MVCAR: Multi-View Collaborative Graph Network for Private Car Carbon Emission Prediction

Chenxi Liu, Zhu Xiao, Cheng Long, Dong Wang, Tao Li, Hongbo Jiang

Abstract: As urbanization accelerates, the rise in private car usage has become a double-edged sword, symbolizing economic growth while exacerbating urban air pollution due to increased carbon emissions. This paper studies the problem of carbon emission prediction of private cars in urban environments, enabling effective carbon emissions reduction and energy conservation guidance. Existing methods struggle with costly carbon emission collection and rely on precise emission factors, inaccuracies in modeling spatial similarities across urban regions, and complexities in modeling global temporal variations. To solve these issues, the Multi-View CollAboRative graph network (MVCAR) framework is proposed for private car carbon emission prediction. MVCAR employs a trajectory-based method to estimate carbon emissions from private car mobility to represent the spatial-temporal carbon emissions effectively. To capture the geo-spatial and semantic regional similarities of the carbon emissions, MVCAR constructs multi-view graphs and utilizes multi-view graph convolution networks. Furthermore, MVCAR integrates collaborative gated recurrent networks to model the spatial-temporal correlations of carbon emissions. The collaborative gated recurrent networks include a multi-view gated recurrent unit (GRU) and a sequential GRU. The multi-view GRU models the multiple spatial-temporal correlations of carbon emissions. A learnable temporal module embeds various temporal features, and further feeds these features into sequential GRU to capture the temporal variations of carbon emissions. Finally, a collaborative strategy that synergistically combines multi-view and sequential GRUs through stacked training. Extensive experiments on real datasets demonstrate the superiority of the proposed MVCAR.