DeepPQDs-DWT-STNet: A novel DWT-ST and hybrid deep learning framework for power quality disturbance classification

The prevalence of power quality disturbances (PQDs) has grown significantly due to the widespread integration of nonlinear power electronic equipment and distributed generation systems, leading to energy losses and operational disruptions. Traditional classification methods struggle with noise and complex feature selection. This research introduces DeepPQDs-DWT-STNet, a hybrid deep learning framework combining Discrete Wavelet Transform (DWT) and Stockwell transform (ST). The DWT-ST contour plots visualize disturbance timing, location, and intensity across frequencies, while specific sub-bands (Q₁ to Q₄, P₄) quantify disturbance magnitude. The key features such as time, frequency, and shape-based characteristics are extracted and fed into the DeepPQDs-DWT-STNet model and then evaluated using synthetic and real time PQD data. The model achieves 99.98 % accuracy in noise-free conditions and maintains high performance with SNR levels of 50 dB (99.92 %) to 20 dB (99.65 %). Hardware-based experiments confirm its superior efficiency, processing each sample in 85 ms. Extensive validation was performed on five diverse datasets: DS1 (Clean Simulation), DS2 (Noisy Simulation), DS3 (Hardware Setup), DS4 (Solar PV) and DS5 (Wind Farm). The results demonstrate that DeepPQDs-DWT-STNet consistently delivers high classification accuracy across all conditions, including realistic renewable-based and hardware-induced disturbances, making it a reliable and effective approach for real-time PQD detection.