Neural network driven fuzzy inference system
R.J. Kuo, Paul H. Cohen, S.R.T. Kumara · 2002
Based on theoretical results, fuzzy systems are universal approximators. In this paper, the authors propose a novel learning approach, self-organizing and self-adjusting fuzzy modeling (SOSAFM), for inference rules. Basically, the proposed system consists of two stages, the self-organizing stage (SOS) and the self-adjusting stage (SAS). In the first stage, the input data is divided into several groups by applying Kohonen's feature maps. Gaussian distribution functions are employed as the standard form of the membership functions. Methods of statistics are used to determine the center and width of the membership function for each group. Regarding the consequences, the linear regression method is used. After the above procedures, one can decide the initial parameters of fuzzy systems. Then, the error backpropagation-type learning method is used to fine-tune the parameters. The simulation results show that the proposed approach is better than conventional neural networks in both accuracy and speed.>