A module-based scalable identification system for power system overvoltage events

It is desirable to detect and identify different overvoltage waveforms based on underlying causes to guarantee the safe operation of power system and improve the reliability of power supply. This paper builds a module-based scalable identification system for power system overvoltage events. Each module is able to extract predefined features and identify one specific overvoltage event by integrating one or two signal processing techniques with Support Vector Machines (SVM). Firstly, based on the priori knowledge about signals caused by various overvoltage events, one or two signal processing techniques are selected to analyze recorded overvoltage signals. The signal processing techniques include RMS method, Fourier and Wavelet transforms. Then, a feature vector different from others is defined for each category of overvoltage events. Finally each SVM is trained by using predefined feature vectors as inputs. The system is scalable and robust. If a new overvoltage event needs to be identified, a new module can be added without retraining the existed modules. The prototype of the system is cross-validated using 247 field-measured overvoltage waveforms which cover six types of overvoltage events and 46 unknown overvoltage waveforms. The total identification rate is 97%. It shows the system can classify overvoltage events effectively and smartly.