Detection and Classification of Islanding and Nonislanding Events in Distributed Generation Based on Fuzzy Decision Tree

This paper presents a new approach for islanding detection of distributed generation systems (DGs), using the features obtained from a new time-frequency transform with the negative sequence voltage, negative sequence current and the 3-phase voltage and current signals as inputs. The well-known S-transform suffers from high computational complexity, for on-line applications and hence, a new time-frequency transform (frequency filtering S-transform or simply the FFST) similar to it but faster by almost 30 times is proposed here using the frequency sampling and band pass filtering only for neighborhood of significant frequency components, determined from FFT of the power disturbance signals at the DG terminals. Thus using only the important frequency components present in the signal, it results in a significant reduction of computational complexity of the discrete S-transform. A data mining approach using a certainty factor-based fuzzy decision tree is used to yield fuzzy rules with the extracted features from the FFST output to recognize disturbance events like islanding or non-islanding for a variety of operating conditions of both the DGs and the electric power system. The results achieved during testing using existing hybrid distribution systems show that the proposed method is very reliable, and fast even in the presence of large disturbances like faults and capacitor bank switching.