Go to content

Using machine learning to achieve better electricity price forecasting

Using machine learning to achieve better electricity price forecasting

The use of machine learning methods in electricity price forecasting can have a significant impact on the profitability of stakeholders in electrical energy markets.

The above finding was confirmed in a study done at the School of Economics and Business, University of Ljubljana, led by Prof. Miroslav Verbič, PhD, and Prof. Jelena Zorić, PhD, which compared the success of alternative methods for electricity price forecasting in daily markets with an extremely high degree of volatility. Improvement in forecasting accuracy may have a significant impact on the business operations of companies. This article analyses the forecasting accuracy of algorithms from the families of machine learning, data mining and deep learning algorithms, compared to the time series econometric model.

Modern forecasting techniques are applied to the poorly researched electrical energy markets in the countries of Central and South-Eastern Europe, with emphasis on data from the Greek and Hungarian power exchanges. The results of this analysis have shown that forecasting using the supporting vectors method is statistically significantly better than the use of the time series econometric model, whereby the choice of the calibration window size has a marked effect on forecasting accuracy. The latter also has a direct effect on cost reduction or improvement in the profitability of business operations of the stakeholders in the electrical energy market one day ahead.

This is also the first article based on publicly available data from the ENTSO-E platform, which was designed in 2015 with the aim of increasing market transparency. The exclusive use of public data all available in one place enables the reproducibility of the results; at the same time, it can contribute to the reduction of information asymmetry between large and small players, thus leading to greater transparency of the power market and a lower likelihood of distorted market competition.

Source: HALUŽAN, Marko, VERBIČ, Miroslav, ZORIĆ, Jelena. Performance of alternative electricity price forecasting methods: Findings from the Greek and Hungarian power exchanges. Applied energy, 2020, vol. 277, art. 115599. ISSN 0306-2619. DOI: 10.1016/j.apenergy.2020.115599. [COBISS.SI-ID 25225731]