Fuzzy System Modeling Studies Based On Sample Data

The related theories of fuzzy control, simulating techniques and fuzzy logic are developed rapidly. They are applied widely in practical projects. In these backgrounds, the theories and techniques of fuzzy modeling are created and developed. It became an effective tool to resolve complicated, uncertain, nonlinear problems of fuzzy modeling and fuzzy control. This paper makes some research and discussions about fuzzy systems structure identification and parameters identification and fuzzy rules base simplification.The paper presents a new method for fuzzy partition and initial fuzzy rules gathering that based on monotonous areas of real systems, then we construct a initial fuzzy systems, which has a good initial state and semantic. Then parameters identification implement by ANFIS learning algorithm. The validity of this method is conformed by simulation. This paper proposes a new way to define a norm of rule importance and its adjust algorithm on line in ANFIS, and proved if ANFIS runs for a enough long time, the value of the norm was approximately equal transformation of the fitting error due to decided rule deletion. This paper conforms the range of the error when we use the value of the norm to replace the transformation of the fitting error. Simulation results show the validity of this method. In practical applying, it's often very difficult to get enough rules. This paper proposes a method to create new rules based on existent rules when needs. This algorithm roots the idea of interpolation. It designs a simple fuzzy system to implement interpolation to satisfy running on line. The stability and continuity of the system's output in the full universe got improvement.