The HDGP-PSF: A Hybrid DRL-GNN Framework with Predictive Supply Forecasting for Solar Plant Integration into Power Systems

Abstract

The exponential growth of renewable energy resources, particularly solar photovoltaics (PV), is transforming the operational landscape of modern power systems. However, this transition introduces challenges related to variability, intermittency, and real-time control. To address these issues, this paper presents HDGP-PSF, a Hybrid Deep Graph Processing framework that integrates Deep Reinforcement Learning (DRL), Graph Neural Networks (GNN), and Predictive Supply Forecasting (PSF). Building on recent advances in renewable energy modeling and graph-based power flow forecasting, the framework adopts a data-driven, topology-aware approach to dispatch optimization. It is validated using real data from the 132/11 kV Peshawar Grid Station and NREL solar datasets. Results show that HDGP-PSF improves operational performance compared to traditional strategies, achieving a 14% reduction in carbon emissions while enhancing grid stability. The LSTM-based PSF model achieved an RMSE of 0.25 MW and a MAPE of 21.93%, demonstrating strong forecasting capability. These findings highlight the effectiveness of combining DRL, GNN, and predictive forecasting in building smarter, greener, and more resilient power systems.