Partial Discharge Intelligent Monitoring System Based on Multifractal Detrended Fluctuation Analysis and Gaussian Mixture Model

• Main Authors: Jia-SingWang, 王嘉興
• Other Authors: Cheng-Chi Tai
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/wmd3z2

碩士 === 國立成功大學 === 電機工程學系 === 107 === In this study, the multifractal detrended fluctuation analysis combined with the Gaussian mixture model is used to identify the partial discharge type of the power equipment, and to verify by the actual measurement signal. In order to construct a partial discharge intelligent monitoring system, the overall structure is divided into three parts: real-time detection system, communication link, and remote monitoring system. First, the real-time detection system uses the high-frequency current transformer to detect the current pulse signal on the ground line of the power equipment, and uses the NI PXI-5105 for data acquisition and transmission to the high-bandwidth embedded controller NI PXIe-8135. Then, feature parameters of the actual measurement signal are extracted by the multifractal detrended fluctuation analysis and is transmitted to the remote monitoring system using the protocol of IEC 61850. The remote monitoring system simultaneously reads the training data from the database to construct the category model, which is the Gaussian mixture model. The model optimizes the initial parameters through the K-means++ algorithm to make the system have more stable effects. Finally, the Bayesian decision theory is used to identify the measurement data received. In this study, all analytical methods and algorithms were designed on the human-machine interface of LabVIEW for measurement, and the feasibility of the system was verified by PD calibrator and partial discharge experiments.