Calibration of a Multi-Residential Building Energy Model - Part II: Calibration Using Surrogate-Based Optimization

The present study deals with a calibration of a white-box Building Energy Model (BEM) by using an optimization technique. The studied building is a multi-residential building which consists of 41 apartments having from 1 to 3 bedrooms. The energy consumption data are collected by digitalmeters installed for each building apartment. As the calibrated energy model will be used for the purpose of energy load prediction as well as for fault detection and diagnosis (FDD) at building and apartment levels, the energy model considers many parameters. Even after a reduction step, i.e., selecting the most influent parameters by a sensitivity analysis (i.e. work presented in a separate paper: Part I - Cluster-Based Sensitivity Analysis), the total number of parameters for the optimization-based calibration is still large. Moreover, the white-box energy model simulation is quite expensive in terms of computational time. Therefore, the meta-modeling technique (e.g., Radial Basis Function network, Kriging model) is employed to reduce the computational cost of the white-box energy model simulation. The evolutionary algorithm, implemented within the Cenaero's in-house design space exploration and multi-disciplinary optimization platform, called Minamo, is used to figure out the global optimum of the constrained optimization problem.