Toggle Main Menu Toggle Search

Open Access padlockePrints

Optimizing carbon tax for decentralized electricity markets using an agent-based model

Lookup NU author(s): Alex Kell, Dr Stephen McGough, Dr Matthew ForshawORCiD


Full text for this publication is not currently held within this repository. Alternative links are provided below where available.


Computer simulation allows practitioners to model real-world systems using software. These simulations allow for `\textit{what-if} ' analyses which can provide an indication as to how a system may behave under specific policies, environments and assumptions. These simulations become particularly important in real systems which have high costs, impacts or risks associated with them. A `digital twin' is a concept which has emerged in recent years. This is defined as a simulation of a specific instance of a system. This digital twin can then be used to learn of optimization techniques that can be applied to the real system. This foregoes the necessity of experimenting with the real system, avoiding potential adverse side effects.An electricity market is an example of a complex system which can be modelled using a digital twin. Disruptions to electricity supply, a substantial increase in the cost of electricity or unrestrained carbon emissions have the potential to destabilize economies~\cite{Kaseke2013,Masson-Delmotte2018}. It is for reasons such as these that electricity market models are used to test hypotheses, develop strategies and gain an understanding of underlying dynamics to prevent undesirable consequences \cite{Jebaraj2006}. In this paper, we use the electricity market agent-based model ElecSim to find an optimum carbon tax policy \cite{Kell}. Specifically, we use a genetic algorithm to find a carbon tax policy to reduce both average electricity price and the relative carbon density by 2035 for the UK electricity market. Carbon taxes have been shown to quickly lower emissions and lower the costs to the public \cite{Wittneben2009}. Carbon taxes are able to send clear price signals, as opposed to a cap-and-trade scheme, such as the EU Emissions Trading System, which has shown to be unstable~\cite{Wittneben2009}.In this paper, we use the reference scenario projected by the UK Government's Department for Business \& Industrial Strategy (BEIS) with model parameters calibrated by Kell \textit{et al.} \cite{DBEIS2019,Kell2020}. This reference scenario projects energy and emissions until 2035. We undertake various carbon tax policy interventions to see how we could reduce relative carbon density whilst at the same time, reduce the average electricity price.The parameter space we optimize over is the carbon tax price over a 17 year period from 2018 to 2035. The carbon price is used to influence the objectives of average electricity price and relative carbon intensity in 2035. Grid and random search are approaches which trial parameters at evenly distributed spaces and random spaces respectively. These approaches are often inefficient, however, and require an increased number of simulations due to their static nature. Genetic Algorithms, in contrast, use an evolutionary computing approach to find global optimal solutions faster.In order to optimize over two potentially competing objectives, i.e. average electricity price and relative carbon intensity, we use the Non-Dominated Sorting Genetic Algorithm II (NSGA-II) \cite{Valkanas2014}. The NSGA-II algorithm can approximate a Pareto frontier ~\cite{Pareto1927, Stadler1979}. A Pareto frontier is a curve in which there is no solution which is better than another along the curve for different sets of parameters. In this context, better means that a solution is closer to the optimal for a particular combination of objectives.We find that the rewards of average electricity price and relative carbon intensity are not mutually exclusive. That is, it is possible to have both a lower average electricity price and a lower relative carbon price. This is due to the low short-run marginal cost of renewable technology, which has been shown to lower electricity prices \cite{OMahoney2011}.The main contribution of this paper is to explore carbon tax strategies using genetic algorithms for multi-objective optimization. The following sections are set out as follows. Section \ref{sec:lit_review} covers examples of optimisations using genetic algorithms and different carbon strategies. Section \ref{sec:optimization_methods} details the optimization techniques applied. Section \ref{sec:sim_environment} explores the electricity market model used. We present our results in Section \ref{sec:results}, and conclude in Section \ref{sec:conclusion}.

Publication metadata

Author(s): Kell AJM, McGough AS, Forshaw M

Publication type: Conference Proceedings (inc. Abstract)

Publication status: Published

Conference Name: The Eleventh ACM International Conference on Future Energy Systems (e-Energy’20)

Year of Conference: 2020

Pages: 454–460

Print publication date: 22/06/2020

Online publication date: 24/07/2020

Acceptance date: 22/05/2020

Publisher: ACM


DOI: 10.1145/3396851.3402369

Library holdings: Search Newcastle University Library for this item

ISBN: 9781450380096