Non-intrusive load monitoring

Working group:

Vojkan Tasic, Thorsten Staake


Non-intrusive load monitoring (NILM) is a concept known for more than 20 years. Despite the long presence of this concept, it has recently received considerable attention in the realm of energy conservation. In this domain, the main objective of NILM is to gain insights into electrical energy consumption of individual devices in private households.Therefore, NILM techniques aim at disaggregating consumption data in order to identify individual devices or groups of devices. The retrieved information can support a number of services, including:

  • Better consumption feedback: NILM can provide targeted information on which appliances or behavior make up a considerable share of overall consumption. This can help users to take targeted countermeasures to reduce energy demand.
  • Consulting services: Energy consulting today is tedious as it often requires an expert visiting a site. NILM can offer cheap and convenient data acquisition for utility companies which can base their power grid management on this information. In the future, the consulting reports may even be generated automatically and sent frequently with billing information.
  • House activity detection: NILM can be used to detect presence or absence of people in an apartment. The information is valuable for security and home automation systems. Energy efficiency application, for example, can put heating or air conditioning systems into “sleep mode” when the inhabitants left the house.

NILM has the potential to be widely adopted as it could be integrated into smart meters which measure the consumption and send it automatically to the utility companies. With large number of installed devices, marketing options emerge. For example, the utility companies can offer services to customers to identify inefficient devices. With this information, they can send the special offers to the users and even put into practice payment concepts from energy contracting.

So far, the user-appliances interaction data is obtained by intrusive techniques solely. Intrusiveness means that utility companies need to install dedicated measurement units to individual devices of importance in order to track their usage for specific household. This approach has many flaws. One of them is the cost of installation and maintenance due to the number of installed devices. Intrusiveness can also mean that technicians need to “intrude” the household to install, calibrate, and maintain the devices.

NILM constitutes a “low-intrusion” approach. It requires one-time installation as well as an initial training period. Installation in most cases is at the mains of the household.

There are small challenges which come with this approach. Voltage and current are measured at only one point in the household’s electric network. This means that only the aggregated data is provided. The main question which naturally appears is: How to disaggregate this data properly in order to get the energy information about each individual appliance in the household. New concepts are published almost every day. Under laboratory conditions, reasonable detection rates are achieved, showing about 80% rate of accurate appliance recognition. The performance under real conditions, however are not so good, where recognition rates rarely reach 60%. There are methods which can go even above 60% but they require high-end algorithms and very high computational power.

The main goal of this project is to find a commercially attractive compromise between the computational power needed and recognition accuracy achieved. The results will be used for the design of a low-cost version of a NILM device which can be deployed in many households.