Meters to Models: Using Smart Meter Data to Predict Home Energy Use

Access to smart meter data in the United States presents an opportunity to better understand residential energy consumption and energy-related behaviors. Air-conditioning (A/C) use, in particular, is a highly variable and significant contributor to residential energy demand. Most current building simulation software tools require intricate detail and training to accurately model A/C use within an actual house. However, integrating existing modeling software and empirical data has the potential to create highly portable and accurate models. Reduced-order models (ROM) are low-dimensional approximations of more complex models that use only the most impactful variables. In this paper, we report on the development of ROMs for 41 physical houses in Austin, Texas, using smart meter data. These models require outdoor dry bulb temperature, thermostat set points and A/C energy use data to regress model coefficients. A nonintrusive load monitoring technique is used to disaggregate A/C electricity consumption from whole-house electricity data reported by smart meters. Thermostat set points are provided by smart thermostats. Once trained, the models can use thermostat set points and dry bulb temperatures to predict A/C loads. The ROMs are used to evaluate the potential of automated thermostat control to reduce the aggregate peak demand. A centralized model predictive controller reduces the aggregate peak load by adjusting the thermostat set points to pre-cool houses and staggers the time A/C units turn on.