A Comparative Study of ANN and KNN for Power System Disturbance Classification

This study provides a comparative analysis of Kth Nearest Neighbors (KNN) and Artificial Neural Networks (ANN) for the classification of power system disturbances. To study the same, 5 events such as healthy, islanding, fault, capacitor switching, non-linear load switching evets are simulated rigorously in a PV integrated microgrid. KNN, an instance-based learning algorithm, is evaluated against ANN, a model-based learning approach, in terms of both training and prediction phases. Since, these two are feature based learners, 5 different features are also chosen as decision variables and 2000 cases are simulated to prepare a dataset of 2000*5. It is found that, KNN takes around 15000 operations during testing a single sample whereas ANN requires only 115 and detection accuracies are found to be 95.65% and 98.2% respectively presenting the superiority of ANN over KNN.