| 摘要 |
PURPOSE: To find a degree of importance for respective inputs corresponding to outputs in order to decrease trial errors by enabling learning with a little data while efficiently utilizing previously recognized fuzzy rules, and extracting the fuzzy rules without reconstituting a network after learning. CONSTITUTION: This device is provided with a fuzzy neural network 10 equipped with a learning function for finding the degree of coincidence of rules from the combination of membership functions for concretely providing the fuzzy rules, constructing the network for leading the output corresponding to this degree of coincidence based on the number of items to be inputted/outputted and performing learning in order to correctly simulate the input/output relation of sample data expressed by an input pattern defined as an object and an output pattern corresponding to it, fuzzy rule setting part 30 for setting the fuzzy rules prepared by an engineer to the network 10, fuzzy rule extracting part 40 for extracting the fuzzy rules from the network 10 after learning, and importance degree extracting part 50 for finding the rate of contribution for the respective inputs to the output.
|
