Buildings account for significant amount of final energy consumption and therefore there is an intensive research aimed at its optimization. Predictive control has become a very popular approach in many industries buildings included. The main bottleneck of this method is a need for a good model. There are many different identification frameworks, plenty of methods and approaches, some of them more or less suitable for the building modeling. A common situation is that there are number of models at hand (often of different complexities), and there is a need for selection of the “best” model for predictive control. A logical choice is to start testing the statistical significance of an additional complexity (in a sense of the structure and number of disturbance inputs) of the model. This paper proposes a systematic way of building-up the model, starting from a simple structure. Then, more complex models are considered in an iterative manner. In each iteration, the statistical significance of the additional information due to the more complex model is checked. The procedure stops when selecting more complex models brings no quality improvements. In this paper, a semi-physical modeling using CTSM1 and model selection based on statistical tests are presented. Finally, the properties of the proposed algorithm are investigated.
展开▼
