ORIGINAL ARTICLE Comparison of modified teaching–learning-based optimization and extreme learning machine for classification of multiple power signal disturbances P. K. Nayak 1 • S. Mishra 2 • P. K. Dash 3 • Ranjeeta Bisoi 3 Received: 21 September 2014 / Accepted: 9 August 2015 Ó The Natural Computing Applications Forum 2015 Abstract This paper presents a modified TLBO (teach- ing–learning-based optimization) approach for the local linear radial basis function neural network (LLRBFNN) model to classify multiple power signal disturbances. Cu- mulative sum average filter has been designed for local- ization and feature extraction of multiple power signal disturbances. The extracted features are fed as inputs to the modified TLBO-based LLRBFNN for classification. The performance of the proposed modified TLBO-based LLRBFNN model is compared with the conventional model in terms of convergence speed and classification accuracy. Also, an extreme learning machine (ELM) approach is used to optimize the performance of the pro- posed LLRBFNN and is compared with the TLBO method in classifying the multiple power signal disturbances. The classification results reveal that although the TLBO approach produces slightly better accuracy in comparison with the ELM approach, the latter is much faster in implementation, thus making it suitable for processing large quantum of power signal disturbance data. Keywords LLRBFNN  Power disturbance signals  Feature extraction  Cumulative sum average filter  Pattern recognition  Simultaneous power quality events  Teaching–learning-based optimization  Extreme learning machine 1 Introduction The quality of electric power has become a pressing con- cern for electric utilities and their customers over the last decade. This is primarily due to the use of solid-state devices in power control resulting in voltage and current waveform distortions; voltage quality disturbances such as sags, swells, and interruptions; oscillatory and impulsive transients; multiple voltage notches due to solid-state converter switching; generation of harmonics; and the use of renewable energy sources. Therefore, power quality event detection and classification assume considerable importance in determining the sources of these distur- bances, thereby making it possible for taking appropriate actions in mitigating them. In the past decade, researchers analyzed the power quality issues with the increasing amount of measurement data from power quality monitors. The time of occurrence and the frequency of power quality disturbances [1–4] are unknown, so the monitoring is often required over an extended period. Further, the power quality disturbances can appear simultaneously as in real- istic power networks there are multiple sources of different disturbing events. Till date, most of the power quality analysis techniques have only considered a few power quality simultaneous disturbances namely voltage sags with harmonics, voltage swells with harmonics, and har- monics with interruptions. However, it is not enough in a realistic application scenario when there is a possibility of the occurrences of several simultaneous disturbances cre- ating a more adverse impact on the power distribution network. Thus, it will be desirable to evolve a methodology which will be able to analyze other combinations of two or more power signal disturbances. According to signal processing point of view, the non- stationary power signal disturbance analysis could be & P. K. Dash pkdash.india@gmail.com 1 Synergy Institute of Engineering and Technology, Dhenkanal, India 2 Centurion University, Bhubaneswar, India 3 S.O.A. University, Bhubaneswar, India 123 Neural Comput & Applic DOI 10.1007/s00521-015-2010-0