For better building control, and for buildings to be better integrated with the grid operation, high fidelity building energy forecasting model that can be used for short-term and real-time operation is in great need. With the wide adoption of building automation system (BAS) and Internet of things (IoT), massive measurements from sensors and other sources are continuously collected which provide data on equipment and building operations. This provides a great opportunity for data-driven building energy modeling. However, the performance of data-driven based methods is heavily dependent on the quality and coverage of data. The collected operation data are often constrained to limited applicability (or termed as “bias” in this paper) because most of the building operation data are generated under limited operational modes, weather conditions, and very limited setpoints. The fact impedes the development of data-driven forecasting model as well as model-based control in buildings. The proposed framework of active learning in short-term data-driven building energy modeling aims to choose or generate informative training data, either to defy data bias or to reduce labeling cost. In the framework, a disturbance categorization is applied to divide the disturbance space into several categories. Then, in each disturbance category, independently apply active learning strategy to decide the controllable inputs in the current time step. In this way, the variations of controllable inputs and disturbances are both considered. In the case study, A virtual DOE reference office building with large-size and simulated in EnergyPlus environment is used as the testbed. A group of hierarchical setpoints, including zone temperature setpoint, supply air temperature and static pressure setpoints and chiller leaving water temperature setpoint, are the controllable inputs in this study. Regression tree is used as disturbance categorization algorithm and estimated error reduction is used as active learning algorithm. Improved model accuracy (lower testing error) is observed in the model trained by data from proposed framework, compared with models trained by normal operation data.
展开▼
