Neural Prophet driven day-ahead forecast of global horizontal irradiance for efficient micro-grid management

This study introduces an innovative approach to day-ahead solar irradiance forecasting, utilizing the NeuralProphet model—a deep learning-based extension of the Prophet tool—to effectively manage the complexities of time-series data in solar energy prediction. Recognizing the critical role of accurate solar irradiance predictions in optimizing the operation of multi-vectored energy hubs, this research integrates NeuralProphet's advanced neural network components, including its trend and seasonality modules, to enhance forecasting accuracy. The innovative integration of NeuralProphet's trend, seasonality, and autoregressive components allows for superior performance in forecasting compared to traditional models. When the model's performance is compared to historical solar irradiance data, it is evident how well it captures underlying trends in comparison to more conventional approaches. In contrast, Dataset 1 has a daily forecast MAE for the model that is about 38.6 % lower than Dataset 2, but Dataset 1 has weekly and monthly forecast MAEs that are 6.25 % and 5.6 % higher, respectively. Better day ahead accuracy is also shown by the daily forecast MAPE for Dataset 1 being 45.1 % lower than for Dataset 2. Furthermore, Dataset 1 has a daily R2 value of 99.5 %, while Dataset 2 has a value of 99.0 %. This suggests that Dataset 1 has 0.5 % more accurate day ahead forecasts. There is a 0.1 % increase in accuracy as evidenced by the weekly R2 values for Dataset 1, which is 98.4 %, while Dataset 2 is 98.3 %. The R2 for monthly projections shows that Dataset 1 has a 0.5 % poorer accuracy over longer time horizons, with 95.6 % for Dataset 1 and 96.1 % for Dataset 2. These results demonstrate the model's potential to optimize the operation of energy hubs by accurately forecasting GHI, contributing to more efficient micro-grid management and a reduction in dependency on fossil fuels. The findings demonstrate that deep learning techniques can be integrated into renewable energy forecasting, offering substantial benefits for the design and management of future energy systems..