Automatic classification of voltage dip root causes via pattern recognition

Voltage dips (VDs) contribute significantly to the total annual cost resulting from poor power quality. This power quality disturbance can be induced by several root causes such as short circuits, transformer energizing, or due to the start-up of large electrical loads. The aim of this study was to develop a classifier which is able to automatically identify the probable root cause of a VD based on characteristic features contained within its corresponding RMS voltage curve. To this aim, mathematical functions were fitted through the characteristic section of VD RMS measurements. These measurements were obtained from the real-life distribution network. Subsequently, the coefficients of the fitting functions served as features for supervised pattern recognition schemes. In this study, 4 classifiers were developed and compared. The proposed approaches provided effective identification of VD root causes. Ultimately, effective classification schemes are a preliminary step to automatically localize VD sources.