Solar Power Forecasting using EEMD followed by LSTM

To properly optimize and manage a smart power grid with high penetration of Solar Power (SP), the accurate prediction of the SP is extremely important. Due to the uncertain and volatile nature of SP, its prediction is very difficult, particularly when involving drastic changes in a short period of time. This paper proposes a decomposition of the data through Ensemble Empirical Mode Decomposition (EEMD) followed by different Long Short-Term Memory (LSTM) architectures to forecast SP with the additional capability to effectively capture the sudden changes in power. The proposed model was trained, validated, and tested with SP data available for three years in samples of 10 minutes. The proposed model performance is compared with the baseline LSTM model, the former showing superior performance mainly for the peak variations of SP than the latter.