- Big Data Methods and Social Sciences
- Time Series Methods
- Modelling Heterogeneous Patterns of Behaviour
- Missing Data and Attrition in Longitudinal Studies
- Innovations/challenges in the Teaching of Statistics in Social and Applied Sciences
Big Data to the Rescue? Challenges in Analysing Granular Household Electricity Consumption in the United Kingdom (with S. J. Mikhaylov), Energy Research and Social Science, Vol. 64, 2020
Predicting Energy Customer Vulnerability from Consumption Behaviour (with S. J. Mikhaylov), Working Paper
Work in Progress
Generalized Additive Models for Residential Energy Load Prediction, In Preparation
Decision-tree Methods for Prediction of Attrition in Large Longitudinal Studies (with Aja Murray), In Preparation
Ushakova, A., & Murcio, R. (2018). Interpreting Smart Meter Data of UK Domestic Energy Consumers. In Longley P., Cheshire J., & Singleton A. (Authors), Consumer Data Research (pp. 120-137). London: UCL Press.
Overview of PhD Thesis work
Thesis Title: Generating Insights from Smart Meter Data: Challenges and Opportunities’
The introduction of smart meter technology has been central to recent innovations in energy provision for the UK residential sector. Smart meters have the potential to give greater insight into energy consumption behavior for energy providers and researchers alike. For example, they may aid our understanding of how the consumption of gas and electricity may be replaced by the energy from renewable sources, or how consumer behaviors can be changed to reduce overall energy consumption, increase efficiency, and lessen the pressure on the national grid networks. The advantage of a thorough understanding of the insights generated from smart meter data for policy issues may sound obvious at a first glance. However, there are significant challenges associated with the availability of methods and computation necessary to perform a complete analysis of the available data.
The thesis provides an in depth look at the nature of energy consumption through an analysis of big data that is recorded by around 400,000 smart meters installed at residential properties across the UK. It further discusses how this data is different from perhaps more conventionally collected retail consumer data, and in what way does the temporal nature of these data reveal information about the customers dynamics without compromising their anonymity. Various machine learning methods are applied and surveyed against more conventional methods often used by researchers and industry practitioners. Some extensions to improve the accuracy and reliability of methods for both segmentation of the behaviour, and prediction are also suggested. This latter point is particularly relevant if one is interested to forecast future energy use on both national and individual scale levels. Lastly, a case study looking at identifying the fuel poor from smart meter data is presented as an illustrative example of potential research questions one may answer with smart meter data records.