Conference Paper
Proceedings of Building Simulation 2019: 16th Conference of IBPSA
A Framework to Quantify Data Informativeness in Risk-Conscious Building Performance Simulation Applications
A Framework to Quantify Data Informativeness in Risk-Conscious Building Performance Simulation Applications
Qi Li 1,2, Godfried Augenbroe 1, Jason Brown 11 School of Architecture, Georgia Institute of Technology, Atlanta, GA 30332, USA 2 Energy Systems Division, Argonne National Laboratory, Lemont, IL 60439, USADOI: https://doi.org/10.26868/25222708.2019.211110Abstract: This study proposes a framework to quantify data informativeness, i.e. the value of data in informing building performance management and retrofitting decisions, in building performance simulation applications under uncertainty. It comprises of the following elements: (1) initial quantification of baseline prediction uncertainties based on generic information, (2) data-driven model refinement through uncertainty propagation and inverse modeling, (3) prediction accuracy evaluated with probabilistic error metrics to reflect general data informativeness, and (4) explicit risk analysis to support specific decision-making. A case study of a hypothetical intervention analysis on a campus building section is presented to demonstrate the framework. Results suggest that inverse modeling with Bayesian inference are effective in constraining uncertainty and improving accuracy of model predictions under normalized mean probabilistic absolute error (NMPAE), a proposed probabilistic error metric. In addition, explicit risk analysis is feasible and necessary to represent more decision-specific data informativeness and guide data monitoring and collection strategies. Keywords: building performance simulation, data informativeness, uncertainty quantification, model calibration, decision makingPages: 4690 - 4697 Paper: