Predicting indoor air temperatures by calibrating building thermal model with coupled airflow networks

Abstract

Building models that can accurately predict hourly indoor air temperatures in free-running situations are key to understanding overheating conditions and the resilience of passive cooling strategies under a changing climate. To accurately predict indoor temperatures it is necessary to properly model pressure-driven infiltration and natural ventilation. This can be achieved by coupling a building thermal model to an airflow network model. In this paper, the development of coupled building thermal and airflow network models is described to calibrate building models using field measurements of indoor air temperature. Building models of three types of buildings were configured: long-term care building, primary school and multi-unit social housing. The building models were developed in Design BuilderTM and exported for use in an EnergyPlus simulation package. From information obtained from building surveys, site visits and architecture drawings, building parameters and operation schedules were collected. The unknown parameters, which included envelope thermal properties, shading devices, internal heat gains, envelope air leakage, window and door openings, were then calibrated based on measured values of indoor temperature. Reasonable ranges in value of the unknown parameters were first retrieved from applicable building construction practice documents and building energy standards. Two rounds of calibration were conducted through parametric simulations using the Monte-Carlo sampling method. A sensitivity analysis was also conducted for ranking the importance of all building parameters. The values for indoor air temperature as obtained through simulation were compared with measurements and the RMSE (root mean square error) was calculated for all values. The parameter value combinations corresponding to the minimum RMSE were adopted for the building models. The calibration process ended when the value for RMSE was <1.5℃. Results showed that the detailed building model was capable of predicting room air temperatures with minimum error levels (0.56℃ ≤ RMSE ≤ 1.50 °C) within the limits of applicable building model calibration standards (MBE±10%, CVRMSE<20%).