Partial Discharge Detecting System Based on Convolution Neural Network and Data Fusion

Abstract

This research mainly uses the convolutional neural network (CNN) and data fusion technology to diagnose the partial discharge (PD) type of power equipment The system proposed in this paper is mainly divided into hardware and software parts In hardware we use NI PXI-5105 for data acquisition and transmit the PD signal of power equipment measured by sensors to the high-bandwidth embedded controller NI PXIe-8135 In software LabVIEWTM is used to perform signal preprocessing manufacture feature maps and construct user interface to provide user information We also use Python to implement the CNN algorithm First of all the discrete wavelet transform is used to filter the noise to get a noise-free PD signal After the signal pre-processing the system will manufacture phase resolved partial discharge (PRPD) patterns with 50 power cycle period as the input of the CNN The system will use the CNN as the classification model to identify the PD of power equipment In addition to CNN we also adopt data fusion technology to improve the system performance When the PRPD pattern is done feature-level data fusion technology combines the information of different sensors to provide complete event information for the classification model The pre-trained CNN model will be called by Python to identify the PD type At last Python returns the discrimination results to the user interface The user can look over the discrimination results through the user interface The system proposed in this paper is constructed by the technologies mentioned above and verified through the typical PD experiment

Date of Award	2020
Original language	English
Supervisor	Cheng-Chi Tai (Supervisor)