STRUCTURE CONVERSION SYSTEM AND FUZZY MODEL EXTRACTION SYSTEM FOR HIERARCHICAL NETWORK STRUCTURE FUZZY MODEL EXECUTION DEVICE

摘要

PURPOSE: To provide an efficient means to automatically extract a fuzzy model from a hierarchical network of neuron by extracting a membership function and a fuzzy rule after converting a pure neuron to a rule part fully coupled neuron or rule part pre-wired neuron, and the rule part fully coupled neuron to the rule part pre-wired neuron. ;CONSTITUTION: An internal rule can be extracted with a format of fuzzy model by performing conversion to the rule part pre-wired neuron 6 or the rule part fully coupled neuron 7. In other words, the membership function for input can be extracted by deleting the connection with low influence of weight between the input layer of the pure neuron 5 and a second layer and that between the second layer and a third layer, and the fuzzy rule can be extracted by deleting the connection with low influence between the third layer and an (n-2)th layer, and finally, a consequent part membership function is extracted. Thereby, it is possible to convert the pure neuron 5 to the pre-wired neuron 7.;COPYRIGHT: (C)1994,JPO&Japio