Integrating Weather Conditions to Forecast the Electricity Price in Belgium Market: A Novel Mixed-Frequency Machine Learning Algorithm

The fluctuations in weather conditions, such as temperature and wind speed, can impact the process of solar and wind power generation, thereby influencing electricity prices. Real-time price settlement requires a higher resolution than daily frequency to forecast electricity prices. This research proposes a novel mixed-frequency model to address the issue of frequency inconsistency problem between electricity prices and related factors, and successfully applies it to Belgium electricity price forecasting. Firstly, the RU-MIDAS method is used to analyze the dynamic impact of weather conditions on prices. Then, RU-MIDAS is combined with machine learning algorithms to predict the prices. The results of error metrics and MCS test indicate that humidity can improve the prediction accuracy of all four sequences. The prediction ability of wind gusts is comparable to that of the highest price; the former helps predict the last two subsequences, while the latter improves the prediction accuracy of the first two subsequences. Temperature can only help predict the fourth electricity price series, and the forecasting ability of weather factors is consistent with the order of feature importance.