Short-term Forecasting of Solar Radiation Using a Hybrid Model of CNN-LSTM Integrated with EEMD

Nowadays, renewable energy gradually become indispensable sources all over the world such as solar energy, wind energy, tidal energy, etc. In terms of solar energy, solar radiation fluctuates and depends on various other factors. Therefore, short-term forecasting of solar radiation plays a consistently important role in many fields of solar energy applications, especially in generating electricity. In this paper, we proposed a Convolutional Neural Network - Long-Short Term Memory (CNN-LSTM) network integrated with the Ensemble Empirical Mode Decomposition (EEMD) method to make a short-term forecast of solar irradiation in Vietnam. Firstly, we used EEMD method to separate the original irradiation series into intrinsic mode functions (IMFs). Secondly, each IMFs were fed into a predictive model that combined CNN and LSTM network and then composed into final forecasting of the solar irradiation. Finally, the results were compared with that of other methods such as the single model of CNN, LSTM and Bi-directional-LSTM to find out the benefits. The comparison illustrated that the performance of the proposed model was better than the others, namely the n-RMSE was 0.098 while that of LSTM, Bi-LSTM and CNN was 0.187, 0.169 and 0.177, respectively.