PM_2.5 concentration prediction algorithm integrating traffic congestion index.

Area of Science:

• Environmental Science
• Traffic Engineering
• Air Quality Monitoring

Background:

• Particulate matter (PM2.5) concentrations are influenced by complex traffic conditions.
• Accurate prediction of PM2.5 requires incorporating traffic dynamics.
• Existing models may not fully capture the impact of traffic congestion on air pollution.

Purpose of the Study:

• To propose and evaluate a novel strategy using the congestion index as an input feature for PM2.5 concentration prediction.
• To assess the effectiveness of integrating traffic congestion data into air quality models.
• To improve the accuracy of PM2.5 forecasting, especially during peak traffic hours.

Main Methods:

• Developed a multi-scale input model incorporating a congestion index.
• Conducted ablation experiments to validate the contribution of the congestion index.
• Compared the performance of the proposed model against conventional benchmark models.

Main Results:

• The proposed strategy significantly reduced the root mean square error (RMSE) of benchmark models by over 6.07% on average.
• The best-performing model achieved a 12.06% reduction in RMSE.
• Maintained RMSE below 9.83 µg/m³ during peak hours, even with high traffic emissions, effectively addressing pollution hotspots.

Conclusions:

• The congestion index is a valuable input feature for enhancing PM2.5 prediction models.
• The multi-scale input model demonstrates superior performance in capturing the effects of traffic on air quality.
• This approach offers a promising strategy for improving urban air quality management and public health initiatives.