Received 19 April 2023, accepted 6 May 2023, date of publication 10 May 2023, date of current version 16 May 2023. Digital Object Identifier 10.1109/ACCESS.2023.3274732 Power Quality Disturbances Detection and Classification Based on Deep Convolution Auto-Encoder Networks PORAS KHETARPAL 1 , NEELU NAGPAL 2 , (Senior Member, IEEE), MOHAMMED S. AL-NUMAY 3 , (Senior Member, IEEE), PIERLUIGI SIANO 4 , (Senior Member, IEEE), YOGENDRA ARYA 5 , (Senior Member, IEEE), AND NEELAM KASSARWANI 2 1 Information Technology Department, Bharati Vidyapeeth’s College of Engineering, Delhi 110063, India 2 Electrical and Electronics Engineering Department, Maharaja Agrasen Institute of Technology, Delhi 110086, India 3 Electrical Engineering Department, College of Engineering, King Saud University, Riyadh 11421, Saudi Arabia 4 Department of Management and Innovation Systems, University of Salerno, 84084 Fisciano, Italy 5 Department of Electrical Engineering, J. C. Bose University of Science and Technology, YMCA, Faridabad 121006, India Corresponding authors: Pierluigi Siano (psiano@unisa.it) and Neelu Nagpal (nagpalneelu1971@ieee.org) The work of Mohammed S. Al-Numay and Pierluigi Siano was supported by the Distinguished Scientist Fellowship Program, King Saud University, Riyadh, Saudi Arabia. ABSTRACT Power quality issues are required to be addressed properly in forthcoming era of smart meters, smart grids and increase in renewable energy integration. In this paper, Deep Auto-encoder (DAE) networks is proposed for power quality disturbance (PQD) classification and its location detection without using complex signal processing techniques and complex classifiers. In this technique, Gabor filter is used to extract a set of general Gabor features from the convolution of PQD image. Subsequently, through sparse based DAE network, essential and optimal features are extracted and learnt which are used by a simple classifier (SoftMax) to classify the PQD type. Furthermore, the temporal information of the PQD is obtained using the PQD image to correctly locate the disturbance’s initiating and terminating instants. The proposed DAE network has the benefits of Deep Learning-based networks in terms of automatic feature selection, but it requires smaller data sets. The issue of obtaining optimised, robust, and strong features from the PQD signal is thus resolved. Excellent classification accuracy of PQD is obtained with appropriate network parameter setting of the proposed DAE network. The proposed technique is compared with three other methods i.e. support vector machine (SVM), stacked auto-encoder (SAE) and principal component analysis (PCA) for PQD classification by implementing all the four techniques on python platform using the same data set. It has an overall classification accuracy of more than 97% at a signal to noise ratio (SNR) of 20dB, which is on the higher side when compared to other methods of PQD detection under noisy environment. Additionally, this method requires less computation time with the same data set than alternative approaches like SVM. Thus, the proposed method outperforms existing methods for PQD classification and detection of single disturbance and multi-disturbance in terms of greater accuracy and reduced computation complexity and computation time. INDEX TERMS Power quality monitoring, power quality disturbance, deep auto-encoders, optimal feature extraction, power quality event detection. The associate editor coordinating the review of this manuscript and approving it for publication was Nagesh Prabhu . 46026 This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/ VOLUME 11, 2023