The analysis of voltage collapse induced by nonlinear loads in an arc furnace utilising deep learning-driven TabNet and NODE models

Electric arc furnaces (EAF) represent a class of nonlinear loads distinguished by high energy consumption during metal melting processes. Voltage collapses in these systems adversely affect power quality, reduce energy efficiency, and cause significant disruptions in production processes. Consequently, this study investigates the feasibility of deep learning-based approaches for forecasting voltage collapse events induced by the dynamic and nonlinear loads generated by electric arc furnaces. The analysis employs methods developed using the Tabular Network (TabNet) and Neural Oblivious Decision Ensembles (NODE) models to assess the characteristic variations of arc furnaces and their impacts on power systems through both experimental and simulation data. The characteristic behavior of the electric arc was modeled using an exponential-hyperbolic function validated by real-time data, while the simulation model was established in the MATLAB/Simulink environment to identify voltage collapse events. Critical features such as I1 (current) and V1 (voltage) were found to play a decisive role in predicting voltage collapse, and the decision mechanisms of the models were elucidated in detail using Explainable Artificial Intelligence (XAI) techniques such as SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME). The obtained results indicate that both models achieve an accuracy rate of approximately 95% with balanced classification performance, thereby offering a reliable approach for the early diagnosis of voltage fluctuations and collapse events in power systems. Furthermore, the findings contribute to strategic decision support systems aimed at enhancing safety and energy efficiency in industrial applications and offer a novel perspective on the potential of deep learning models in complex systems.