Optimised autoencoder-based ensemble deep learning approaches for cyber-physical event classification utilizing synchrophasor PMU data
Dhinu Lal M, Ramesh Varadarajan · Results in Engineering · 2025
The increasing integration of Phasor Measurement Units (PMUs) into smart grid infrastructure necessitates accurate and efficient event classification of Cyber-Physical System (CPS) events to ensure real-time system resilience. This study tackles key limitations, including class imbalance, high-dimensional feature spaces, and computational constraints, in PMU-based event classification. The primary objective is to develop a reliable and optimised Deep Learning (DL) framework capable of classifying Power System (PS) events using real-world Synchrophasor datasets acquired from Mississippi State University and Oak Ridge National Laboratory (MSU-ORNL). The proposed methodology incorporates Adaptive Synthetic Sampling (ADASYN), Random Oversampling (ROS), and Synthetic Minority Oversampling Technique (SMOTE), followed by the selection of features by three variants of the Grey Wolf Optimiser (GWO): Standard GWO, Adaptive Weight GWO (AW-GWO), and Binary GWO (B-GWO), with B-GWO providing the optimal balance among accuracy and computing efficiency. DL architectures, including Long Short-Term Memory (LSTM), Bidirectional LSTM (BiLSTM), and Gated Recurrent Unit (GRU), are assessed, with improved performance using Autoencoder-based ensemble models. Among them, the Autoencoder-GRU ensemble attained a superior classification test accuracy and loss of 94.92 % and 0.1815, respectively. Further improvement through Random Search-based hyperparameter optimisation led to a peak test accuracy of 95.82 % and a reduced loss of 0.1681. The proposed framework shows low latency and high precision by leveraging GPU-accelerated processing, making it suitable for real-time deployment in smart grids. This work offers a scalable and generalisable solution for high-fidelity CPS event classification, providing timely insights for operational decision-making and fault response in Synchrophasor-enabled power systems.