Hierarchical-stochastic model predictive control for a grid-interactive multi-zone residential building with distributed energy resources

Abstract

Listen

Translate article

Using Model Predictive Control (MPC) is a promising method for enabling grid-interactive efficient buildings. Since MPC relies on a building model and the forecasts of external disturbances to derive optimal inputs, the uncertainties due to forecast errors and model inaccuracies can deteriorate the control performance. Most existing MPC studies for the built environment are deterministic MPC without consideration of uncertainties even though several different methods (e.g., stochastic MPC) are available in the literature to deal with them. Even studies that consider forecast uncertainties often neglect model inaccuracies. Hence, in the present paper, a novel hierarchical-stochastic MPC is proposed considering forecast uncertainties and model inaccuracies, and its performance is compared with deterministic, stochastic, and hierarchical MPC for power management in a residential building with distributed energy resources. The control objective is the cost-optimal scheduling of a heat pump, a battery for energy storage, and a rooftop photovoltaic system. Measurement data is used to identify the building model. The results for one-week simulation in winter show that (1) the deterministic MPC results in an unacceptable level of temperature constraint violations in two out of five rooms; (2) the hierarchical MPC can reduce the temperature constraint violations to an acceptable level at the expense of increased cost; (3) the stochastic MPC achieves the same reduction in temperature constraint violations as the hierarchical MPC but at slightly lower costs; and (4) the new proposed hierarchical-stochastic MPC results in both lower temperature constraint violations and lower financial expenses than the use of stochastic or hierarchical MPC individually.