A hybrid LSTM and agent-based modeling framework for forecasting carbon emissions in China's transport sector

As a key contributor to global carbon emissions, the transportation sector confronts substantial hurdles in properly forecasting emissions due to the complexity of individual choices and the diversity of transportation architecture. This study identifies critical drivers of carbon emissions within the transportation sector, conducting an in-depth analysis of the distinct dynamics inherent to passenger and freight transport structures. We applied a Long Short-Term Memory (LSTM) model, achieving superior prediction accuracy compared to Backpropagation Neural Networks (BP) and Support Vector Regression (SVR). Furthermore, a high-fidelity passenger volume forecasting model was developed using an Agent-Based Modeling (ABM) approach. This model integrates time costs, monetary costs, environmental costs, and comfort costs into a generalized cost function, yielding an exceptionally low Mean Absolute Percentage Error (MAPE = 1.94 %). Simulating 16 distinct scenarios spanning four dimensions—economic growth, demographic shifts, railway policy, and road freight policy—from 2025 to 2040, our projections indicate that sectoral carbon emissions will peak around 2034. Crucially, our analysis reveals that the expansion of railway passenger capacity and a strategic reduction in road freight dependency serve as the central mechanisms through which this peak can be achieved. Contributions from economic growth and population decline exerted relatively modest impacts. Moreover, the analysis demonstrates that strategically integrating railway expansion with a reduction in road freight share constitutes the most effective pathway towards the sector's minimal feasible emission pathway.