Predicting Natural Gas Prices Based on a Novel Hybrid Model with Variational Mode Decomposition

Natural gas firmly plays the crucial role in the transition through low-carbon energy systems and country’s energy security. Accurately and reliably forecasting gas price can be considered as a significant issue to both related investors and policymakers. This study adopts a decomposition-ensemble model to improve the accuracy of predicting gas price. The process is as follows: Firstly, the original energy price sequence is decomposed into subsequences with different frequencies through variational mode decomposition (VMD) with genetic algorithm. Secondly, Elman neural network (ELMAN) predicts the last high-frequency subsequence and bidirectional gated recurrent unit neural network model (BiGRU) forecasts other subsequences. Thirdly, the final forecast of the subsequences with various models is integrated with nonlinear integration approach. After the empirical study, three loss functions including root mean square error, mean absolute error, mean absolute percent error, the determination coefficient $${R}^{2}$$ R 2 and the modification in Diebold-Mariano statistics (the MDM test) are adopted as the evaluation criteria. The empirical results prove that the proposed method outperforms other compared models under different test set lengths and different periods of business cycle. Overall, experiments on the natural gas prices demonstrate the validity and superiority in the established VMD-BiGRU-ELMAN through the nonlinear ensemble method. The proposed method can simultaneously develop different advantages of corresponding model providing a useful prediction method to governments and enterprises.