Thermal power forecasting of solar power tower system by combining mechanism modeling and deep learning method

Abstract

Concentrating solar power systems are proliferating around the world. However, the non-schedulable characteristic of thermal output presents considerable challenges to safe utilization of solar heat. So, seasonable forecasting of thermal output is going to be essential to both power grid and plant. In this paper, a hybrid forecasting method based on mechanism modeling and deep learning method is proposed. By mechanism modeling, the major meteorological factors used for forecasting are identified accurately, avoiding the subjectivity of chosen of model input features. By convolutional neural network and long short-term memory network, the input features are reconstructed, and the spatial-temporal coupling characteristics between meteorological factors are fully explored. Finally, the thermal power is output by fully connected layers. As a case study, the thermal power of a Solar Two-like concentrating solar power plant based on meteorological conditions of Zhangbei is forecasted, and the mean values of root mean square error and mean absolute error are 5.061 MWt and 2.871 MWt on the whole year, respectively. The results show that the methods proposed in paper can be effectively used for thermal power forecasting of concentrating solar power system.