Abstract
It is well known that calibration of building energy models is an under-determined problem, whether subjected to hourly or monthly calibration criteria. In fact, while it is possible to identify a large number of calibrated models, it is not clear which ones offer a good representation of the building behaviour. For calibration methodology of building energy models to be effective it should automate and speed-up calibration process, which is especially important when the number of model parameters is too large to tune manually. Moreover, when the number of model parameters is too large, the probability to find the real parameter combination using statistical sampling methods is very small. Instead we suggest performing a guided search of the parameter space, e.g. solving a parameter optimization problem. Since Tikhonov-type regularization has been applied successfully to many ill-posed inverse problems, we propose adopting the same methodology to find optimal parameters of building energy models. Regularization term can be interpreted as imposing certain a-priori distributions on model parameters as identified from the energy audit. As an illustration, a case study residential apartment is calibrated and it is shown that regularization more accurately predicts the energy demand estimate after the retrofit for the case study.