Short-term multi-step forecasting of rooftop solar power generation using a combined data decomposition and deep learning model of EEMD-GRU

Abstract

In this study, an integrated forecasting model was developed by combining the ensemble empirical mode decomposition (EEMD) model and gated recurrent unit (GRU) neural network to accurately predict the rooftop solar power output at a specific power unit located in Tay Ninh province, Vietnam. The EEMD method was employed to decompose the solar power signals into multiple frequencies, allowing for a more comprehensive analysis. Subsequently, the GRU network, known for its ability to capture long-term dependencies, was utilized to forecast future values for each decomposition series. By merging the forecasted values obtained from the decomposition series, the final prediction for the solar power output was generated. To evaluate the efficacy of our proposed approach, a comparative analysis was undertaken against other forecasting models, including a single artificial neural network, long short-term memory network, and GRU, all of which solely considered the solar power series as input features. The experimental results provided compelling evidence of the superior performance of the EEMD-GRU model, especially when incorporating weather variables into the forecasting process, achieving the best results in all three forecasting scenarios (1-step, 2-step, and 3-step). For both forecasting targets, Inverter 155 and 156, the n-RMSE indices were 1.35%, 3.5%, and 4.8%, respectively, significantly lower than the compared single models. This integration of weather variables enhances the model's accuracy and reliability in predicting rooftop solar power output, establishing it as a valuable tool for efficient energy management in the region.