Evaluation of Air Pollutants in Extra-Long Road Tunnel with the Combination of Pollutants Nonlinear Evolution and Machine Learning Method

The extra-long expressway tunnel has a high socio-economic effect on inter-regional development, with high traffic and strong traffic winds. Nevertheless, the impacts of the tunnel traffic volume on pollutant evolution are rarely considered. This study conducted a field measurement in a real-world extra-long highway tunnel for 578 days. For the first time, the nonlinear dynamics of traffic pollutants (CO, VOCs, NO2, PM2.5, PM10) were analyzed using the Multifractal Detrended Fluctuation Analysis approach. Using the Random Forest model, the impacts of traffic and environmental parameters on air quality were quantified. The findings indicated that COVID-19 had a considerable impact on tunnel traffic, although the variance in pollutant concentration was not very noteworthy. The bidirectional effect of traffic was the main reason for this phenomenon. The Canonical Correlation Analysis was unable to quantify the correlation between pollutants and environmental parameters. The pollutant concentration evolution has a steady power-law distribution structure. Further, an inverse Random Forest model was proposed to predict air pollutants. Compared with other prediction models (baseline and machine learning), the proposed model provided higher goodness of fit and lower prediction error, and the prediction accuracy was higher under the semi-enclosed structure of the tunnel. The relative deviations between the predictions and measured data are less than 5%. These findings ascertain the nonlinear evolutionary mechanisms of pollutants inside the expressway tunnel, thus eventually improving tunnel environmental sustainability. The data in this paper can be used to clarify the changes in the traffic environment under the COVID-19 lockdown.

Automated summary

This study conducted a field measurement in a real-world extra-long highway tunnel for 578 days and analyzed the nonlinear dynamics of traffic pollutants using the Multifractal Detrended Fluctuation Analysis approach. The impacts of traffic and environmental parameters on air quality were quantified using the Random Forest model. The findings indicated that COVID-19 had a considerable impact on tunnel traffic and the bidirectional effect of traffic was the main reason for this phenomenon. An inverse Random Forest model was proposed to predict air pollutants, which showed higher goodness of fit and lower prediction error.