Research on Hybrid LNG Price Forecasting Model Based on Decomposition-Integration and Error Correction
QIN Qing
MIAO Kunlin
SUN Lifan
Abstract：The nonlinear and non-stationary characteristics of liquefied natural gas price series limit the forecasting accuracy of traditional models.To address this,a hybrid forecasting model is developed following the framework of"mode decomposition-sequence reconstruction-heterogeneous modeling-error correction."Specifically,variational mode decomposition is employed to decompose the original price series into multiple modes,which are then reconstructed into low-frequency and medium-frequency subsequences using sample entropy clustering.ARIMA and GRU networks with integrated attention mechanisms are applied to forecast these subsequences respectively,followed by XGBoost for secondary residual correction.Single-step rolling forecasting experiments on China's daily LNG ex-factory prices demonstrate that the proposed model outperforms all benchmark models,achieving a 44％reduction in MAE compared to the best baseline GRU-Attention,with the Diebold-Mariano test confirming the statistical significance of its forecasting superiority.Ablation analysis validates the effectiveness of the error correction module,while tests across different market scenarios further confirm the model's robustness.
Keywords：variational modal decompositionsample entropyheterogeneous modelingerror correctionLNG price forecasting
Publication Date：2025-10-25
Online Publishing Date：2025-11-17(First online date of this platform, not the publication date of the document)
Pages：9( 96-104 )