entropy approach, combined with ID3 machine learning methods, leads to a generic and automated procedure for establishing fuzzy models for dynamic systems. The approach offers the advantages of inherent ability of model complexity selections, fast convergence in searching for directions, and the ability to handle large samples of data. Self-consistency and convergence of the approach are established by employing the basic properties of non-negativity of mutual information. The approach is further incorporated with a tuning algorithm which adjusts the central points of membership functions to minimize modeling errors between fuzzy systems models and physical systems. The standard inverse control strategy is utilized as a typical fuzzy control method to illustrate potential applications of modeling approach in fuzzy logic control problems. Simulation results for various dynamic systems, as well as an automatic fuzzy control application, are presented.