Deep learning augmented medium-term photovoltaic energy forecasting: A coupled approach using PVGIS and numerical weather model data
Muhammad Ehtsham, Marianna Rotilio, Federica Cucchiella
Abstract
Integrating PV energy resources into energy grids is crucial for PV energy organizations, making medium- and short-term PV forecasts important. PV organizations look forward to modern tolls for efficient systems for the most beneficial operations for PV systems. This research proposes, applies, and assesses a modern machine learning and deep learning based short-term PV energy forecasting system. Numerical weather model-based data is utilized for real-time forecasting at an hourly scale for the next four days, an additional analysis is performed by leveraging the PVGIS data in addition to NWM data. The proposed methodology is developed and applied to more than 200 PV installations, both BIPVs and BAPVs. The system was able to produce effective PV energy forecasts with high accuracy and efficiency analysis of 3 different PV installations ranging 17 kWp, 91 kWp, and 386kWp are reported in this paper. The research concluded with the feasibility of the proposed systems and findings further support the efficacy of the proposed framework, which can be adopted by organizations seeking to optimize PV system performance and reliability. • Proposing DL-based systems for short-term PV energy forecasting. • Leveraging NWM and PVGIS data to enhance PV energy prediction accuracy. • Applying the methodology to 200 + BIPV and BAPV installations. • Evaluating ML, DL, and hybrid models for accurate PV energy forecasting outcomes. • RF LSTM hybrid model demonstrated robust performance, particularly with NWM data alone. • Optimizing PV system reliability and grid integration using advanced models.