The system is a so-called Non-Intrusive Load Monitoring (NILM) and it identifies the usage of the individual electricity consumers by monitoring their voltage and current. After that, it uses DSPs (digital signal processors) and machine-learning algorithms to identify the device and its level of consumption, by matching the signals against a library of learned devices.
“There are many opportunities for reducing electricity consumption in buildings, but identifying and quantifying them is often very difficult, particularly in single-family homes,” said Dr Mario Berges from Carnegie Mellon University. “This means that for most residents the only indicator of consumption they have is their monthly electricity bill.”
The NILM devices need a special training period, during which it learns the different modes of operation for the various electric devices in the household. The researchers are currently studying the case in which they could embed the most used household consumers in NILM’s factory memory.
A continuous monitoring of the electricity consumption could lower the existing 37 percent used electricity in the U.S. only by residential buildings. This would account for large decreases in carbon dioxide and particulate emissions, because many of the power stations in the US are still powered by coal.