A study on machine learning models in predicting volatile spot prices : a case study on Norway’s electricity market

Abstract

The uncertainty caused by the increased use of renewable energy sources makes it more essential to find good forecasting tools that can offset the increased risk in predicting elspot prices. Different supervised machine learning models are applied in this thesis to predict electricity prices for the different price areas in Norway using hourly data for elspot prices, energy prices and temperature collected for the period 2014-2020. The results show that some models are better suited for predicting elspot prices compared to others, with the Linear regression model, Gradient Boosting and Extra Randomised Trees regressor (ET) giving the best results out of the 11 tested models. The findings also suggest that choosing seasonal forecasting horizon together with adding more explanatory variables such as system load and wind power will improve the predictive performance of the models by capturing price spikes and anticipating changes in the elspot prices that longer forecasting horizon fail to capture.