Improved building energy simulations and verifications by regression

Sammanfattning: It is common with significant differences between calculated and actual energy use in the building sector. These calculations are often performed with whole building energy simulation (BES) programs. In this process the analyst must make several assumptions about the studied building and its users. These calculations are often verified with measured data through the EUI benchmark indicator which is calculated by normalizing the annual energy use (from the grid) with the floor area. Due to the highly aggregated nature of the EUI indicator it is problematic to use this indicator to deduce erroneous assumptions in the calculations. Consequently, the learning process is often troublesome.Against this background, the main aim of this thesis has been to develop methods that can provide feedback (key building performance parameters) from measured data which can be used to increase simulation accuracy and verify building performance. For the latter, regression models have been widely used in the past for verifying energy use. This thesis has the focus on the use of regression analysis for accurate parameter identification to be used to increase the agreement between BES predictions and actual outcome. For this, a BES calibration method based on input from regressed parameters has been developed which has shown promising features in terms of accurate predictions and user friendliness. The calibration method is based on input from regressed estimations of air-to-air-transmission losses, including air leakage (heat loss factor) and ground heat loss. Since it is known that bias models still can give accurate predictions, these parameters have been evaluated in terms of robustness and agreement with independent calculations. In addition, a method has been developed to suppress the bias introduced in the regression due to solar gain. Finally, the importance of calibrated simulations was investigated.The regressed parameters were found to be robust with yearly variations in the heat loss factor of less than 2%. The regressed estimates of ground heat loss were also in good agreement with independent calculations. The robustness of the heat loss factor based on data from periods of substantial solar gain was also found to be high, with an average absolute deviation of 4.0%. The benefit with calibrated models was mainly found to be increased accuracy in predictions and parameters in absolute terms. With increased access to measured data and the promising results in this thesis it is believed that the presented regression models will have their place in future energy quantification methods for accessing energy performance of buildings.