Project Highlights:


  • Create new insights for the design of near real time energy demand response interventions
  • Develop digital tools to implement in real world with active participation of end users
  • Test and validate the effectiveness of digital interventions


Two days in 2017 were noteworthy towards the decarbonisation of UK grid: a coal-free day on 22nd April and a mostly renewable energy powered day on 8th June. However, the intermittency of renewable resources mean that gas generation (more specifically, combined cycle gas turbines, CCGT) accounts more than 40% of power generation on average (see Figure 1 for hourly generation profiles on a random day). In order to reduce the need for economic and environmentally costly gas generation, alternative approaches to shift demands to match the availability of renewable supply are needed. Currently, these so-called ‘demand response’ interventions mainly focus on developing tariffs that vary to reflect the actual delivery cost of electricity in a given time period. By concentrating on total demand and ignoring the purpose of energy use (e.g. heating vs running dishwasher), it is not surprising that the success of these tariffs, such as critical peak pricing or time of use, to shift demands for electricity has been limited.

The aim of this PhD project is to develop a set of digital tools to create demand response for specific energy activities in near real time and test their effectiveness by collaborating with an energy supplier with actual households. Near real time interventions are used commonly in transport (e.g., taking an alternative route instead of a busy road) but the scope to create flexibility in energy domain is yet to be tested. Focusing on the duration, frequency and types of energy activities, this PhD studentship will bring insights from behavioural science literature to design data-driven innovative interventions to shift demands for electricity in near real time. Specific objectives of the project are as follows:

  • To bring insights from environmental psychology and sociology to inform the design of innovative near real time demand response interventions in energy domain
  • To develop a predictive model of energy activities using diverse datasets and several machine learning techniques and assess their performance
  • To design and develop digital tools to create flexibility and test their effectiveness in field trials in collaboration with an energy supplier


Figure 1. Power generation mix on 20 June, 2017


The project will make use of artificial intelligence and big data analytics methods to design and test near real time demand response interventions. Using published data on household electricity profiles (e.g. Ofgem funded Low Carbon London or Customer-led Network Revolution projects), we will develop insights into the sequence of energy activities by different socio-economic groups in the first 18 months of the project. Then, these developed tools will be tested in the field by collaborating with an energy supplier between 18-30 months. The last six months of the project will focus on synthesis of the results and their dissemination.

Training and Skills

The research will require skills to use artificial intelligence and big data analysis methods. The successful candidate will benefit from training courses on data and information fusion, Machine Learning for Computer Vision Applications and programming with Python or R languages.


Year 1: Main activities will focus on providing the applicant with strong analytical and theoretical background as well as start developing some online tools to be tested in the field. Some training courses and literature review will be undertaken to find out about energy practices and new approaches to extract economic value of consumer’s flexibility to shift their demands to other times. This review will identify types of artificial intelligence (AI) methods as well as insights from behavioural science on creating behaviour change.

Year 2: This phase will build a data-driven modelling approach of the energy behaviours and their disruptions with near real time interventions. Using machine learning techniques, such as neural networks and support vector machines, the model will aim high degree of generalizability and adaptability and minimum computing efforts. A comprehensive set of meta-features will be used as inputs to the learning models. Also, different types of feature reduction techniques will be investigated to improve the performance of the learning models. Based on this understanding of sequence of energy activities, near real time interventions via digital tools like text messages or apps (e.g. ‘tomorrow it’s going to rain, do your laundry today’) will be implemented to shift patterns of energy use. We will experiment with different socio-economic groups or urban/rural contexts.

Year 3: This phase will test the effectiveness of the interventions to disrupt the households’ energy activity patterns and dissemination of the research findings.

Partners and collaboration (including CASE)

Subject to further approval, we have discussed the project with one of the innovative energy suppliers. They currently provide demand side responses to commercial and industrial sectors. However they would like to expand their customer base to the domestic sector. It is very likely that the PhD student will have the opportunity to work with this company and benefit from in-kind support to access data, test the online tools developed with the households in the field as well as to have a placement at the company’s offices in London. It is anticipated that the student will be seconded for up to 6 weeks during months 18-30.

Further Details

The student will be based at the Cranfield campus at Cranfield in Bedfordshire - https://www.cranfield.ac.uk/About/How-to-find-Cranfield