Open AccessAn implementation of type‐2 fuzzy kernel based support vector machine algorithm for power quality events classification
Author(s) -
Naderian Sobhan,
Salemnia Ahmad
Publication year - 2017
Publication title -
international transactions on electrical energy systems
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.428
H-Index - 42
ISSN - 2050-7038
DOI - 10.1002/etep.2303
Subject(s) - support vector machine , kernel (algebra) , artificial intelligence , computer science , sliding window protocol , pattern recognition (psychology) , algorithm , iterated function , radial basis function kernel , window function , power quality , kernel method , mathematics , window (computing) , power (physics) , spectral density , physics , combinatorics , quantum mechanics , mathematical analysis , telecommunications , operating system
Summary This study presents a new method for detection and classification of power quality (PQ) events, based on discrete Gabor transform (DGT) with a finite impulse response window (FIR‐DGT) and type‐2 fuzzy kernel based support vector machine (T2FK‐SVM). The features were extracted through FIR‐DGT and T2FK‐SVM classified PQ events. Using proper window function for DGT is essential. Iterated sine window was used as window function to extract events features. Iterated sine window function is 4 times faster than the default window function of DGT. Kernel design is a main part of many kernel‐based methods such as support vector machine (SVM), so using type‐2 fuzzy sets as an SVM kernel, the total accuracy of classification is enhanced. This method can reduce the features' size of the disturbance signals significantly, so less time and memory are required for classification via T2FK‐SVM method. Nine types of events are simulated in the classification problem. The simulation results revealed accurate classification and fast learning and execution in the detection and classification of PQ events. The findings were compared with other methods, and the accuracy of the proposed method was evaluated under noisy and real conditions.