Online power quality events detection using weighted Extreme Learning Machine

This paper proposes a novel method for online power quality events classification using a machine learning based Weighted Extreme Learning Machine (W-ELM) classifier. For its fast response, easy to build structure and superior generalization, ELM is a preferred algorithm for different fields. W-ELM is an enhanced version of basic ELM whose striking feature is using a weight function for more effective classifying. Permutation entropy, local peaks, and LombScargle periodogram compose the powerful feature set with their powered ability to reveal the distinctiveness and low computational cost comparing to transform based methods. Dataset consisting of real site actual signals has been gathered from Turkish electricity transmission system. Most occurred events like voltage sag, swell, interruption and harmonics are included. We also assign conventional methods as artificial neural network and support vector machine with a basic ELM for a though analyze. Results prove that proposed system has the ability to operate both normal and faulty operations with its fast computational effort capable of online systems. Analyze studies on real-world PQ signals prove the validity of the presented algorithm.