Direct normal irradiance forecasting for high-temperature concentrated solar thermal systems

Concentrated solar thermal (CST) systems, reliant on the beam component of solar radiation, experience operational challenges during abrupt changes in the direct normal irradiance (DNI) caused by cloud cover. These fluctuations in DNI can induce thermal stress and compromise the system efficiency. However, limited research has focused on forecasting DNI within a timeframe suitable for addressing these operational challenges. Accurate short-term DNI forecasts can help mitigate the effects of DNI fluctuations by making timely adjustments to the CST system. Therefore, the objective of this study is to develop a forecasting algorithm for short-term fluctuations in DNI. To support this objective, we devised a new experimental CST setup that incorporates inexpensive camera components mounted on a solar tracker. A small amount of black tape was adhered to the camera lens, obscuring direct sunlight from the image sensor of the camera. This configuration enabled the capture of granular details in the circumsolar region and reduced the glare without obstructing a significant portion of the image. We evaluated various algorithms using these sky images to forecast DNI, with a forecast horizon of 20 s. We observed that models combining convolutional neural networks and recurrent neural networks were the most accurate, achieving the highest forecasting skill score (25.5%) of all implemented models. Considering the practical application of this forecasting algorithm, we analyzed a thermal storage setup consisting of a CST system and thermophotovoltaic cells over a four-minute intermittently cloudy period and found that the forecasting algorithm can increase power generation by 10.8%. The findings provide fundamental insights into the operation of CST technologies at elevated temperatures.