A weekly crude oil price interval-valued prediction architecture on fusion of decomposition technique and adaptive integration

Accurate prediction of crude oil price is significant for energy supply stability, investment decision, and sustainable energy economic development. In fact, even within the same week, crude oil price can fluctuate significantly at different time stages, and point prediction methods cannot capture this phenomenon. Therefore, this paper proposes a multivariate weekly crude oil price interval-valued prediction architecture based on interval decomposition and adaptive integration. Firstly, interval grey incidence analysis (IGIA) was used to select feature interval-valued time series (ITS). Next, multivariate variational mode decomposition (MVMD) was employed to decompose oil price and optimal features ITS simultaneously. Thirdly, an attention weight multi-output least squares support vector regression (AW-MLSSVR) was applied to predict each subseries, respectively. Finally, an adaptive integration (AI) strategy was employed to weight the predictive ITS of each subseries. In this framework, MVMD solves the defect of univariate decomposition techniques that cannot extract multivariate scale common mode features. AW fully learns the similarity between features and targets, while AI addresses the limitation of simple addition as integration operation. The two datasets demonstrated that compared with the optimal interval-valued model, this architecture promoted on the six evaluation indicators, with average reductions of 39.97 %, 37.83 %, 39.06 %, 37.95 %, 5.79 % and 59.85 %, respectively.