Fast Islanding Detection for Distribution System including PV using Multi-Model Decision Tree Algorithm

Modern distribution system including Distributed Generation (DG) requires reliable and fast islanding detection algorithms in order to determine the grid status. In this paper, a new multi-model classification-based method is proposed, in order to detect islanding condition for photovoltaic units. Decision tree is chosen as the classification algorithm to classify input feature vectors. The final result is based on voting among three decision tree algorithms. First order derivatives of electrical parameters are employed to construct feature vectors. To cover intermittent nature of renewable sources, different generating states for PV unit are assumed. Probable events are simulated under different system operating states to generate classification data set. The pro­po­sed method is tested on typical distribution system including the PV unit, different loads, and synchronous generator. This study sh­o­wed that this method succeeds in highly fast islanding det­ec­tion. This quick response can be used in micro-grid application as well as anti-islanding strategy. The results revealed that the proposed vot­ing-base algorithm could classify instances with very high acc­ur­a­cy which leads to reliable operation of distributed gene­rat­i­on units.

Keywords:

Data mining , distributed generation , Intelligent Classification , Micro-grid , Passive Islanding Detection