A Hybrid Prediction Model for International Crude Oil Price Based on Variational Mode Decomposition with BiTCN-BiGRU-Attention Deep Learning Techniques

Abstract Predicting the price and volatility of international crude oil futures is a complex task. This paper presents a novel hybrid prediction model, the VMD-BiTCN-BiGRU-Attention, which integrates variational mode decomposition (VMD) and advanced deep learning techniques to forecast the nonlinear, non-stationary, and time-varying characteristics of crude oil price sequences. Initially, the price sequence is decomposed into multiple modes using VMD, enabling the capture of different frequency components. Each mode is independently predicted using a bidirectional time convolutional network (BiTCN), which captures temporal sequence information and enhances long-term dependencies through dilated convolution. Subsequently, a bidirectional gated recurrent unit (BiGRU) models the temporal dependencies more effectively, while an attention mechanism adjusts the weights of the BiGRU outputs to emphasize critical information. The model’s predictions are optimized with the Adam algorithm. Empirical results demonstrate that the model is adept at forecasting non-stationary and nonlinear international crude oil prices. Furthermore, the Diebold-Mariano (DM) test confirms that this model surpasses 15 other models regarding accuracy and performance, achieving optimal results with key metrics: R² = 0.9953, RMSE = 1.4417, MAE = 0.7973, and MAPE = 1.5213%. These findings underscore its potential for enhancing crude oil price prediction.