Self-organising fuzzy modeling for nonlinear system control

The authors propose an algorithm for self-organizing fuzzy modeling (SOFM) which models the system by learning rules from input and output data, even though the rule set is empty at the beginning. The concept of tuning the cut value and gain introduced to SOFM helps to model the system more accurately and quickly. The cut value, like a filter, can be used to prune unimportant data and decide suitable rules from data possibilities. The gain, like a scaling factor, can adjust membership functions to determine the fuzzy model sensitivity. Simulation of a nonlinear system has shown the fuzzy model obtained in this way to be satisfactory. Pseudorandom binary sequence and Gaussian random noise signals have been added to the system to validate the modeling robustness. Using these results, the design of a hierarchical self-organizing fuzzy logic control structure which will include modeling, control and fault detection, is being investigated