Passive Islanding Detection For Distributed Generation System

Islanding detection is indispensible for integrating distributed generators (DGs) into electrical power system (EPS). Passive islanding detection methods do not impact negatively on the power quality of EPS, and can easily be applied to DGs with different circuit topology and control algorithm. Passive methods may be classified as direct comparison based and intelligent based. Existing direct comparison based passive methods commonly have large non-detection zone (NDZ) and difficulties for threshold setting. Existing intelligent based passive methods focus on combining developed detection variables with intelligent algorithms, but seldom exploit the nature of the related detection variables changing during islanding occurrence.This work analyzed the impedance change as seen by DG before and after islanding. Three novel direct comparison based passive methods and one novel intelligent based passive method are proposed from perspectives of frequency dependent impedance change, reference impedance pre-calculation, signal energy spectrum analysis, and multi-impedance feature vector construction.(1) The interconnection topology of DG connecting to the utility grid is analyzed. The impedance change as seen by DG is characterized by an impedance based analytic model. A passive method based on tracking the frequency dependent impedance change (FDIC) is proposed. The feasibility of FDIC method is demonstrated with simulation and experimental results. FDIC method is shown to reduce the (NDZ) when combined with the UF/OF method.(2) A feature that characterizes the islanding condition is explored mathematically, a passive method based on reference impedance pre-calculation (RIP) is proposed. The robustness of RIP method is verified by cases study of typical islanding conditions and non-islanding conditions. Performance analyses are discussed from two perspectives of non-detection zone (NDZ) and fault detection zone (FDZ). With multiple DGs interconnecting to the distribution network, RIP method shows good robustness when different grid disturbance occurs. RIP method is able to detect islanding in the NDZ of UV/OV and UF/OF methods.(3) An energy spectrum analysis (ESA) based passive method is proposed, simulation and experimental results are provided for verification. The effects how different measurement quantization influence the proposed impedance based ESA method, RIP method and FDIC method are analyzed, possible solutions are suggested.(4) A multi-impedance feature vector is constructed and used for the design of a probabilistic neural network based classifier. A passive islanding detection method based on multi-impedance feature vector and probabilistic neural network (MIFVPNN) is proposed. Compared with using other feature vectors for the design of probabilistic neural network based classifier, analytic results show that the high detection accuracy of the proposed MIFVPNN method may be achieved with the appropriate selection of the multi-impedance feature vector.