Paper Conference

Proceedings of Building Simulation 2017: 15th Conference of IBPSA

     

Efficiency and Reliability of Bayesian Calibration of Energy Supply System Models

Kathrin Menberg1,2, Yeonsook Heo2, Ruchi Choudhary1
1University of Cambridge, Department of Engineering, Cambridge, UK
2University of Cambridge, Department of Architecture, Cambridge, UK


DOI: https://doi.org/10.26868/25222708.2017.315
Abstract: In this study, we examine the efficiency and reliability of a Bayesian calibration setup using temperature point measurements. Hamiltonian Monte Carlo sampling is found to be significantly more efficient with regard to convergence of the posterior distributions, which is assessed using different visual and quantitative measures. The examination of posterior realizations from different data sets and different prior distributions reveals that inference about model parameters is in general quite reliable, while learning about the magnitude of different error terms, such as model discrepancy and random errors, proves to be more difficult. Finally, predictive simulation results based on these inferred posterior distributions are generally in good agreement with measured data.
Pages: 1212 - 1221
Paper:
BS2017_315