One of the key components of the UK government's strategy to reduce carbon emissions is to carry out a nationwide rollout of smart energy monitoring devices.

This is set to take place over the next decade and will result in over 27 million households being equipped with intelligent metering systems that can monitor electricity consumption in 15 minute intervals and facilitate easy communication of usage data. 

At a cost of approximately £10 billion to implement, it is imperative that the vast amount of data that will be generated by these devices is handled in an efficient and effective manner. The implementation of smart meters will facilitate the creation of a smart grid to allow for much more efficient energy generation management, but these devices also allow a unique opportunity to observe an individual's electricity consumption and help them improve on their own carbon footprint. This is supported by the government proposal because it states that all households must also have an in-home display (IHD) to inform them of their electricity consumptions.

This project is supported by Green Energy Options (GEO), a Cambridge based company who are one the UK and Europe's leading suppliers of smart energy management display systems. The data being used in this project was generated from a trial carried out by GEO using a number of their display devices to monitor individual appliance and complete household usage in more than 150 homes. Each file of the data typically has around 365 days worth of electricity consumption data; each day is represented as 96 readings taken at 15 minute intervals. The overall electricity can simply be recorded from the total amount of electricity coming into a household, whilst specific device usage is observed using ‘plugbugs'. The device to be monitored is plugged into the wall socket via a plug bug, which communicates wirelessly with an IHD. 

Some examples of daily usage from 10 common household devices are shown below. The examples are one dimensional profiles of typical usage, sampled every 15 minutes throughout a single day. The ten devices shown are: screen group (television, computer), oven/cooker, washing machine, immersion heater, dishwasher, cold group (fridge, freezer, fridge/freezer) and kettle.

However, simply collecting data such as this and presenting it to a consumer is clearly not going to modify long term behaviour or cause any significant reduction in energy consumption. Whilst many reports suggest that energy display systems can reduce electricity usage by around 2.5%, very little is known about how people will react to smart meters and how best to use their output to encourage reduced consumption without a detrimental effect to a household's lifestyle. The long-term success smart metering to alter consumer behaviour will be strongly influenced by:

  • What information can be extracted from the usage data
  • How this information is presented to best inform the consumer
  • Whether the consumer can be encouraged to interact with the device in order to act on this information

From a data mining standpoint, the primary concern is obviously how to extract knowledge from the data. However, the models that are formed from this data are likely to be influenced by how information can be presented to users and how they can be encouraged to interact with it. For example, a system developed to detect anomalous behaviour of devices would likely encourage engagement with a smart metering system if it could warn consumers when devices are malfunctioning or behaving incorrectly. However, for an automatic system like this to exist, a prerequisite is to be able to determine the class of a given device when presented only with of unlabelled previous usage. In response to this, we investigated the problem of classifying household devices based only on their usage profiles over both daily and weekly intervals. We found that with a weekly profile, we can accurately discriminate between classes of device by deriving a set of descriptive features and classifying them with either Random Forest or nearest neighbour classifiers.

Future aims of the project include developing an anomaly detection system and also investigating and developing data mining algorithms that can be applied to this kind of data in the future.

References

Classification of Household Devices By Electricity Usage Profiles (IDEAL 2011, LNCS Vol. 6936), J. Lines, A. Bagnall, P. Caiger-Smith, S. Anderson.

Research Team

Mr Jason Lines, Dr. Anthony Bagnall, Dr. Richard Harvey

Collaborators