Advanced Machine Learning Islanding Detection Scheme in PV Grid Tied System by Scalogram Image Based Features

The rapid growth of Distributed Generation (DG) is driven by the persistent need for electricity. Over time, nearly all DG sources are sustainable. Islanding is a significant issue that arises when there is an excessive integration of photovoltaic (PV) inverter-based DG sources. Islanding poses a potential risk to both the clients and their equipment. This study presents a novel approach using machine learning (ML) classifiers to detect islanding by categorising images. In order to distinguish between non-islanding and islanding events, images are utilised to generate the histogram of orientated gradient (HOG) features. The scalogram images are obtained from the time-series signal derived from the point of common coupling (PCC). The Histogram of Orientated Gradients (HOG) algorithm is employed to extract the distinctive characteristics from the images. The generated feature vector is then utilised as an input for the ML classifier during both the training and testing phases. The scalogram image is generated by utilising the total harmonics distortion (THD) characteristic of the voltage. We are assessing the efficacy of the ML classifier through the utilisation of 5-fold cross-validation. The classification results demonstrate that employing a ML classifier with the HOG feature for islanding detection when classifying images yields excellent outcomes and low validation error.