Sreenu Sreekumar

Power Quality Disturbances (PQDs) must be accurately detected and classified to preserve stability, efficiency, and power reliability in modern power systems as nonlinear loads and renewable energy sources become more common. Despite the excellent classification accuracy of Deep Learning (DL) models, their computational complexity and implementation feasibility are often overlooked. This work uses time-frequency representations obtained from the Fourier Synchrosqueezing Transform (FSST) to assess several DL architectures for PQD classification. To assess performance and efficiency, the standalone Convolutional Neural Networks (CNNs) and CNN–recurrent models are analyzed. With DenseNet-based architectures achieving the best performance, the analyzed models exhibit high, closely spaced classification accuracies, typically ranging from 98% to 99.4%. The study underlines the importance of evaluating both computational performance and efficiency, which also shows that such slight accuracy gains often come at the expense of significantly greater computational complexity. In addition to classification accuracy, this study also highlights model explainability. Gradient-weighted Class Activation Mapping (Grad-CAM) is used to show discriminative regions in FSST representations, which sheds light on the DL models’ decision-making process.

Electric Arc Furnaces (EAFs) are vital to sustainable steelmaking by enabling scrap metal recycling and reducing raw material demand. However, their high energy usage typically 400–470 kWh per ton poses economic and environmental challenges. This study proposes a machine learning framework to predict energy consumption through model-level optimization in EAFs using real data from a 120 MVA DSP-120 furnace. The dataset includes variables like transformer tap stages, oxidation rates, oxygen/gas injection rates, and post-process chemical compositions, all mapped to unique Heat IDs. After preprocessing and standardization, five regression models Random Forest, CatBoost, XGBoost, Bi-LSTM, and ANN were developed and evaluated. Random Forest hyperparameters were optimized using a Bayesian search strategy to improve model robustness and forecasting accuracy. To improve robustness, the top three models were combined through ensemble averaging. The final model achieved an MAE of 0.2533 MW, RMSE of 0.5858 MW, R2 of 0.9918, SMAPE of 48.51%, and 99% of predictions within ±5 MW, clearly demonstrating the industrial reliability of the proposed BO-RF approach. SHAP explainability identified key drivers of energy use, supporting interpretability and operator trust. This optimized ensemble approach offers both high predictive accuracy and a industry ready solution for intelligent energy control in modern steel production.