Real-time heliostat field aiming strategy generation for varying cloud shadowing using deep learning.

Tower solar power generation technology has been proved to be one of the most promising electricity generation alternatives with renewable resources. A reasonable aiming strategy is essential for the efficient operation of tower solar plant. Meanwhile, the real-time calculation of focusing strategy is strictly required for the changing sun angle and possible local cloud occlusion at the front end and the demand of working medium at the back end. However, It is difficult to solve the optimization problem in real time because of the high complexity of calculation. In our work, a supervised learning scheme based on data optimized by genetic algorithm (GA) is proposed. We try to use the idea of image translation to learn the aiming strategy of thousands of mirrors under different conditions. In a case study of a heliostat sector, our deep learning model can give an aiming strategy with an effect close to that of genetic algorithm in a twinkling. [ABSTRACT FROM AUTHOR]