Addressing data informativeness in risk-conscious building performance simulation applications
MetadataShow full item record
Building performance management remains an important aspect in reducing building energy consumption and enhancing occupants’ thermal comfort and work productivity. Recent decades witnessed the maturity and proliferation of numerous methods, software and tools that span the whole spectrum of common building performance management practice. Among those related research and applications, the use of information and data in calibration and validation of building performance simulation (BPS) models constitutes an important subject of study especially in fault detection, operations management, and retrofit analysis. An extensive review of BPS model calibration and validation studies reveals two major research gaps. First, contemporary model calibration practice calls for an effective and robust method that can systematically incorporate a variety of information and data, handle modelling and prediction uncertainties, and maintain consistent model performance. Second, current approaches to collecting information and data in real practice largely depend on individual experience or common practice; further study is needed to understand the value of information and data, i.e. assess data informativeness, such as to support specific decision-making processes in choosing data monitoring strategies and to avoid missed opportunities or wasted resources. To this end, this dissertation develops a new framework to address data informativeness in model calibration and validation to answer two major research questions: 1) how to make optimal use of available information and data to calibrate a building simulation model under uncertainty, and 2) how to quantify the informativeness of information and data for risk-conscious building performance simulation applications. This framework builds upon uncertainty propagation using detailed measurements, and inverse modelling using Bayesian inference. It introduces probabilistic performance metrics to assess model prediction consistency and quantify data informativeness. Following an explanation of the framework’s theoretical soundness, this dissertation provides two case studies to demonstrate its practical effectiveness. The first is a controlled experiment in the Flexlab test facility at Lawrence Berkeley lab. A new validation methodology is proposed to validate a simulation model under uncertainty, in which the validation criteria build upon the introduced probabilistic performance metrics. Given the experiment setup, uncertainty propagation based on synthetic measurements is applied, which effectively improves prediction agreement and reduces the risk of accepting invalid simulation outcomes. The second is to determine the appropriate model form and metering data for a hypothetical intervention analysis of an existing building with hydronic heating on the Cambridge, UK campus. A three-level modelling method is proposed to enable modelling all thermal processes occurring in individual rooms while efficiently modelling the whole building to estimate heating system performance. Different sets of metering data are then used to calibrate the physical model, and the result indicates the superiority of Bayesian inference in exploiting the value of data, the necessity of room temperature and electricity monitoring under uncontrolled conditions, and the potential of daily metering data for calibration in real building performance management practice.