Generator partial discharge identification method based on feature fusion word bag and parallel twin network

• Main Authors: PENG XIAOSHENG, ZHANG YUE, XIE ZHIHUI, ZHOU JIN, CHEN YUZHU
• Format: Patent
• Language: Chinese; English
• Published: 01.08.2023
• Subjects: CALCULATING; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; MEASURING; MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES; PHYSICS; TESTING

The invention discloses a generator partial discharge identification method based on a feature fusion word bag and a parallel twin network, and the method comprises the steps: collecting the partial discharge data and phase signal data of a generator stator bar in a laboratory, obtaining a corresponding phase map, constructing a word bag word book of SIFT, LBP, HOG and Haar-like multi-descriptor feature detection, and retrieving a similar partial discharge map; the method comprises the following steps: training AlexNet, GoogLeNet, VGG-16 and ResNet neural networks on the basis of partial discharge samples collected in a laboratory, then carrying out twinning and weight sharing on the trained neural networks, calculating loss by utilizing a full-connection layer cross entropy function, and selecting a label with the minimum loss as an identification result of each channel; and during testing, multi-channel identification results based on different phase quadrant segmentation of the atlas are integrated by usin