How can machine learning help in developing very accurate prediction forecasts in solar thermal power plants? Read in this week’s Helmholtz AI project showcase how researchers at the German Aerospace Center (DLR) and Forschungszentrum Jülich (FZJ) are working on this challenge.
Could you introduce yourself, giving your affiliation, area of work, and of course, the project title?
My name is Dr. Daniel Maldonado Quinto. I work at the German Aerospace Center in the Institute of Solar Research. I lead the group "Simulation and Digitalization" and we investigate modern algorithms to make solar thermal power plants of the future more efficient and competitive. Our collaboration partner from FZJ provides additional and crucial machine learning know how for generative models to use these data for a reliable prediction of the essential flux density maps, together with computational resources for conducting extensive training experiments. The GANCSTR: GAN-based calculation of concentrated radiation on solar tower receivers project helps us to adapt our numerical models to reality.
In simple words, what specifically is your project about? And, how and why do you think it is a high risk, high gain endeavour?
Concentrating solar power plants (CSP) play an important role in the world's transition to a sustainable energy supply. Solar tower systems offer a high potential for cost reduction due to the high temperature heat that is generated through highly concentrated solar radiation. In these systems, a large number of mirrors (heliostats) reflect the incoming sun radiation onto one point (the receiver). There, the concentrated radiation and especially its surface distribution are the key variables for a safe and optimal operation of the entire power plant. Their prediction and control are decisive for the levelized cost of energy (LCOE) and thus a key success factor for reaching competitiveness of CSPs with fossil fuel power plants. Up to now there are no measuring methods to determine the concentrated solar radiation distribution at large power plants. In addition, classical calculation methods using physical models are computationally expensive. The goal of this project is to predict an exact radiation distribution on the receiver using Generative Adversarial Networks (GANs) networks. As the available data for each heliostat is small, we employ advanced transfer learning techniques. The small amount of data poses a major risk to success. However, if we achieve our goal, it can give solar power plants a huge boost and accelerate the energy transition.