Fuzzy neural networks based on rough sets for process modeling
Jianming Zhang · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2003
A new constructive method of the fuzzy neural network based on rough sets is proposed. First, we generate an initial fuzzy rule base from the history input-output data pairs. Then, in order to obtain the optimal rule base, the inconsistent or redundant rules of the initial fuzzy system are reduced by means of the rough set theory. Finally, we implement an optimal and simple fuzzy neural network by mapping from the optimal rules and train it with the initial data pairs. Since we determine the proper network structure and initial weights in advance, we can train the fuzzy neural network rapidly. The application to modeling of a nonlinear process reveals that it is effective and has good performances. The merit of this new method is to optimize the overall structure of fuzzy neural networks as well as to adjust each parameter of fuzzy rules without doing the complicated clustering process.