Islanding Detection Based On The Multi-classifier Combination In Distributed Generation System

These days, with the increasing environmental problems and the continuous decrease of fuel energy, the exploiting of renewable energy has been paid widely attention. Using renewable energy, the distributed generation system (DG) is friendly to environment. It has advantages of safety and efficient operation as well as the ability to improve quality of power. Therefore, it has become the hotspot in the power system. The phenomenon of islanding is the key issue of DG. Rapid and efficient islanding detection techniques play an import role in ensuring steady operation of DG. Our study is in research of efficient islanding techniques, which is of significance and value.The main content contains the following three aspects:(1) Feature extraction consists of three parts:steady features, structural parameters based on system identification, wavelet singular entropy and local energy via wavelet analysis. The paper acquires the islanding classifier by data processing and machine learning. Classification algorithm includes logistic regression, SVM, random forest and Adaboost.(2) We build the multi-level, multi-model and multi-state islanding detecting system by model combination.(3) The simulation platform is built for verifying the effectiveness of the algorithm.There are three innovation points:(1) The paper proposes an islanding detection method based on system identification.(2) Optimal features are selected by combining Fisher criterion and sequential floating backward selection.(3) In consideration of various operation modes, the paper proposes a multi-level dynamic weighted voting combining modes for islanding detection.The innovations play an active role in islanding detection, which improves the performance of the detection algorithm.