Power Quality Disturbance Classification using Hilbert–Huang Transform and Deep Learning

Power quality disturbances (PQDs) pose significant challenges in power systems, affecting operational efficiency and equipment longevity. This paper presents a novel approach for PQD classification using the Hilbert–Huang transform (HHT) combined with deep learning models. The methodology converts sampled PQD signals into time–frequency–energy matrices via HHT, which serve as inputs for four machine learning models: Support Vector Machines (SVM), Naïve Bayes (NB), Vision Transformers (ViT) and a pre-trained MobileNet-v2 model. The classification performance is evaluated using accuracy, precision, recall, specificity and [Formula: see text]1-score. Among the models, the pre-trained MobileNet-v2 achieves the highest classification accuracy of 98.88%, outperforming ViT (98.22%), NB (95.69%) and SVM (94.71%). Additionally, MobileNet-v2 exhibits superior precision (96.48%) and recall (96.50%), indicating its ability to balance false positives and false negatives effectively. The confusion matrix analysis further highlights MobileNet’s robustness, with high true positive rates and minimal misclassifications, demonstrating its strong generalization capability. In contrast, traditional models like SVM and NB exhibit lower performance, particularly struggling with classes that have overlapping statistical characteristics, leading to lower recall and [Formula: see text]1-scores. The study validates the approach using a publicly available PQD dataset, confirming its effectiveness in accurately identifying and categorizing six types of PQDs. This research advances PQD classification techniques by offering a high-precision, computationally efficient and scalable solution for power system monitoring.