Research

I’m currently working on completing a PhD in Computer Science in the Electronics and Computer Science department, at the University of Southampton. I’m undertaking my PhD as part of the iDEaS Project, in the Agents, Interaction and Complexity Group.

This page is dedicated to the research that I’m doing as part of my PhD. On this page you will find information about my research interests, my PhD, and publications that I have had a part in writing. The next section provides a brief list of what is on the page, where clicking on each item will skip you straight to the relevant section.

On This Page

Summary of my Research Interests

A Wordle of my Transfer Thesis

A Wordle of my Transfer Thesis

My PhD

I’m currently studying for a PhD in computer science under the supervision of Dr. Alex Rogers, and Dr. Enrico Gerding, in the Agents, Interaction and Complexity group, at the University of Southampton. Broadly, my PhD follows the research themes of agent-based computing, and computational economics. More specifically, in my PhD, I’m studying how we can use mechanism design in order to obtain specific behaviours from electricity consumers in the smart grid. I’m looking at a problem in which there are numerous consumers, and a single aggregator whose job it is to optimally purchase electricity for the home agents. The supply of electricity must always match the demand — if there is surplus demand, the grid becomes unstable, and if there is surplus supply, money is wasted generating electricity which is unused. Consequently, electricity companies are penalised for purchasing electricity at the ‘last minute’, and thus they receive better rates for planning for their customers’ demand ahead of time.

Currently, companies make predictions of their customers’ future consumptions through the use of historical data and statistical models. However, there exists a wealth of information within each house that can be used in order to more precisely predict the future consumption of that building.

In my PhD, we say each home has an home agent — an intelligent piece of software that autonomously gathers information, which it can then use to construct a prediction of its houses’ consumption. The home agent, is completely autonomous, and interacts with the electricity aggregator, represented by an aggregator agent, in order to transmit a prediction to the aggregator that it believes will maximise its net payoff (i.e. the amount it is rewarded minus the cost of generating the prediction).

The rewards given to the home agents must be carefully designed in order to ensure the home agents behave appropriately. If the home agents find that they are able to gain a higher reward when either they ‘guess’ their consumption, without taking into account any information, or lie about their estimate (for example, to make it seem as though they’re more confident in it), then they will do just this. Of course, for the aggregator, this isn’t beneficial as it will be using incorrect data in order to try to optimally purchase electricity for its customers. Consequently, rewards must be designed such that the home agent expects to maximise its reward only when he reports its true estimate.

I’m also incorporating other incentives into the rewards in order to promote the efficient use of electricity. For example, wildly varying loads being placed on the grid are responsible for a large amount of waste. In order to cope with unexpected demand, there are reserve generators, constantly spinning but not outputting to the grid until the demand is placed. This results in increased emissions, and a greater cost of electricity. By encouraging users make their use of electricity more constant (or by employing technology to do so autonomously), we can reduce the need for standby generation and increase the efficiency of the grid.

An important part of this work is to learn the behaviours of the home agents under such schemes. For example, agents’ behaviours should be analysed in order to discover any undesirable side effects. It could be, for example, that when basing agents’ rewards on their predictions, they have an incentive to needlessly burn electricity to match their predictions when they discover they will consume less than they initially predicted.

Selected Publications

Below is a list of publications I have authored (in reverse chronological order). Clicking on the title will take you to that paper’s corresponding ePrints page where you will be able to download a copy of the paper in PDF format.