Knowledge-based fault detection and identification for high-voltage power systems

This thesis introduces a Knowledge Based approach for High-Voltage power system faults detection and identification. Based on the feature of the typical signals obtained from the Transcan Recording System (TRS), a dual approach is pursued. Feature extraction is central to this thesis. Various features of power system signals are extracted to provide a basis for a decision support system for power system fault and identification. First, faults that have periodic signals such as phase current and 6 pulse signals, and Valve currents are analyzed using FFT and auto-correlation to identify the type of the waveform of the input signal. Second, for faults that have non-periodic signal such as pole line voltage, pole current and pole current order, a new method called Fuzzy Wavelet Analysis is introduced to determine the type of the faults. In addition, there are also some other attributes like the Ratio of Phase current and current order, ac Phase voltages Error that are analyzed using granular computing methods. Finally, we use the above attributes to set up a decision table and then use Rough Set rule generation tool called Rosetta to generate fault-classification decision rules. Performance evaluation of detectability and identifiability are defined to assist in assessing the performance that is achieved through a learning mechanism based on the detectability and identifiability measures.