| 摘要 |
PROBLEM TO BE SOLVED: To increase the classifying accuracy of the category of a signal identification device, by making proper learning data to be used for the learning of a competitive learning type neural network. SOLUTION: The presence/absence of the abnormality of equipment X is decided by a competitive learning type neural network 1, by using the featured values of an object signal including oscillating components generated in the operation of equipment X. A data set, comprising a plurality of data whose categories are judged to be normal are stored in a learning data storage part 6. The competitive learning type neural network 1 is made to learn the data set. A learning data selecting part 4 searches the separation level of the weight vector of an ignited neuron in the output layer of the competitive learning neural network 1, after learning and each piece of data. The data whose separation level is a specific separation level threshold or higher are erased from the data set, until the mean value of the separation levels, calculated from the data of the data set, becomes smaller than the distribution threshold. COPYRIGHT: (C)2008,JPO&INPIT
|
