True RMS-Weighted KNN Based Underlying Cause Recognition of Power Quality Disturbance

Traditional approaches for recognizing the fundamental causes of power quality (PQ) disturbances typically involve time-consuming manual analysis of PQ data by experts. However, with the increasing complexity of power systems, there is a growing need for more automated approaches to PQ diagnosis. To this end, this paper proposes a true RMS and weighted K-Nearest Neighbor (KNN) based approach for recognizing three underlying causes of PQ disturbances i.e., capacitor energizing, three-phase fault, and induction motor starting. This true RMS value is calculated for each voltage sample in the power quality data. Using true-RMS voltage amplitude vs time curves, feature extraction has been taken place for several cases of single-stage and multiple disturbances. This feature set has further been provided to a weighted KNN based classifier. The weights are calculated using a Gaussian function, which gives higher weights to neighbors that are closer to the test point. The proposed classifier's accuracy has been compared with the other classifiers using classification leaner app in MATLAB. As a result of which, the highest training accuracy of 99% and testing accuracy of 90.47% has been obtained with weighted KNN algorithm. Thus, the weighted KNN results recommends that the proposed approach performs adequately for classifying single and multiple PQ disturbances and their causes.