An Efficient Global Optimization Scheme for Building Energy Simulation Based on Linear Radial Basis Function
Efficient Global Optimization, Building Simulation, Surrogate Model, High Performance Buildings
Motivation: The building performance optimization is considerably increasing since the design goals are moving from the solely energy saving target to the optimization of the overall performances, cost and sustainability objectives. The evolutionary algorithms coupled with building simulation codes are often used in academic researches, however, they are limited applied in actual building design. Indeed, the high number of expensive simulation runs required by evolutionary algorithms strongly limits their suitability for the professional practice. For this reason, an efficient optimization scheme is essential for the diffusion of the building performance optimization tools outside the academic world. What was done: The research focuses on the development of an efficient global optimization scheme (EGO) based on a radial basis function network (RBFN) meta-model to emulate the expensive function evaluation by means of the building energy simulation. In this surrogate model, each cost function can be approximated by a linear combination of unknown coefficient multiplied by a set of linear radial-basis function. In the proposed method, the surrogate model is firstly used in the evolutionary algorithm code to find the optimal solutions. Then, the actual fitness functions are evaluated for the optimal points by means of building simulation and the surrogate model is then update. These steps are continued until the convergence criterion is met. This efficient optimization scheme has been implemented in Matlab and verified on some test cases. The test bed of the method is the optimal building refurbishment of three simplified existing buildings, for which the optimal solutions have been also calculated by using the brute force approach. Finally, the EGO performances were also compared with those offered by the popular Non Sorting Genetic Algorithm (NSGA-II). Expected benefits of what was done: The results of this research show how the EGO algorithm is able to find a large number of optimal solutions with a reduced number of expensive simulation runs. This makes it possible to apply the algorithm to the optimization of building projects that use expensive simulation codes such as lighting models, CFD codes or coupled dynamic simulation of building and HVAC systems.