Wavelet Multivariate Adaptive Regression Splinesand Their Application to the UK Electricity Market

The growing effect of electricity prices on energy markets appeals for more accurate forecasting techniques since the market suffers from high volatility, high frequency, nonstationarity and multiple seasonality. Aiming at maximum utilities under highly-volatile conditions, both the supplier and the consumer sides seek to monitor and response to the ongoing changes of the electricity prices. In this study, we use a new hybrid approach, called Wavelet - Multivariate Adaptive Regression Splines (W MARS), to forecast day-ahead electricity prices by considering their challenging structures. Here, wavelet transform captures multiple seasonality, unusual behaviors and volatility, whereas MARS eliminates the selection of explanatory variables problem. Hence, W MARS method combines the advantages of Wavelet Transform and Multivariate Adaptive Regression Splines in terms of all prediction performances, computational efforts and learning effects. In contrary to classical regression and time-series techniques, no assumption is required for the data structures or during the implementation process. This new hybrid method overcomes the drawbacks of previously recommended techniques for huge data sets by a fast and automated machine learning method. The performance of the new W MARS method is tested by using United Kingdom electricity market data. That market is characterized by day-ahead forecasting and half-hourly metering. The presentation ends with a conclusion and an outlook to future studies