A Hybrid PLO-Transformer-CEEMDAN Model for Natural Gas Price Forecasting

To address the nonlinear and non-stationary nature of natural gas prices, this study proposes a hybrid PLO-Transformer-CEEMDAN forecasting framework. The model integrates weight optimization, trend extraction, and residual correction, leveraging 27 heterogeneous features—including Google Trends—to enhance predictive performance. Empirical tests on 2004–2024 Henry Hub monthly data demonstrate that the model achieves an RMSE of 0.0257 and an <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">R</i>² of 0.9998, reducing RMSE by 92.87% compared to the baseline Transformer. Notably, incorporating Google Trends alone improved accuracy by 30.91%, highlighting its significant value in capturing market sentiment ahead of price movements. The proposed method significantly outperforms conventional benchmarks in both accuracy and generalization, providing a robust tool for energy price forecasting and market risk management.