Machine Learning Based Islanding Detection for Grid Connected Photovoltaic System

This paper focus on developing a new islanding detection method with the help of machine learning and signal processing technique. The islanding detection method make sure that there is a proper remote monitoring of the grid integrated photovoltaic (PV) system. In case of grid fault or maintained of the grid, a proper informed signal is provided to various distributed generation (DG) networks so that they can disconnect with the grid and operate in isolated mode. A simulation of 1kW grid connected PV system is performed. The signal such as voltage, current and frequency are recorded at point of common coupling (PCC). The feature of recorded signals are extracted using wavelet transformation. The extracted features are used to form a islanding scenarios matrix. The matrix is further utilized to train a classifier using machine learning algorithm. From the result it can be observed that the trained classifier depicted 97.9% training accuracy with a training time of 16.9 sec which is better when compared with the literature. Further the trained classifier is subjected to test with an unknown islanding condition to observe the robustness of the classifier.