Lightweight Real-Time power quality disturbance recognition using Time-Frequency fusion with Cross-Attention mechanism
DOI:
https://doi.org/10.4108/ew.12734Keywords:
Power quality disturbances, Deep Learning, Dual-Pathway Architecture, Cross-Attention Mechanism, Time-Frequency Fusion, Smart Grid MonitoringAbstract
INTRODUCTION: Accurate power quality disturbances (PQDs) classification is critical for maintaining grid stability and reliability in modern power systems. However, existing deep learning methods predominantly rely on single-domain feature extraction, limiting their discriminative capability for complex composite disturbances under noisy conditions. This study addresses these limitations by proposing a dual-pathway architecture that synergistically integrates time-domain and frequency-domain representations through cross-attention fusion.
OBJECTIVES: This work aims to develop a robust PQDs classification framework capable of accurately identifying 24 disturbance classes, including complex composite types, while maintaining high noise immunity and computational efficiency for real-time monitoring applications.
METHODS: A dual-pathway deep learning architecture is proposed, comprising parallel CNN-BiLSTM branches for time-domain temporal modeling and FFT-based frequency-domain spectral analysis. A cross-attention mechanism dynamically fuses complementary features from both pathways. The model is trained and evaluated on a comprehensive dataset containing 24 PQDs classes under multiple noise levels.
RESULTS: The proposed model achieves 99.73% accuracy on the validation set and maintains 98.94% accuracy under 30dB noise conditions. Ablation studies confirm the dual-pathway structure improves accuracy by 6.51 percentage points over single-branch variants, while the cross-attention mechanism contributes an additional 2.08 percentage points. The model converges within 43 epochs with inference latency of 251μs per sample, satisfying real-time requirements.
CONCLUSION: The proposed dual-pathway cross-attention architecture demonstrates superior performance in PQDs classification, effectively balancing accuracy, noise robustness, and computational efficiency. This approach provides a viable solution for intelligent power quality monitoring in practical smart grid applications.
Downloads
References
[1] Aleem SHEA, Zobaa AF, Aziz MMA. Optimal C-type passive filter based on minimization of the voltage harmonic distortion for nonlinear loads. IEEE Trans Ind Electron. 2022;69(2):1583-93.
[2] Mishra S, Ray PK, Mallick RK. Power quality event classification under noisy conditions using EMD-based de-noising techniques. IEEE Trans Instrum Meas. 2023;72:1-11.
[3] Kumar C, Liserre M. A LSTM-based deep learning method for predicting grid voltage sag. IEEE Trans Ind Appl. 2023;59(1):954-63.
[4] Tan RHG, Ramachandaramurthy VK, Mansor M, Ekanayake JB. A comprehensive review on power quality disturbances and machine learning applications. Renew Sustain Energy Rev. 2024;183:113515.
[5] Shen L, Wang F, Cao L, Zhao J, Qiao W. Novel fusion-based deep learning network for multi-label power quality disturbances detection. Electr Power Syst Res. 2022;208:107881.
[6] Wang J, Xu Z, Che Y. Power quality disturbance classification via combining advanced S-transform and CNN with feature ranking. Measurement. 2023;203:111975.
[7] Shen L, Wang F, Cao L, Zhao J, Qiao W. Novel fusion-based deep learning network for multi-label power quality disturbances detection. Electr Power Syst Res. 2022;208:107881.
[8] Chen Y, Zhao D, Luo L, Wang Y. A light gradient boosting machine with Shapley additive explanation for power quality disturbances classification. Electr Power Syst Res. 2022;211:108163.
[9] Wang J, Xu Z, Che Y. Power quality disturbance classification via combining advanced S-transform and CNN with feature ranking. Measurement. 2023;203:111975.
[10] Yao Y, Wei H, Chen J, Zhou G, Liu Y. Power quality disturbance classification based on multiresolution convolutional neural networks. IEEE Trans Ind Electron. 2023;70(8):8421-30.
[11] Sha H, Mei F, Zhang C, Pan Y, Zheng J. Identification method for voltage sags based on K-means-singular value decomposition and least squares support vector machine. Energies. 2022;15(8):2831.
[12] Kumar C, Liserre M. A LSTM-based deep learning method for predicting grid voltage sag. IEEE Trans Ind Appl. 2023;59(1):954-63.
[13] Ahila R, Thangavel S, Jeyabharath R. A hybrid particle swarm optimization technique for power quality disturbance classification using wavelet transform and extreme learning machine. Electr Eng. 2022;104(5):3101-16.
[14] Tan RHG, Ramachandaramurthy VK, Mansor M, Ekanayake JB. A comprehensive review on power quality disturbances and machine learning applications. Renew Sustain Energy Rev. 2024;183:113515.
[15] Rodrigues Junior WL, Borges FAS, Rabelo RdAL, Fernandes RAS. A methodology for detection and classification of power quality disturbances using a real-time operating system in a computational embedded platform. Int J Electr Power Energy Syst. 2023;147:108860.
[16] Kiranyaz S, Avci O, Abdeljaber O, Ince T, Gabbouj M, Inman DJ. 1D convolutional neural networks and applications: A survey. Mechanical Systems and Signal Processing. 2021;151:107398.
[17] Jiang Y, Xie B, Wang J, Xia Y. Hybrid LSTM-based deep learning framework for fault diagnosis in power systems using PMU data. IEEE Trans Power Syst. 2023;38(5):4523-35.
[18] Li X, Wang W, Hu X, Yang J. Selective kernel networks with attention mechanism for power quality disturbance recognition. IEEE Trans Ind Inform. 2023;19(4):5aborr-54.
[19] Baig MAA, Ratyal NI, Amin A. A multi-modal deep learning framework for power quality disturbance classification: an integration of 1D time-series signals and 2D scalograms. Comput Electr Eng. 2025;128:110716.
[20] Luo Y, Wong Y, Kankanhalli M, et al. G-Softmax: improving intraclass compactness and interclass separability of features. IEEE Trans Neural Netw Learn Syst. 2020;31(2):685-699.
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Fei Yu, Haoyang Xue, Jiaming Qi, Xiaoqian Yue, Xinrui Pei
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
This is an open-access article distributed under the terms of the Creative Commons Attribution CC BY 4.0 license, which permits unlimited use, distribution, and reproduction in any medium so long as the original work is properly cited.
