A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Assessment of linear emulators in lightweight Bayesian calibration of dynamic building energy models for parameter estimation and performance prediction
AbstractCalibration of building energy models is widely used in building energy audits and retrofit practices. Li et al. (2015) proposed a lightweight approach for the Bayesian calibration of dynamic building energy models, which alleviate the computation issues by the use of a linear regression emulator. As a further extension, this paper has the following contributions. First, it provides a brief literature review that motivates the original work. Second, it explained the detailed calibration methodology and its mathematical formulas, and in addition the prediction using meta-models. Third, it introduced new performance metrics for evaluating predictive distributions under uncertainty. Fourth, it used the standard Bayesian calibration method as the benchmark, assessed the capability of regression emulators of different complexity, and showed the comparison result in a case study. Compared to the standard Gaussian process emulator, the linear regression emulator including main and interaction effects is much simpler both in interpretation and implementation, calibrations are performed much more quickly, and the calibration performances are similar. This indicates a capability to perform fast risk-conscious calibration for most current retrofit practice where only monthly consumption and demand data from utility bills are available.
Assessment of linear emulators in lightweight Bayesian calibration of dynamic building energy models for parameter estimation and performance prediction
AbstractCalibration of building energy models is widely used in building energy audits and retrofit practices. Li et al. (2015) proposed a lightweight approach for the Bayesian calibration of dynamic building energy models, which alleviate the computation issues by the use of a linear regression emulator. As a further extension, this paper has the following contributions. First, it provides a brief literature review that motivates the original work. Second, it explained the detailed calibration methodology and its mathematical formulas, and in addition the prediction using meta-models. Third, it introduced new performance metrics for evaluating predictive distributions under uncertainty. Fourth, it used the standard Bayesian calibration method as the benchmark, assessed the capability of regression emulators of different complexity, and showed the comparison result in a case study. Compared to the standard Gaussian process emulator, the linear regression emulator including main and interaction effects is much simpler both in interpretation and implementation, calibrations are performed much more quickly, and the calibration performances are similar. This indicates a capability to perform fast risk-conscious calibration for most current retrofit practice where only monthly consumption and demand data from utility bills are available.
Assessment of linear emulators in lightweight Bayesian calibration of dynamic building energy models for parameter estimation and performance prediction
Li, Qi (author) / Augenbroe, Godfried (author) / Brown, Jason (author)
Energy and Buildings ; 124 ; 194-202
2016-04-10
9 pages
Article (Journal)
Electronic Resource
English
Using Emulators to Evaluate the Performance of Building Energy Management Systems
| British Library Online Contents | 1994
Using Emulators to Evaluate the Performance of Building Energy Management Systems
| British Library Conference Proceedings | 1994