Machine Learning Driven Islanding Detection using Random Forest and Scalogram

Islanding is a problem caused by distributed generation sources, causing potential damage to clients and equipment. To meet IEEE 1547 DG interconnection requirements, it must be detected within two seconds. This research introduces a novel method using image pattern recognition and machine learning classification algorithms, using random forest (RF) classifiers and histogram of oriented gradient (HOG) features for identifying islanding. The scalogram images are derived from the time-series signal collected from the point of common coupling. The images are utilised to extract HOG features, which are subsequently used as feature vector inputs for training and testing the RF classifiers. The scalogram image is derived using the rate of change of the voltage parameter. The proposed method achieved 98.62% accuracy and a detection time of 142 milliseconds.