Investigation on characteristics of polluting particulate matter at urban traffic intersections in Hong Kong

Abstract

The aim of this thesis is to investigate the variation in polluting particulate matter (PM) at urban traffic intersections. At the intersections, vehicles stop frequently with idling engines during the red-light periods and speed up rapidly during the green-light periods, which generally produce severe pollution than that in other situations. Meanwhile, the pedestrians are just exposed to high levels of particulate pollutants as they walk-by or cross the zebra areas. The inhaled particulates may affect the functions of hearts and lungs of human beings and cause adverse health effects. With these considerations, a detailed investigation on characteristics of PM variation at urban traffic intersections in Hong Kong is carried out and reported in this thesis. The main contents of the thesis are as follows. I. Measurements at a typical traffic intersection in Hong Kong In the experiments, particles with the diameters larger than 0.3μm were detected and classified into six categories. The traffic conditions were recorded by a digital camera while the meteorological conditions, such as wind speed, temperature and relative humidity, were simultaneously measured. To acquire the representative datasets, the measurements were performed for one and half hours every time and lasted for one week in spring and winter seasons, respectively. Within multi-day measurements, the data about the PM levels, traffic conditions, and meteorological conditions were collected as the original database. II. Statistical analysis of PM variation Based on the measurements, statistical analysis was conducted to explore the characteristics of PM at intersections. The results show that the majority of measured particles vary periodically with traffic signal intervals. In particular, the variation in particles within the range of 0.5-5μm appears to correspond with the traffic signal periods, implying close correlation between them. The auto-correlation function was used to reveal the periodic behaviour and the results indicate that the variation in PM in afternoon exhibits a more pronounced periodic behaviour than that in morning, especially for particles within the range of 0.5~5μm. The stable variations in measured data in the red-light periods were also selected to seek an appropriate statistical distribution via the goodness-of-fit test. Lognormal distribution was recognized as the most suitable one for particles smaller than 5μm. III. Impacts of traffic conditions and meteorological conditions on PM variation The multivariate statistical techniques of cluster analysis and principal component analysis (PCA) were applied to identify the relationships among PM concentrations, traffic conditions, and meteorological conditions in two respects. One was to clarify the relationships between the instantaneous PM concentrations and corresponding meteorological variables such as temperature, relative humidity, and wind speed on two days in different seasons. The other was to explore the parallel relationships among the average PM concentrations during green-light periods with traffic conditions and meteorological variables over one week in different seasons. Through analyses, a strong relationship between particles within the range of 0.5-5μm and the diesel vehicle count was identified, and these particles were thus concluded to be mainly related to the number of diesel vehicles. Meteorological conditions, particularly wind speed, were found to have a significant influence on the number concentrations of particles larger than 5μm. The results also revealed that the number concentrations of ultrafine particles within the 0.3-0.5μm range were strongly dependent on the relative humidity. IV. Prediction of PM concentrations using multilayer perception (MLP) model with assistance of PCA In view of the nonlinear relationships among the PM concentrations, traffic conditions, and meteorological conditions, the PM concentrations were predicted using the nonlinear approach of MLP model with three algorithms, i.e., the Levenberg-Marquardt (LM) algorithm, particle swarm optimization (PSO) algorithm, and PSO algorithm with chaotic mapping (CPSO). Through comparing the obtained results, the CPSO algorithm was shown to be more effective than the others in both predictive capability and precision. It might be due to the fact that the MLP-CPSO model combines the merits of PSO and chaos, with the former overcoming the over-fitting problem and the latter avoiding the model becoming trapped in local optima. In addition, PCA was employed to generate principal components (PCs), rather than the original data, as the input variables to reduce the model complexity and eliminate data collinearity. The performance of the models using PCs as the input variables was shown better than that of those using the original data. The MLP-PC model trained with the CPSO algorithm again outperformed the model trained with the other two algorithms. V. Prediction of PM concentrations using semi-empirical box model with instantaneous velocity and acceleration Apart from the indirect estimating approach of the MLP model, a direct approach of semi-empirical box model was applied to predict the PM concentrations from vehicle emission. In order to minimize the influence of meteorological conditions, the application of the model was firstly carried out in a sunny day and then further testified between two sunny days. From the evaluation and predication results, the semi-empirical model with instantaneous vehicle velocity and acceleration performed well in all cases, although certain values were overestimated or underestimated. The motive of using instantaneous vehicle velocity and acceleration in semi-empirical box model was to attempt to describe the traffic flow and calculate the vehicle emission with a more realistic representation from microscopic point of view. Through application of the improved model in two situations, the attempt was proved to be significant and informative. It was shown to be a practical alternative approach for evaluating the PM dispersion at urban traffic intersections with a reasonable accuracy.