Development, implementation, and impact analysis of model predictive control-based optimal precooling using smart home thermostats

Abstract

Model Predictive Control (MPC)-based precooling strategies are typically analyzed through simulations. Their field studies in residential buildings are quite limited due to hardware retrofitting costs, lengthy parameter identification periods, and the need to have homeowners' participation. Therefore, there is a lack of MPC-based precooling impact analysis using field data which may have prevented widespread adoption of the technology. To address this issue, in this paper we developed an MPC agent and implemented it on an online cloud-based platform. The MPC agent makes use of a gray-box model to capture the home thermal dynamics, a novel set of constraints to ensure that the model parameters identified are realistic, and a node sensor to provide interior wall surface temperature measurement. To evaluate its performance and conduct impact analysis, we introduced a modified matrix profile-based weather clustering algorithm and two performance indicators for cost saving and energy flexibility. We also carried out extensive field tests on nine homes over a period of four months. Our results show that the MPC agent can reduce energy cost by 28.72%–51.31% on hot summer days and by up to 60.32% on mild summer days, in addition to achieving significant energy flexibility. Moreover, the agent's performance is found to be most impacted by weather conditions, AC performance, user comfort preferences, and floor areas of the homes.