Cross-city carbon emission estimation for electric mobility services: a transfer learning approach

Artikel i vetenskaplig tidskrift, 2026

The estimation of carbon emission reductions in electric mobility services remains challenging due to the scarcity of battery data and the heterogeneous characteristics of urban contexts. To achieve cross-city adaptation under data scarcity scenario, this study proposes a framework that integrates a spatiotemporal network with lightweight transfer learning. Experiment results show that the prediction model achieves superior prediction accuracy and robustness. With only 30 labeled samples, the model generalizes well across different cities and vehicle types. To further demonstrate the practical implementation of this framework, we conduct a case study of electric taxi service in Shenzhen, China. The findings indicate that the proposed framework provides a scalable approach for evaluating the decarbonization potential of electric mobility services, supporting sustainable urban transport planning.