Neural network model from Wisconsin University enhances CSP simulation

Researchers at the University of Wisconsin-Madison's Department of Mechanical Engineering have developed a neural network-based model to improve the operation of Concentrating Solar Power (CSP) plants. The model is efficiently simulated solar field dynamics for operator training and optimizing control strategies. It also calculated heat absorption with a 0.3% error compared to a detailed model while running simulations 100 times faster. Validation using data from the Solana Solar Generating Station in Gila Bend, AZ has shown accurate temperature predictions throughout the day, including startup and shutdown. The model also accounted for changes in net optical efficiency due to collector defocusing. Case studies demonstrated its effectiveness as both a training tool and a strategy optimization resource. The research was published in TIB Open Publishing.