Distribution network fault detection and diagnosis using wavelet energy spectrum entropy and neural networks

This paper develops a hybrid fault detection and diagnosis method using Discrete Wavelet Transform (DWT) to extract characteristic features from transient waveforms obtained from disturbance recorders in electric power distribution networks. Entropy per unit indices are computed from the DWT decomposition of substation measurements made up of three phase and zero sequence currents, and are used as input to rule-based decision-taking algorithms and multilayer Artificial Neural Networks (ANNs). Different learning algorithms and architectures were experimented upon to obtain the structure of the ANNs. Comparisons, verification, and analysis made of the results obtained from the application of this method have shown good performance for different fault types, fault locations, fault inception angles, and fault resistances. The proposed method is distinct because of the processing stage done with DWT/wavelet energy entropy per unit formulation, and the use of practical equipment such as the Real-Time Digital Simulator (RTDS) and an Intelligent Electronic Device (IED) configured as a disturbance recorder