Image pattern recognition technique for the classification of multiple power quality disturbances

In a practical power system, a power quality (PQ) event often consists of various types of disturbances occurring at the same time. However, most of the existing techniques are trained to classify single disturbances, and hence have very limited usage in classifying real-time data. This work proposes a novel technique to automatically detect and classify multiple disturbances that coexist in PQ monitoring data. The data obtained from PQ monitors are represented in the form of 2-dimensional (2D) images. This 2D data visualization has significant benefits, in that each type of disturbance produces a different type of edge in the image; a sag produces horizontal edges, an oscillatory transient diagonal edges, and so on. The 2D discrete wavelet transform is applied to decompose these 2D images into 4 subbands using the Haar wavelet transform. Each of the events constituting the multiple disturbances has its own signature in each of the subbands and these signatures aid in the detection of these events. Pertinent features extracted from the wavelet subbands are utilized to train decision trees. These decision trees describe the rules needed for multiple event classification. Comparison results with the S-transform classification technique indicate that the proposed technique can offer faster classification and comparable accuracy, not only for a single disturbance, but also for multiple ones. Nevertheless, its memory requirement is very meager compared to S-transform. Moreover, it performs better than 1D wavelet-based event detection and classification techniques.