Carbon Emission Forecasting for Urban Road Traffic at Topological Level Based on Improved Informer

The continuously increasing carbon emission from road traffic has a negative impact on the global climate, mak-ing its forecasting a hotspot in intelligent transportation research. There are complex spatiotemporal characteristics and multidimensional features in road traffic network data, which cannot be effectively processed by previous forecasting models. To address the problem, this paper proposes TSD-Informer as the forecasting model. Considering the distraction of attention on temporal dependency, the input mapping method is changed from feature dimension to time dimension. To capture the spatial characteristics of the road traffic network, the spatial correlation is enhanced through a multi-source information fusion gate. The coupling relationship and contribution of muti-source data is studied by introducing the Gated Residual Network. Eventually, an improved Informer model is developed to forecasting the carbon emission for urban road traffic at topological level. The experimental results show that each strategy is effective and compatible with each other. The improved Informer model outperforms Informer model by an average reduction of 27.47%, and 21.38% in the root mean squared error and the mean absolute error with the best R2 closest to 1. All evaluation indicators are superior to the current mainstream forecasting models, which proves the superiority of the proposed TSD-Informer model.