A Novel Algorithm Based on Wavelet and Parallel Neural Networks for Diagnosing Power Transformer Status

作者：Ma Tsao Tsung^*

来源：International Review of Electrical Engineering, 2010, 5(3): 970-979.

• Power Transformer
• Magnetizing Inrush Current
• Status Monitoring System
• Probabilistic Neural Network

This paper presents a new algorithm based on wavelet and parallel neural networks for diagnosing power transformer status. The proposed algorithm is developed by analyzing the energy distribution patterns derived from the inrush currents under various conditions, i.e., normal inrush currents, internal faults and insulation deteriorations. In this method, the discrete wavelet transform, a powerful tool in extracting information from the transient signals simultaneously in both time and frequency domains, is firstly applied to decompose the online detected inrush currents of the power transformer into a series of approximated and detailed wavelet components. Then, the energy distribution features of the wavelet components are calculated and employed to train the probabilistic neural networks to discriminate a variety of transformer internal status. The merit of this method is demonstrated by simulation of different faults and switching conditions on a power transformer using MATLAB/SIMULINK software. The simulated results clearly show that the proposed technique can accurately recognize various internal faults, insulation deteriorations and normal magnetizing inrush currents in power transformers. Also the proposed algorithm is tested with data practically collected from a real-life power transformer of Taiwan Power Company. Test results verify the feasibility and effectiveness of the proposed algorithm.

• 出版日期2010-6