The global energy transition and geopolitical uncertainty have intensified energy-market volatility, leading to sharp fluctuations in liquefied natural gas (LNG) prices and increasing risks for energy security and market stability. This study develops a robust and interpretable forecasting framework for China’s LNG prices while identifying the key drivers of price fluctuations. We propose a “Trend–Seasonal–Random” multiscale time series decomposition learning framework for LNG price forecasting. The framework integrates multiple linear regression (MLR) and XGBoost for trend modeling, employs seasonal autoregressive integrated moving average (SARIMA) to extract seasonal components, and applies exponential smoothing (ES) to capture short-term disturbances. The empirical analysis is based on China’s LNG spot prices from 2014 to 2024 combined with multidimensional variables covering supply–demand, inventory, alternative energy, and financial markets. Monte Carlo combination optimization is used to select the most relevant factors. The results show that the proposed framework achieves a mean absolute percentage error (MAPE) of approximately 8% and a directional accuracy (DA) exceeding 90% for out-of-sample forecasts, significantly outperforming traditional autoregressive integrated moving average (ARIMA) and direct forecasting models. Comparative results indicate that the MLR model is robust and interpretable—well suited for long-term trend analysis—while the XGBoost model excels in nonlinear fitting and short-term forecasting but tends to systematically overestimate under extreme market conditions. The optimal factor selection reveals that carbon prices, coal prices, nuclear power outages, China’s LNG import volume, and U.S. natural gas inventories are the primary determinants of price fluctuations. The proposed framework enhances the accuracy and stability of energy price forecasting and provides practical insights for corporate risk management and policy formulation.
Wang et al. (Wed,) studied this question.