multi-level modelling, MPC, system identification, thermal energy storage
This paper presents results obtained by applying a multi-level methodology for the implementation of a model-predictive control (MPC) strategy in a large institutional building. The case study building, is a model of a two storey school building, with a floor area of 24,000 m2 (258,000 ft2) with 46 thermal zones. The zones considered include a large diversity of spaces: small offices, classrooms, long hallways and two gymnasia. A detailed thermal model of the building was created in EnergyPlus. The EnergyPlus was used to generate input and output data employed for a systematic system identification exercise, which resulted in a set of multi-input single-output (MISO) linear models. Three control levels were considered: a thermal zone level (46 models), â€œwingâ€ level (7 models) and a building level (one model). The models identified are state-space representations with order ranging between 4 and 12. This hierarchical, multi-level methodology enables the use of low-order models for each system under consideration: for example, a simple 9th order model at the building level can be used to predict its thermal load over a 48-h horizon, with a relatively coarse sampling time of 2 hours (24 samples). At the other extreme, a zone level model has a prediction horizon of 2 hours, and a much finer sampling time of 10 minutes (12 samples). For the MPC studies, a mechanical system considering thermal energy storage devices (ice bank + hot water tank) was considered in the calculations. An optimization routine was carried out to minimize the electricity cost, while maintaining comfortable conditions in the space: a time-of-use rate was employed in the definition of the objective function. The results presented in this paper illustrate how the multi-level concept discussed in this paper can be used to harmonize the performance of building control systems, from the supervisory BEMS to the local thermostat controllers.