Neural-network-driven dynamic simulation of parabolic trough solar fields for improved CSP plant operation

Abstract

Concentrating Solar Power plants face challenges in achieving and sustaining high performance levels partially due to complexities in plant operations. This study addresses these challenges by developing a computationally efficient, high-fidelity parabolic trough solar field model capable of emulating CSP plant dynamics for use as an operator training simulator and as a tool for optimizing operation strategies. Leveraging a neural network methodology, the model efficiently computes heat absorbed by heat transfer fluid in a solar field with various receiver conditions. The trained neural network model achieves heat absorption error of 0.3% compared to a detailed model while increasing the simulation speed by a factor of 100. The solar field model is validated with data from the operational Solana Solar Generating Station near Gila Bend, AZ (US), and computes temperatures resulting in a mean absolute error of $2.2[°C]$ over an entire day including start up and shut down. The model is further validated with respect to net optical efficiency that accounts for time-varying collector defocusing. Lastly, this work concludes with case studies that demonstrate the model’s capabilities both as the engine for a training simulator and as an tool for optimizing solar field control strategies.