Development and performance assessment of a new opensource Bayesian inference R platform for building energy model...

Area of Science:

• Building Science
• Computational Modeling
• Energy Systems Analysis

Background:

• Building energy consumption models face inherent uncertainties, necessitating robust calibration and sensitivity analysis.
• Existing deterministic calibration methods often lack quantified uncertainties and rely on user experience for parameter selection.
• A rigorous, automated approach is needed to improve the reliability of building energy models.

Purpose of the Study:

• To develop an automated Bayesian Inference calibration platform for building energy models.
• To integrate sensitivity analysis and Bayesian inference for parameter determination and uncertainty quantification.
• To enhance computational efficiency using a meta-model for the Markov Chain Monte Carlo process.

Main Methods:

• Developed an R package for automated Bayesian Inference calibration.
• Implemented sensitivity analysis to identify key calibration parameters.
• Utilized Bayesian inference to quantify uncertainties in model parameters.
• Employed a meta-model to accelerate the Markov Chain Monte Carlo (MCMC) simulations.

Main Results:

• Successfully demonstrated the platform on synthetic and real high-rise buildings in a hot and arid climate.
• The platform effectively determines calibration parameters and quantifies associated uncertainties.
• The meta-model significantly reduces computational time for the calibration process.
• Established the relationship between calibration parameters, performance, and meta-model accuracy.

Conclusions:

• The developed Bayesian Inference platform offers significant advantages over existing methods for building energy model calibration.
• The platform provides reliable building energy performance estimations in reduced computation times.
• Automated calibration with uncertainty quantification enhances the trustworthiness of building energy models.