Non-Intrusive Load Monitoring (NILM) and similar methods

A combination of active and reactive power used when appliances are switched on can be used to identify individual appliances and measure energy use for those appliances. Each appliance can be identified and monitored using this unique signature within the household supply.

Intel have created prototype appliance signature detection products. Intel's devices so far require limited training of the monitoring device to recognise each individual appliance profile. Intel demonstrated the system during the Intel 2010 developer forum smart energy presentation (@19:49). Ireland's Episensor also requires specific training.

A Cambridge University start-up, Navetas offers an advanced home energy monitor which does not require training of the device to recognise each individual appliance. The system can be retrofitted by electrical utility providers. Navetas is a UK company and it possible their product will work in New Zealand without significant modification.

In recent press Navetas state: "while home energy monitor products have recently become popular, these have limited functionality as they only provide an estimate of usage. The problem for the householder is that the final bill can vary widely from this 
estimate. Our solution provides an accurate measure of the actual energy usage and can be broken down by appliance. It is possible then to compare the efficiency of appliances and see, at a glance, where energy is being used".

The Energy Detective's (TED) 'Footprints' software available for the TED 1000 and TED 5000 devices includes a load profiling feature. This feature is best able to recognise larger constant loads but has trouble recognising some variable and complex appliance loads. It is unlikely this system uses complex NILM methods.

Enetics in the USA offer professional systems for integration with existing smart meters, and standalone systems. Enetics is associated with George Hart, an early researcher in the NILM field. Enetics also provide data collection and analysis services for utilities if needed.

Many users of power monitoring products already perform their own simple profiling and analysis. For instance, if the electrical load increases by approximately 3000 watts and the only 3000 watt load in the house is the hot water heating element: It is a safe conclusion that a 3000 watt increase or decrease can be attributed to the hot water system switching on or off. Appliances like refrigerators and freezers have patterns which are more complex but still predictable. Experienced users who track data visually on a PC can identify these patterns without difficulty. 

Distribution of appliance use data and use patterns has obvious privacy and security implications. In addition there are large potential benefits for energy efficiency and conservation through highly accurate knowledge of in-service energy profiles. Both general and specific appliance faults may be diagnosed using these methods.

Related links and papers:

The Smart Electricity Grid and Scientific Research [PDF, View online]
A Taxonomy of Load Signatures for Single-Phase Electric Appliances [PDF, View online]
A Field-Based Approach to Non-Intrusive Load Monitoring [PDF, View online]
NILM uses in appliance and HVAC fault detection (Norford [Web], Leeb [PDF, View online])
George Hart (an original NILM researcher, page no longer maintained)


Presentation slides. Lucio Soibelman, H. Scott Matthews, Mario Berges, Ethan Goldman. (Carnegie Mellon)

"Individual appliances’ electricity consumption is automatically disaggregated from a single custom metering system on the main feed to an occupied residential building. A data acquisition system samples voltage and current at 100 kHz, then calculates
real and reactive power, harmonics, and other features at 20Hz. A probabilistic event-detector using the generalized likelihood ratio (GLR) matches human-labeled events to the time-series of features. Machine-learning classification was most successful
with the 1-nearest-neighbor algorithm, correctly identifying 90% of the laboratory-generated training events and 79% of validation examples. The challenge of obtaining adequate training data for the real-world home leads to the development of the Wire
Spy, a wirelessly-networked event detector with an inductive sensor which clamps to the cable of an appliance"


"[...] The use of standard digital electricity meters could avoid this problem but the loss of information of the measured data has to be compensated by more intelligent algorithms and implemented rules to disaggregate the total load trace of only the active power measurements. The paper presents a new NIALM approach to analyse data, collected from a standard digital electricity meter" 

"[...] the consumption data could be collected by a low cost optical sensor clamped on the existing electricity meter cf. [10]. Typical NIALM systems with automatic setup use pattern recognition methods based on a priori knowledge from connected
expert systems to detect appliances. Some other (cf. [4] and [6]) need initial training periods with manual entering of data. This NIALM approach tries to find recurring appliance patterns without manually entering of initial data"

"The presented algorithm is designed to analyse rough data from ordinary electricity meters and has been tested with simulated and real data from different domestic homes. The analysed real data was collected with an optical sensor clamped on the already installed electricity meter in domestic homes"

The Flukso open source power monitor blog links to additional papers.

Updates:

Oliver Parson (Phd in NILM) maintains a list of companies providing NILM technologies and service and blogs regularly on NILM and load disaggregation topics.

Jack Kelly has taken some steps to establish an online community for NILM researchers.


Comments