Fast estimation and pattern discovery of dynamic vehicle emissions on road networks based on the traffic performance index: A case study of Beijing

Abstract

The estimation of dynamic emissions from mobile sources such as vehicles presents a major challenge in air quality modeling. Current studies indicate strong positive correlations between traffic congestion and emission levels. However, there is a notable lack of research defining the relationship quantitatively. The dynamic and rapid estimation of total emissions from road networks holds critical importance using real-time Traffic Performance Indices (TPIs). This paper focuses on rapidly estimating and identifying patterns in dynamic vehicle emissions across road networks by leveraging TPIs. At first, relationships between vehicle emissions and TPIs are rigorously examined. Subsequently, a quantitative sine function model is developed to characterize the relationship. Finally, employing the SOM neural network algorithm, temporal emission patterns are investigated, and the impact of the COVID-19 pandemic on emissions is assessed. The results demonstrate: 1) Non-linear positive correlations exist between total emissions and TPIs with emissions escalating with increasing TPI; 2) The proposed sine function model predicts total network emissions with average relative errors of 9.26 % for NO_x, 9.04 % for HC, 8.86 % for CO₂, and 9.02 % for CO; 3) Utilizing the Self-Organizing Map (SOM) neural network clustering algorithm identifies eight emission variation patterns including Holydays, Saturdays, Sundays, Workdays during vacations, Mondays, Fridays, Ordinary workdays, and Congested workdays, which effectively represent 91.54 % of emission scenarios across the network. The findings are expected to broaden TPI applications while providing robust scientific support for integrated policies targeting congestion management and pollution control.