Influence of Data Pre-Processing Techniques and Data Quality for Low-Order Stochastic Grey-Box Models of Residential Buildings

Abstract

Model Predictive Control (MPC) has proved to be a key technology to activate the energy flexibility of buildings. A reliable control-based model should be developed to implement an efficient optimal control. Grey-box models, as a combination of physical knowledge and experiment data, have been widely used in the literature. However, in the identification process of grey-box models, many factors affect the results. This paper uses data from virtual experiments in IDA-ICE to investigate the influence of the optimization methods, the filtering methods, the training dataset and the sampling time interval on stochastic grey-box models. It shows that global optimization increases the chance to avoid a local minimum. Pre-filtering methods have a small influence on the model quality. Larger data sampling time will cause the identified parameters to become non-physical. However, the simulation performance of the model is kept almost unchanged.