Surrogate modeling for the fast optimization of energy systems.

Abstract: While software simulations in the building technology field are essential for designing efficient systems, for specific applications, the tools currently available are computationally too expensive. This paper presents a novel approach for the fast computation of accurate simulation results and demonstrates its efficacy for solving a multiobjective optimization problem. A surrogate modeling approach is proposed, which is based on an effective interpolation algorithm and an optimal distribution of sampling nodes. The highest accuracy is shown to be achieved by a Poisson disk node distribution and an interpolation function developed from a cubic radial basis function. Furthermore, since global interpolation algorithms in high-dimensional spaces are fundamentally inefficient, a local adaptation is developed. The fidelity of the surrogate model is shown to generalize to higher dimensions with respect to the average number of nodes per dimension η s . The application potential is tested on a case study consisting of a multiobjective design optimization problem. The results show that the surrogate and the fine model converge to the same solution. Being 150 times faster than the original model, the surrogate model is thus a valid choice for computationally demanding applications. [Copyright &y& Elsevier]