Determining relevant features in estimating short-term power load of a small house via feature selection by extreme learning machine
Monitoring and controlling electrical loads is crucial for demand-side energy management in smart grids. Home automation (HA) protocols, such as X10 and Insteon, have provided programmatic load control for many years, and are being widely deployed in early smart grid field trials. While HA protocols include basic monitoring functions, extreme bandwidth limitations (<180bps) have prevented their use in load monitoring. In this paper, we highlight challenges in designing AutoMeter, a system for exploiting HA for accurate load monitoring at scale. We quantify Insteon's limitations to query device status---once every 10 seconds to achieve less than 5% loss rate---and then evaluate techniques to disaggregate coarse HA data from fine-grained building-wide power data. In particular, our techniques learn switched load power using on-off-dim events, and tag fine-grained building-wide power data using readings from plug meters every 5 minutes.
Unfortunately, ACM prohibits us from displaying non-influential references for this paper.
To see the full reference list, please visit http://dl.acm.org/citation.cfm?id=2434023.