Energy price prediction based on independent component analysis and gated recurrent unit neural network

Abstract

The changes of energy prices are exercising ever deeper influence on the world economic pattern. Meanwhile, due to the inherent non-stationarity and non-linear characteristics of energy prices, improving the prediction accuracy of the energy price is perceived as a challenging area to work in. Inspired by this, a novel hybridization of multi-scale model for predicting the energy price based on independent component analysis (ICA), gated recurrent unit neural network (GRUNN) and support vector regression (SVR), which is abbreviated to IGS, is proposed. Firstly, the energy price is decomposed into several intrinsic mode functions (MFs) by variational mode decomposition (VMD) technique. Then, MFs are modeled through ICA to separate out independent components (ICs) that reflect the inherent features of energy price. Secondly, applying the GRUNN model to the ICs for predicting the inner driving features, each of the predicted features may represents the future trends of different factors of the original data. Finally, by replacing the conventional linear combination or regression with SVR, the forecasting results are integrated into the prediction of energy price. Experiments on three types energy price series: natural gas price, crude oil price and carbon price demonstrate the validity and reliability of the improved IGS.