A model of a building's thermal dynamics is needed for prediction-based control. The task of identifying a thermal dynamic model is made challenging by the presence of large unmeasured disturbances, especially the heat gain due to the occupants. In fact, identification of this "occupant-induced load" is also valuable for predictive control-especially in commercial buildings. We propose a method to identify both a model (of resistance-capacitance network type) and the unmeasured disturbances from measured input-output data. The method is based on the insight that the main contributor to the unmeasured disturbance, the occupant-induced load, is piecewise constant, especially in commercial buildings. This can be used to construct an augmented dynamic model so that disturbance estimation is converted to a state estimation problem. An outer-loop optimization identifies the best-fit parameter values. The effectiveness of the method is evaluated using data from a simulation model (under both open and closed-loop operations) and a real building.

No abstract

We propose a control architecture for distributed coordination of a collection of on/off TCLs (thermostatically controlled loads), such as residential air conditioners, to provide the same service to the power grid as a large battery. This involves a collection of loads to coordinate their on/off decisions so that the aggregate power consumption profile tracks a grid-supplied reference. A key constraint is to maintain each consumer's quality of service (QoS). Recent works have proposed randomization at the loads. Thermostats at the loads are replaced by a randomized controller, and the grid broadcasts a scalar to all loads, which tunes the probability of turning on or off at each load depending on its state. In this paper we propose a modification of a previous design by Meyn and Bušić. The previous design by Meyn and Bušić ensures that the indoor temperature remains within a pre-specified bound, but other QoS metrics, especially the frequency of turning on and off was not limited. The controller we propose can be tuned to reduce the cycling rate of a TCL to any desired degree. The proposed design is compared against the design by Meyn and Bušić and another well cited design in the literature on control of TCL populations, by Mathieu et al. We show through simulations that the proposed controller is able to reduce the cycling of individual ACs compared to the previous designs with little loss in tracking of the grid-supplied reference signal.

No abstract

No abstract