Machine learning and multimedia content generation for energy demand reduction

Domestic energy demand accounts for about 30% of overall energy use. The IDEAL project uses a variety of IT methods to investigate whether, and in which social groups, feedback of personalised, household-specific and behaviour-specific information results in greater reduction in energy use than overall consumption information reported by Smart Meters. It is a sociotechnical study, concentrated on existing housing, with a strong social science component and an experimental design that looks at income levels and household composition as primary factors. Temperature and humidity data related to behaviour is gathered using a small number of wireless sensors in the home, together with data on weather, building factors and household composition. This data is streamed over the internet to servers where it is analysed using Bayesian machine-learning methods to extract household-specific behaviours in near-realtime. Information on the cost, carbon content and amount of energy used for specific behaviours is reported back to the householders via a dedicated wireless tablet. This interactive content is automatically generated using multimedia methods based on natural language generation techniques. The project is in its design phase, with the main project planned (and funded) to run 2013-2016. It is anticipated to demonstrate whether such low-cost sensing, analysis and feedback is significantly more effective than standard Smart Meters in reducing demand, and a business opportunity for green service organisations.