Prediction of Direct Normal Irradiation Based on CNN-LSTM Model

Due to the intermittency and instability of power output of solar power stations (SPS), the safety and stationarity of the electric grid are facing considerable challenges. So, it is essential to predict the power output of SPS accurately. To date, as the most important meteorological factor determining the power output of SPS, the prediction of direct normal irradiation (DNI) has always been a hotspot, however at the same time a challenging problem. In recent years, with the improvement of hardware performance, deep learning methods have witnessed rapid development in the field of forecasting of renewable energies. In this paper, a hybrid method combining convolutional neural network (CNN) and longshort term memory (LSTM) neural network is proposed, the former can be effectively used for extraction of spatial feature, while the latter is used for extraction of temporal feature. As a case study, the proposed method is used for prediction of DNI of four typical days, i.e. spring equinox, summer solstice, autumn equinox, and winter solstice, based on the meteorological conditions of Zhangbei. The results show that deep learning methods can be effectively used for prediction of DNI, however, the proposed method in this paper achieves higher prediction accuracy than that of the compared methods.