Benchmark of optimization techniques for identification of buildings thermal parameters

Overview & Synopsis

This article presents an ongoing work aiming at the development of an optimization tool for the assessment of building intrinsic performances. The objective is to enable for the reliable calculation of as-built envelope thermal parameters (resistance and capacitance) based on measurements collected over a limited period of time. The tool is based on the combination of a basic physical model and of optimization algorithms that automatically calibrate the model from measures. This paper presents a benchmarking study lead as part of this work to select an inverse optimization methodology, which is intended to identify a set of thermal parameters. Three methods have been tested for the building thermal inversion. The first method is based on a simple greedy resolution (Particle Swarm Optimization) whereas the other two are based on substitution model (Support Vector Regression and Metamodels). The study has been validated on a basic use case (monozonal building) and relied on a comparison between the predictions obtained from calibrated models and those obtained from the Energy Plus simulation environment. The study shows that metamodels coupled with a crossvalidation method (kriging) lead to the best results.