Electricity price prediction: a comparison of machine learning algorithms

Abstract

In this master thesis we have worked with seven different machine learning methods to discover which algorithm is best suited for predicting the next-day electricity price for the Norwegian price area NO1 on Nord Pool Spot. Based on historical price, consumption, weather and reservoir data, we have created our own data sets. Data from 2001 through 2009 was gathered, where the last one third of the period was used for testing. We have tested our selected machine learning methods on seven different subsets. We have used the following machine learning algorithms: model trees, linear regression, neural nets, RBF networks, Gaussian process, support vector machines and evolutionary computation. Through our experiments we have found that a support vector machine using an RBF kernel has the best prediction ability for predicting the NO1 electricity price. We have made several interesting observations that can serve as a basis for further work in the topic of electricity price prediction for Nord Pool Spot.