A new islanding detection and classification technique for the grid connected DFIG wind farms

This paper proposes an efficient islanding detection and classification technique based on the Extreme Learning Machine (ELM) algorithm for the wind farm based distributed generation (DG) connected to a power grid. In regard to design of any classification technique, learning time is a vital aspect to be considered. Thus having inherited with a good capability of generalization and an efficient and speedy leaning capacity, the single hidden layered ELM technique is an attractive technique for developing the detection and classification algorithms. The target feature set comprises of the Total Harmonic Distortion of current (THDi), the Harmonic Amplification Factor of voltage (HAFv) and the Negative Sequence Power (NSP) and is extracted from the efficient Adaline Gauss-Newton algorithm (instead of the conventional Widrow-Hoff delta rule), the input to which are the raw voltage and current at the target DG (where the wind velocity is varied from a minimum to maximum cut-off range). These target feature sets are then utilized by the proposed ELM technique that detects and classifies the islanding events to that of the non-islanding events (switching faults). The proposed DG Model is developed in MATLAB/Simulink environment, and the performance of the proposed ELM technique is compared to some of the conventional techniques in the MATLAB/Editor environment and the results analyzed are satisfactory.