Hybrid Denoising Autoencoder–GRU Architecture for Robust Power Quality Disturbance Detection

This article presents a robust triple power quality disturbance (PQD) classification framework integrating spectral analysis, a denoising autoencoder (DAE), and a gated recurrent unit (GRU) network. The system is designed to detect and classify 16 classes of triple PQDs under various noise conditions. Synthetic PQD signals were generated per IEEE 1159 standards and subjected to additive white Gaussian noise (AWGN) at signal-to-noise ratio (SNR) levels of 5–20 dB. The spectral analysis transforms time-domain signals into the frequency domain to enhance class separability, while the DAE effectively denoises and compresses spectral data. The GRU network then models temporal dependencies for final classification. Extensive experiments reveal that the proposed model outperforms traditional baselines across all noise levels, achieving a peak accuracy of 99.7% in noise-free conditions and maintaining 85.6% at 5-dB SNR. Visual analyses, including power spectrum comparisons, t-distributed stochastic neighbor embedding (t-SNE), and DAE reconstructions, validate the model’s discriminative power and noise resilience. Benchmarking against recent methods confirms state-of-the-art performance, while validation on IEEE PES datasets verifies high accuracy and robustness under real-world conditions. These results demonstrate the framework’s strong generalization capability and practical utility for PQD monitoring applications.