This paper describes a novel fault classification and fault detection scheme using current signal data from only one end of a transmission system. Firstly, the measured current signals are decomposed using the wavelet transform to obtain the necessary frequency details and then the spectral energy for a chosen number of wavelet coefficients are calculated using a moving short time window; this forms the feature extraction stage, which in turn, defines the inputs for the neural network which is used for classifying the types of fault. After the fault type is identified, the proposed scheme selects the specific neural network of the fault type to distinguish between internal and external faults by utilizing the same patterns features extracted from the previous stage. The input features comprise both the high and low frequency components to enhance performance of the scheme. An extensive series of studies for a whole variety of different system and fault conditions clearly show that the performance of the scheme both for fault classification and detection is accurate and robust.
|Title of host publication||Power and Energy Society General Meeting, 2012 IEEE|
|Publication status||Published - 2012|
|Event||2012 IEEE Power and Energy Society General Meeting - San Diego, UK United Kingdom|
Duration: 21 Jul 2012 → 25 Jul 2012
|Conference||2012 IEEE Power and Energy Society General Meeting|
|Country/Territory||UK United Kingdom|
|Period||21/07/12 → 25/07/12|