A SELF-ORGANIZING NEUROFUZZY SYSTEM AND ITS APPLICATION TO QUALITY PREDICTION FOR HYDROCRACKING PROCESSING

摘要

In this paper a self-organizing neurofuzzy system (SONFS) is proposed. It is a feed-forward multiplayer network and learns in two main phases. In the first, the system utilizes improved nearest-neighborhood clustering scheme to determine the number of if-then rules and the parameters of fuzzy antecedent membership functions. Then supervised learning is utilized to adjust parameters to minimize the network output error. On the basis of the two-phase learning, the on-line learning algorithm is also proposed. Lastly, the SONFS is used in quality prediction of hydrocracking processing. Simulation results show that SONFS can provide better generalization results and simple model structure.