Improved Shadowed Sets Data Selection Method in Extension Neural Network

Abstract

Due to the weaknesses of the generalization capability and diagnostic accuracy of identification of the improve extension neural network, we have found the method, Shadowed Sets, to improve the ENN. We can get the algorithm description through analyzing the Shadowed Sets and extension neural network system structure, making a comparison on a variety of extension neural network classifier performance, and detailed analyzing the Shadowed Sets of training data selection method. Finally, we can make a detailed experimental comparison through improved extension neural network (IENN) and traditional extension neural network (ENN), and at the same time get the results of the experiments by the Steam Turbine Power Generation Set Vibration Diagnostic tests. The experiments show that: we can use the shadow data set selection method and obtain high-quality training sample to enhance and improve the performance of ENN; selecting the nuclear data and boundary data as training ENN training data is a scientific and reliable; the improved extension neural network has been improved not only in the accuracy of classification, but the generalization ability