Abstract This study proposes an AI-driven framework for designing and optimizing a grid-connected solar photovoltaic (PV)-battery system of hybrid energy tailored for educational campuses. The methodology begins with a comprehensive assessment of campus infrastructure, incorporating high-resolution data on electricity consumption, solar irradiance, weather patterns, and existing energy sources, including grid supply and diesel generators. Diffusion-Based Clustering is employed to identify distinct load profiles across functional zones such as academic buildings, hostels, laboratories, and administrative blocks. These insights facilitate the development of customized energy models reflecting real consumption behaviours. To enhance predictive accuracy, Advanced Dual Branch Graph Neural Networks (DBGNN) are utilized for spatial-temporal energy demand forecasting, while a Temporal Fusion Transformer (TFT) model performs multi-horizon solar irradiance prediction, accounting for seasonal and meteorological variability. The forecast outputs inform a Decision Transformer (DT) framework, which leverages reinforcement learning for intelligent, real-time energy dispatch optimization among PV systems, battery storage, grid power, and diesel generators. This integrated forecast-control pipeline is HOMER Pro software was used for modelling purposes, allowing for the evaluation of multiple hybrid system configurations. Key performance indicators including the use of alternative energy, carbon emissions, Net Present Cost (NPC), and Levelized Cost of Energy (COE) reduction are used to assess system viability. The proposed AI-based approach promotes energy autonomy, operational efficiency, and sustainability, offering a scalable and replicable model for smart campus energy systems.
Balita et al. (Mon,) studied this question.