Ship Pollutant Emission Prediction and Emission Reduction Path Analysis Based on Transformer: A Case Study of Guangzhou Port

With the steady expansion of global maritime trade, pollutant emissions from the shipping industry have become an increasingly urgent environmental concern. Accurate forecasting of ship emissions and understanding pollutant concentration dynamics in port areas are vital for proactive environmental management. In this study, we integrate Automatic Identification System (AIS) data, Lloyd's Register ship data, and a global marine engine database. Following comprehensive data cleaning and enhancement, a dynamic bottom-up approach is applied to construct a high-resolution ship emission inventory, which provides a solid foundation for predictive modeling. We employ the Flowformer model for emission forecasting and compare its performance against several baseline models. To further improve predictive accuracy, Mamba blocks are incorporated into the model architecture, enabling better temporal modeling and feature representation. Experiments conducted on real-world ship data from Guangzhou Port show that the Flowformer model achieves higher accuracy and stronger stability in capturing emission trends compared to other approaches. The Flowformer model with Mamba integration further enhances long-term prediction performance. The findings confirm that our method can support port managers in making data-driven decisions, such as scheduling ships and implementing timely environmental policies. This study demonstrates the practical value of combining high-quality emission inventories with advanced deep learning techniques, offering a scalable and effective solution for port-level air pollution prediction and management.