Home energy made up nearly 25% of all energy consumed in the United States in 2010, and 31% of this energy is estimated to come from home heating, ventilation, and air conditioning systems. Yet, the inputs available to a homeowner for those systems are not easily correlated to the energy consequences, making it difficult (if not impossible) for a homeowner to choose set points to meet a budget each month. This brief presents a novel management framework targeted toward end users, which incorporates a light-weight data-driven prediction component to dynamically learn the data relationships of the underlying system and estimate its future behavior within a given time horizon. Traditional prediction methods to close the loop require a state update function, but such an approach would require installation of additional sensors, and frequent modification of the optimization functions as parameters to these state update functions change. Furthermore, disturbances in the system frequently have higher impact on performance than can be mitigated with more accurate estimates of the state update function. To mitigate these issues, this brief proposes an averaged sectional model built from sampled data, and employs linear regression to learn the data relationships that are used for prediction. In this way, it is possible to perform optimization without the need to develop more accurate state update models. The optimization component is integrated to determine how to best manage the underlying platform, based on end user defined constraints, as the underlying system or its environment changes. The results are applied to gathered data in a single-family Arizona home over several months, to demonstrate validity when applied to temperature variations introduced by daily life.
- Averaged sectional model (ASM)
- and air conditioning (HVAC)
- data-driven prediction
- predictive control
ASJC Scopus subject areas
- Control and Systems Engineering
- Electrical and Electronic Engineering