Hybrid Deep Reinforcement Learning and Graph Neural Network Framework for Renewable Energy Integration in Power Systems (HDGP): A Case Study of Peshawar Grid Station

The increasing penetration of renewable energy, particularly solar photovoltaics (PV), introduces significant uncertainties into power system operations. Traditional Optimal Power Flow (OPF) methods often rely on detailed network models and are computationally expensive, while existing reinforcement
learning (RL) approaches frequently overlook physical constraints. This paper proposes a novel Hybrid Deep Graph Processing (HDGP) framework, integrating deep reinforcement learning (DRL) and graph neural networks (GNN) for optimal real-time integration of solar PV systems in power distribution networks. The framework leverages real-world data from the Peshawar Grid Station 132/11 kV network and incorporates time series solar generation profiles from the National Renewable Energy Laboratory (NREL), allowing accurate modeling of renewable power variability. Experimental results demonstrate a 14% reduction in carbon emissions and a 11% decrease in the Levelized Cost of Energy (LCOE) compared to the conventional
PESCO grid. The framework also ensures computational efficiency, with an average decision-making time of 4.03 ms per step on an Intel i5-8250U CPU @ 1.60GHz with 16GB RAM, significantly outperforming conventional OPF methods. These findings highlight the potential of HDGP to transform renewable
energy integration and smart grid management, enabling realtime, cost-effective, efficient, and environmentally sustainable power dispatch.