A deep learning framework for global transportation energy carbon emission forecasting: integrating generative pre-trained transformer with multi-scale feature analysis

The transportation sector contributes approximately 20% of global carbon dioxide emissions, posing significant challenges for energy transition and decarbonization efforts. We proposed TransCarbon-GPT, an advanced deep learning framework based on a generative pre-trained transformer architecture, designed to forecast transportation-related carbon emissions across 22 major economies. This framework integrates a multimodal dataset encompassing 33 domains and over 29000 feature variables, including energy price indices, fossil fuel consumption patterns, and energy policy indicators. Leveraging transfer learning techniques built upon the open-source LLaMA3 model, TransCarbon-GPT achieves state-of-the-art predictive performance, with SMAPE values ranging from 0.3782% to 5.7329%, significantly surpassing conventional forecasting approaches. The framework employs SHapley Additive exPlanations (SHAP) to identify key drivers of carbon emissions at both global and national scales to enhance interpretability. Our findings highlight energy price volatility, economic policy uncertainties surrounding energy transitions, and geopolitical risks as dominant factors influencing transportation emissions, with distinct impacts observed between developed and developing nations. Notably, natural gas prices influence more than crude oil prices in economies with diversified energy portfolios. Ablation studies reveal that incorporating patching reduces RMSE and MAE by 23.09% and 19.23%, respectively, while channel independence achieves reductions of 20.48% and 17.92%. Combining both components, the hybrid architecture delivers the most substantial improvements, reducing RMSE and MAE by 68.45% and 72.01%, respectively. TransCarbon-GPT provides actionable insights for policymakers to design targeted carbon reduction strategies, supports transportation enterprises in optimizing energy consumption, and facilitates the development of cleaner energy pathways, advancing the transition toward energy-efficient transportation systems.