Reduced-Order-Model to Optimize the Thermo-Hydraulic Performance of a Cavity Receiver

Abstract

The overall performance of central receiver CSP plants highly depends on the design of the solar field and the receiver. The high computational cost of Computational Fluid Dynamics (CFD), Finite Element Analysis (FEA) and raytracing simulations to evaluate the performance of these systems limits the feasibility of using such detailed simulations in coupled optimization schemes. As a result, coupled optimization approaches commonly use correlations and heuristic optimization methods instead of detailed, high-fidelity simulations for quick performance evaluations to enable an optimization loop and iteratively optimize the solar field and the receiver towards a coupled optimum. Although such correlations enable quick performance evaluations, they are often not very precise and only valid for a specific range of boundary conditions. Therefore, this study proposes the integration of Reduced-Order-Models (ROM) based on a limited set of high-fidelity training data points into the coupled optimization framework as an alternative to empirical correlations for receiver performance predictions. In this work, convective and radiative ROMs of a theoretical central tower cavity receiver have been successfully built, and a validation study showed very good estimation performance of 1,19% and 4,47% using the original data tensor consisting of 243 training data points. Theoretically reducing the training data set to 32 resulted in a noticeable increase of estimation error to 5,39% and 9,1%, but also a significant reduction in required training points. Once the ROM has been built, the performance evaluation of a single design point requires approximately t_ROM,i ≈ 0,06s, which is comparable to the time typically required using empirical correlations. Overall, the results demonstrate that data-driven ROMs provide a promising intermediate modeling approach between purely empirical correlations and fully resolved high-fidelity simulations.