Ensemble Learning of Decision Trees for Inverter-Interfaced Microgrid Protection

In this paper, inverter-interfaced microgrid protection in the islanded mode of operation is investigated using data mining approaches. Different disturbances including line and load switching, capacitor bank energization, Transformer inrush current and faults at different locations and with difference arc impedances are simulated. Local raw features are reduced using PCA and two cascaded classifier is used to define the each relay role i.e. primary, backup or idle and faulty phase selection. MATLAB $\backslash$ SIMULINK environment is used to simulate and record the data. Then, Different classifiers including k-nearest neighbors (KNN), Naive Bayes (NB), Linear Discriminant Analysis (LDA), Multilayer Perceptron (MLP) and Decision Tree (DT) are tested for fault detection only using local information and it is seen DT achieves the best performance. Finally, ensemble learning is used to enhance the DT accuracy.