Enhancing Long-and-Short-Term Forecasting for Optimized Microgrid Energy Management Through Advance Hybrid Deep Learning Models

Solar power, while abundant, is unpredictable due to its intermittency, causing instability in energy supply and imbalances between supply and demand, which can threaten grid reliability. To address these challenges, innovative solutions are required, especially as electricity demand fluctuates. Energy Storage Systems (ESS) help manage peak shaving and load shifting, but erratic energy sources can degrade batteries, leading to high costs. Accurate forecasts of energy consumption (EC) and solar energy generation (EG) are crucial for optimizing solar microgrids. This study evaluates deep learning models, including Convolutional Neural Networks (CNN), Gated Recurrent Units (GRU), and a hybrid CNN-GRU model, to predict both EC and EG. A multi-input, parallel processing approach was used to capture temporal and spatial patterns for real-time applications with reduced data drift and improved accuracy. The model was tested using 47 months of historical data from the Sceaux microgrid near Paris, France, spanning from December 2006 to November 2010, from the University of California, Irvine repository. The data was also used to optimize the photovoltaic (PV) system sizing. The proposed method achieved excellent results for EG prediction with a Mean Absolute Error (MAE) of 3.974 and a Root Mean Squared Error (RMSE) of 6.603. For EC prediction, it obtained an MAE of 4.869, an RMSE of 6.527, and a Mean Absolute Percentage Error (MAPE) of 0.113, demonstrating its effectiveness for both short-term and long-term forecasting.