Non-intrusive load monitoring (NILM) is a way for homeowners and building managers to monitor energy consumption on an appliance-by-appliance basis without having to install dedicated sensors across an entire house or office building. NILM is not a new concept – the first non-intrusive devices date back almost twenty years to processes developed at MIT. But those early devices were far from user-friendly: setting them up required installation by a trained electrician and a manual, one-by-one synchronization with every appliance in the household.
Growing public interest in smart meters has put NILM back into the limelight. Smart meters build on conventional energy meters by giving them the ability to communicate with the central office of whichever utility company would normally read them. On one level, this allows utilities to track – and manage – customer energy consumption for their demand response programs. But it has benefits for the end user as well, offering consumers information on their energy consumption in a way that’s easily accessible even to a technical layman.
NILM builds on the basic principles of smart metering by adding another layer of technical sophistication.
In a household, each appliance has a unique energy “signature”. By analyzing the smart meter data, NILM identifies a signature for each device in the household, then uses a sophisticated algorithm to separate those signatures from the overall energy consumption.
Instead of a single aggregate total for energy consumption, NILM can enable users to see just how much their toaster oven or flatscreen TV is tacking onto their monthly electricity bill. Beyond the obvious advantages, there are other benefits to NIALM:
With more than 50 million smart meters deployed in the US, NILM can enable smart meters to be truly intelligent, giving energy-savvy users the tools to predict, monitor, and manage their own power consumption and efficiency on a very granular level.
Over the last five years, several companies and research groups worldwide started pursuing NILM technologies, which resulted in a few commercial products and hundreds of scientific publications. In this context, what is different about Fraunhofer USA's approach?
For a typical smart-meter signal, Fraunhofer USA's patented algorithms consider not only a power “signature” of an appliance, but also its temporal usage properties, such as duration of time on or off. Naturally, incorporating these additional features into the algorithms enhances their disaggregation accuracy. However, a straightforward machine learning approach that combines power and temporal appliance features would increase the run time exponentially with the number of appliances, making it impractical for most applications. Fraunhofer USA’s algorithms overcome this challenge by sophisticated division of appliances into groups, resulting in much more computationally efficient linear scaling.
Whereas smart meters can usually provide power consumption information with up to a second resolution in time, special sensors can sample electric current at much higher resolutions, giving the information on the current waveform. The waveform information can be used to characterize the appliance “signatures,” and a pattern recognition technique can be used to differentiate appliances based on the waveform signatures. For such high-resolution signals, Fraunhofer USA developed an approach that optimally fuses together several pattern recognition techniques, which further increases disaggregation accuracy while, at the same time, controlling for false detections.
