Algorithmic approaches to energy market price prediction

Abstract

A new energy market, the Integrated Single Electricity Market (ISEM), went live in Ireland during October 2018 providing more flexibility and competition to energy traders. This recent development requires energy traders to purchase energy in advance. Therefore, if traders could accurately predict usage and the correct time at which to purchase energy, they could optimise their costs. Price prediction through statistical and computational approaches would be a valuable commercial tool when forecasting electricity prices to capture market trends with the aim of reducing market costs to increase profits.

This thesis explores day-ahead electricity price forecasting within the ISEM and British Electricity Trading and Transmission Arrangements (BETTA) energy markets. Traditional statistical approaches were first considered by utilising time-series prediction models with historical data to observe energy market trends. Appropriate stationarity, integration, and seasonal checks were included in the traditional statistical models for estimation and diagnostic testing. Next, non-linear regression models were applied to model input-output relationships and find key energy-related factors that influence current electricity price. The inclusion of energy-related model inputs were shown to influence price prediction. Refining the statistical models to only include the identified significant factors from the non-linear regression models improved overall accuracy. Technical indicators have shown great promise within the stock market, thus building on this knowledge eight novel energy price technical indicators consisting of trend, oscillator, and momentum types were developed. These technical indicators were used as inputs into machine learning algorithms and demonstrated highly accurate prediction performance. Therefore computational approaches would be advantageous for energy trading prediction modelling.

Overall this thesis demonstrates that many approaches may be used to predict energy prices. However, the combination of novel technical indicators and machine learning provided compact models that accurately represent the variability and dynamics within the energy market.

Date of Award	May 2022
Original language	English
Sponsors	Click Energy
Supervisor	Dermot Kerr (Supervisor) & Sonya Coleman (Supervisor)