Collaborative carbon emission peak actions in urban agglomerations: multi-agent reinforcement learning analysis of the urban agglomerations of Beijing-Tianjin-Hebei and Yangtze River Delta

Coordinating carbon emission reduction strategies across cities helps mitigate the growth of carbon emissions and promote the achievement of carbon peaking and carbon neutrality goals. To optimize collaborative strategies across cities with diverse economic and resource contexts, this study proposes a multi-agent reinforcement learning-based approach. Acting as an independent agent, each city uses a transformer-based model to correlate its socioeconomic actions with its net carbon emissions. The socioeconomic actions of cities within an urban agglomeration are coordinated using the QMIX algorithm, which employs value function factorization to integrate local decisions and feedback on reduced emissions. This approach promotes collaborative policy optimization across cities and is validated by comparing collaborative and baseline development scenarios in the Beijing-Tianjin-Hebei and Yangtze River Delta urban agglomerations. This study demonstrates that collaborative development scenarios can effectively reduce the net carbon emissions of urban agglomerations at their peaks or advance their peak times. Our findings enrich the methodological basis for the effective promotion of regional carbon emissions reduction and provide policy references for the low-carbon development of these two urban agglomerations. © 2025 Wuhan University.