Detection and Classification of Multiple Power Quality Disturbances using a Deep Learning Framework

Power quality disturbance (PQD) detection plays a crucial role in maintaining the stability and quality of electrical power systems. The rapid growth of non-linear loads and integration of renewable energy sources have led to a rise in complex PQDs. Accurate detection of PQD is more helpful in mitigating PQD problems and to take corrective measures. Conventional machine learning (ML) techniques for PQD detection rely on sliding window and manual feature extraction, struggling with noise and multiple disturbances, whereas deep learning (DL) models enhance automatic feature extraction, but face limitations in in detecting multiple PQDs in a single time frame. To overcome these limitations, the proposed study introduces a You Only Look Once (YOLO)-based object detection framework that utilizes Hilbert Transform (HT) derived envelope signals. The simulated power signal is converted into its analytic form to compute envelope signal in a two-dimensional representation, capturing key amplitude, frequency, and phase features for improved PQD interpretation and classification. Performance evaluation demonstrates that the HT-based envelope signal significantly outperforms raw signal inputs in both noiseless and noisy conditions. In the noiseless condition, the model achieved a precision of 0.983, recall of 0.98, and $\mathrm{mAP} \text{@} 50$ of 0.984. Under noisy conditions ($30-50 \mathrm{~dB}$ signal-to-noise ratio), it consistently outperformed raw signals, achieving improvements of 14.91% in precision, 15.65% in recall, 9.68% in $\mathrm{mAP} \text{@} 50$, and 25.63% in $\mathbf{m A P} \text{@} 0.5: 0.95$. These results validate the proposed method’s noise-resilient and discriminative feature representation, enabling accurate real-time detection and localization of multiple PQDs within a single window time frame.