A Hybrid VMD-GRU Model for Enhanced Photovoltaic Power Forecasting

Accurate photovoltaic (PV) power forecasting is crucial to grid stability, energy market operations, and storage optimization. However, the inherently non-linear and variable nature of PV power generation poses significant challenges to traditional forecasting methods. In this paper, we propose a hybrid model that combines variational mode decomposition (VMD) and gated recurring unit (GRU) networks to address these challenges. VMD decomposes the PV power data into intrinsic mode functions (IMFs) to isolate key frequency components, while GRU networks are utilized to model temporal dependencies for each IMF. The proposed VMD-GRU model was evaluated in the short, medium and long-term forecast horizons using realworld photovoltaic power data. Experimental results demonstrated that the hybrid model significantly outperformed the standalone GRU and Long Short-Term Memory (LSTM) models, achieving reductions in Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) and a R2 value exceeding 99% across all horizons. These results highlight the effectiveness of combining signal decomposition with advanced deep learning techniques to improve forecast accuracy. This study contributes to the development of robust and precise PV power forecasting models, which are essential for effective grid management and the integration of renewable energy into modern power systems.