A combined Data Mining and Machine Learning approach for High Impedance Fault Detection in Microgrids

Fault discrimination in the Microgrids has become a pivotal part of the modern Energy Management Systems (EMS) for optimal load dispatch. High Impedance Faults (HIFs), in particular, increase the ambiguity in fault discrimination as the magnitude of voltage-current variation during HIF is similar to the sudden change in load. In the present work, a method based on Machine Learning (ML) is proposed to improve fault discrimination considering HIFs. The data used for the training of ML models is preprocessed using a Density-based Spatial Clustering Algorithm with Noise (DBSCAN). Discrete Wavelet Transform (DWT) is used to process and obtain the combination of the unique parameters required for the input vector of the ML models such as Support Vector Machines (SVM), Decision Tree (DT), Multi-Layer Perceptron Neural Network (MLPNN), and Random Forest (RF). The combination of the data mining and ML algorithm shows improved accuracy in the detection of HIFs in acceptable computation time. PSCAD was used for simulation and PYTHON for data mining and ML model execution respectively.