Scalable identification and control of residential heat pumps: A minimal hardware approach

Abstract While model predictive control (MPC) is a widely studied tool for integrating residential heat pumps with the smart grid, modeling difficulties and hardware requirements are significant barriers to adoption. In response, we present a nonintrusive plug-and-play methodology for model identification and model predictive control using only two increasingly popular smart-home devices: a smart thermostat and smart electricity meter, which makes the overall approach highly scalable. However, this minimal hardware approach requires new methods to overcome modeling challenges due to the lack of data diversity. A data-driven heat pump power model is identified without the need for submetering by using energy disaggregation. By incorporating the heat pump control data from the smart thermostat, we provide implicit load classification data to the model, substantially simplifying the signal decomposition task. Building and heat pump model parameters are then identified by a novel algorithm that combats overfitting caused by limited state excitation. Finally, we show the advantage of this scalable approach in an aggregate model predictive controller enabled by the communication ability provided by the smart thermostat. Results show that this aggregate control approach can provide improved grid services, such as reduced peak demand, while at the same time reducing energy consumption and improving thermal comfort. The proposed method has the potential to greatly lower the barrier for residential energy aggregation.