(345a) Data-Driven Models and Model Predictive Control Formulations for Load Leveling of Residential Buildings Conference: AIChE Annual MeetingYear: 2014Proceeding: 2014 AIChE Annual MeetingGroup: Computing and Systems Technology DivisionSession: CAST Division Plenary Time: Monday, November 17, 2014 - 10:35am-11:00am Authors: Baldea, M., The University of Texas at Austin Edgar, T. F., The University of Texas at Austin Perez, K. X., University of Iowa Buildings account for about 60% of the electricity consumption in the United States . While residential and commercial buildings have an approximately equal share, the impact of residential buildings on the grid is more pronounced, owing to the variability of their energy demand, which changes significantly during a day and between subsequent days. For example, the load placed on the grid by residential consumers is strongly influenced by weather and human activity patterns. Power producers and distributors deal with such demand fluctuations by bringing online additional, peaking, generation capacity, which typically consists of less efficient and more polluting power plants. Eliminating the use of peaking plants provides a strong incentive for mitigating variability in residential energy consumption. The most significant residential energy consumer is the heating, ventilation and air conditioning (HVAC) system. In this paper, we investigate the potential for coordinated control of the HVAC systems in a residential community, with the purpose of achieving reductions in peak electricity demand, i.e. “leveling the residential load”. Our work is based on experimental data obtained from a unique complement of 41 homes situated in Austin, Texas, for which power demand is sampled at high frequency using smart meters. Additionally, data concerning residential comfort preferences are collected using smart thermostats. We present a scheme for developing control-relevant, reduced-order models (ROMs) for each home, that tie energy consumption to outdoor dry bulb temperature and thermostat set points. A non-intrusive load monitoring technique is used to disaggregate A/C electricity consumption from whole-house electricity data reported by smart meters. We show that the proposed time-series models can use thermostat set points and outdoor dry bulb temperatures to accurately predict A/C loads. We use the ROMs to construct a centralized model predictive control (MPC) scheme that manages energy use at the neighborhood level by altering the thermostat setpoints in individual homes. Simulation results confirm that the proposed controller can reduce the peak electricity load by pre-cooling homes and staggering the time A/C units turn on. Our approach is predicated on leveraging the thermal mass of each house and shifting the electricity peak demand of each individual house to a different time in order to level the overall load. On average the centralized MPC is able to reduce the peak load by 27% (15 kW) for the group of 41 houses without substantial increase in overall electricity consumption. We also show through simulations that decentralized control (whereby each house attempts to level its own peak) is less successful at leveling the overall load. References  U.S. Energy Information Administration, Annual energy review 2011 (2012).