Worst-Case Guarantees for DNN Based DCOPF Under High RES Penetration

Addressing DC Optimal Power Flow (DCOPF) challenges in power systems with a high penetration of renewable energy sources (RES) necessitates innovative solutions due to the limitations of data accessibility, dynamic network adaptability, and worst-case scenario planning, including load shedding and RES curtailment. Traditional datasets for neural network training often fall short in quality and relevance, impeding the accurate modeling of machine learning solutions for energy distribution. This study introduces a robust deep neural network (DNN) model tailored to effectively manage high levels of RES integration by solving the DCOPF problem. The proposed model presents a model for simulating DCOPF without running the DCOPF equations. The reliability of the model's predictions is enhanced by minimizing the maximum deviation of the predicted DCOPF results from the actual results. Utilizing a dataset derived from the DCOPF configurations of a multi-bus power grid, the model's efficacy is demonstrated through comprehensive evaluations of its predictions against real-world data, showcasing superior performance over conventional neural network approaches. The proposed model's effectiveness is demonstrated across multiple bus systems, including Garver's 6, IEEE 9, 14, and 30 bus systems. Although the simulation was limited to the IEEE 30-bus system, the model has shown promising results that could be extended to larger power systems. The proposed model outperforms the grey box neural network (GBNN) and Projection-Aware DNN (PA-DNN).