Vol. 17, no. 4, 2025
РусскийEnglish
RADIOELECTRONICS
An Efficient and Robust Framework for Denoising and classification of Power Quality Events Based on the Adaptive Wavelet and SVM Techniques
Fadhel A. Jumaa, Hamdiyah Sekeb Jasim, Mohammed Obayes Yousif
Al-Furat Al-Awsat Technical University, https://atu.edu.iq/
Kufa-Najaf Al-Ashraf–54001, Iraq
Email: dr-fadhela.jumaa@atu.edu.iq, tcm.hamdia@atu.edu.iq, inm.moh2@atu.edu.iq
Anas Fouad Ahmed
Al-Iraqia University, https://en.aliraqia.edu.iq/
Al Adhmia-Haiba Khaton, 6029, Baghdad, Iraq
Email: anas.ahmed@aliraqia.edu.iq
Received May 30, 2025, peer-reviewed June 06, 2025, accepted June 09, 2025, published August 14, 2025
Abstract: The reliability and efficiency of modern electrical power systems hinge on accurately identifying and classifying power quality (PQ) issues. This paper introduces an innovative adaptive wavelet-machine learning framework that combines entropy-based wavelet selection, statistical feature extraction, and Support Vector Machine (SVM) classification. Unlike traditional fixed-wavelet methods, our approach enhances the time-frequency localization of both transient and non-stationary PQ events by dynamically selecting the most suitable wavelet for each signal, guided by a minimal entropy criterion.The signal is broken down with a 10-level discrete wavelet transform (DWT). Features like energy, standard deviation, and peak are retrieved at different scales. An RBF-kernel SVM is then trained with these features to classify nine typical PQ events. The average inference time is only 2.31 ms per signal, and the system remains robust in noisy conditions (SNR > 20 dB). Experimental data indicate that this framework reaches an average classification accuracy of 99.56%. Since the method outperforms several existing deep learning and fixed-wavelet models in terms of accuracy, interpretability, and computational efficiency, it is suitable for real-time PQ monitoring applications.
Keywords: denoising, discrete wavelet transform, categorization, support vector machine, power quality events
UDC 628.517.2
RENSIT, 2025, 17(4):431-438e DOI: 10.17725/j.rensit.2025.17.431
Full-text electronic version of this article - web site http://en.rensit.ru/vypuski/article/689/17(4)431-438e.pdf