Machine Learning Based Protection Scheme for Low Voltage AC Microgrids

The microgrid (MG) is a popular concept to handle the high penetration of distributed energy resources, such as renewable and storage systems, into electric grids. However, integration inverter-interfaced generation units (IIDGs) imposes control protection challenges. Fault identification, classification isolation are major concerns with IIDGs-based active MGs where IIDGs reveal arbitrary impedance thus different fault characteristics. Moreover, bidirectional complex power flow creates extra difficulties for analysis. This makes conventional methods inefficient, new paradigm in schemes needed IIDGs-dominated MGs. In this paper, machine-learning (ML)-based technique developed IIDG-based AC by extracting unique novel features detecting classifying symmetrical unsymmetrical faults. Different signals, namely, 400 samples, wide variations operating conditions an MG obtained through electromagnetic transient simulations DIgSILENT PowerFactory. After retrieving pre-processing 10 feature extraction techniques, including peaks metric max factor, applied obtain 100 features. They ranked using Kruskal–Wallis H-Test identify best performing features, apart from estimating predictor importance ensemble ML classification. top 18 used input train 35 learners. Random Forest (RF) outperformed all other classifiers detection type faulted phase identification. Compared previous methods, results show better performance proposed method.