Detection of Serial Arc Faults on Indoor Low Voltage Power Lines by Using Wavelet Transform and Neural Network

• Main Authors: Yu-Wei Liu, 劉鈺韋
• Other Authors: Chi-Jui Wu
• Format: Others
• Language: zh-TW
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/04468923999811591985

博士 === 國立臺灣科技大學 === 電機工程系 === 104 === This dissertation combines the discrete wavelet transform (DWT) with an artificial neural network (ANN) to identify the occurrence of serial arc faults on indoor low voltage power lines and expect to have better recognition. Dangerous serial electric arc faults on low voltage power lines must be detected in order to turn off the electric power sources before fire hazards occur. The detection technology is required to have high accurate recognition. However, the characteristics of line current waveforms during serial arc faults are complicated. The DWT is utilized to obtain the time-frequency domain characteristics of line current waveforms, and the data of some sub-bands by using signal energy method is useful information to reflect the serial arc fault patterns. And then, the appropriate expected values and the data of signal energy obtained from DWT can train a radial basis function neural network (RBFNN). After the training process, the RBFNN has excellent ability to identify the serial arc-fault circumstances. At last, the accumulative outputs of the RBFNN within 30 power cycles are used to determine whether serial arc faults occur on power lines or not. This dissertation compares the results of detecting serial arc faults with a commercial arc fault circuit interrupter (AFCI) to confirm the advantage of the purposed method. In the future, the proposed detection method in this dissertation can be combined with smart meters or other technologies to improve the factor of using electricity safely.