| Author: | Li, Ang |
| Title: | A unified framework for path travel time prediction using heterogeneous traffic data and weather information |
| Advisors: | Lam, H. K. William (CEE) |
| Degree: | Ph.D. |
| Year: | 2025 |
| Subject: | Route choice -- Data processing Transportation -- Data processing Weather forecasting Traffic flow Hong Kong Polytechnic University -- Dissertations |
| Department: | Department of Civil and Environmental Engineering |
| Pages: | 1 volume (various pagings) : color illustrations |
| Language: | English |
| Abstract: | Advanced Traveler Information Systems (ATIS) usually offer information on travel times for specific paths or routes. They aim to impart timely information to road users and assist them in making route choices under uncertainties in the near future, especially in road networks with frequent adverse weather conditions. Recent research has explored the use of diverse traffic data sources for predicting path travel times in the current and future time intervals. These traffic data include both real-time and historical data from different sources, in which the former is collected on the current day, and the latter is gathered before the current day. There are three challenges integrating them and relevant weather information for path travel time prediction in the current and future time intervals. First, some traffic data are sampled at high frequency (say, once every 1 or 2 minutes) due to the requirement of practical applications in ATIS. Consequently, the sample size per time interval is insufficient to provide reliable information for removing outliers from real-time data. Moreover, ground truth on path travel times is difficult to collect with high cost and limited samples for field surveys (e.g., floating car surveys). Collecting these ground truths is more suitable for validation than model training. Second, existing ATIS generally disseminate the predicted average path travel times in the current time intervals for all vehicle classes in reality. However, the observed path travel times of a significant proportion of vehicles (i.e., private cars) may deviate substantially from the average path travel times. It is specifically true when many other vehicles (e.g., buses and goods vehicles) travel with private cars on the same road. There is a need to integrate traffic data to predict multi-class path travel times. Additionally, different traffic sensors may furnish heterogeneous traffic data (e.g., travel time, flow, speed, etc.), which complicates the path travel time prediction for different vehicle classes. Third, in cities with frequent rainfall, the rainfall intensity can significantly impact the accuracy of travel time predictions. Existing studies have used historical rainfall intensity data to predict path travel times. However, previous studies may not fully consider the temporal relationships between rainfall intensity data and predicted path travel times. Moreover, less attention has been given to the weather forecast information, which can be further investigated as adverse weather can affect the travel behavior of road users (e.g., departure time and route choices). Addressing the usage of weather information is crucial for improving the performance of ATIS on path travel time prediction under varying traffic and weather conditions. Based on the above challenges, this thesis seeks to propose a unified framework for path travel time prediction in ATIS, offering the following three key contributions: Firstly, the proposed unsupervised algorithm is designed to filter limited real-time automatic vehicle identification (AVI) data without relying on ground truth for training. Real-time AVI data can be limited due to the high frequency of collection. Contrastingly, historical data contain adequate information on variations of path travel times on each time interval. This type of variation helps to indicate the typical traffic conditions by different time of day. Therefore, the proposed unsupervised algorithm goes beyond traditional filtering methods (relying purely on real-time AVI data) by incorporating day-to-day variations of path travel times. It consequently offers valuable insights for data filtering, particularly when real-time AVI data is limited. The second contribution involves the development of a novel model for multi-class path travel time prediction in the current time interval. The proposed prediction model effectively utilizes heterogeneous traffic data from various types of traffic sensors. This prediction model incorporates the temporal relationships of path travel times across different vehicle classes inferred from multi-source traffic data. It allows for the fusion of traffic information from diverse traffic data sources. As a result, the proposed prediction model can provide satisfactory predicted path travel times by vehicle class in the current time interval. The third contribution arises with a new model that considers weather information to predict path travel times in future time intervals. This thesis proposes a modeling framework to further capture the relationship between predicted path travel times and weather information. Therefore, the proposed modeling framework can help describe the dynamics of predicted path travel times under future rainy conditions. Additionally, the proposed modeling framework distinguishes the effects of weather information under different traffic conditions and various rainfall categories. Hence, it can ultimately enhance the prediction accuracy. The empirical evidence from real-world traffic data in Hong Kong has demonstrated the effectiveness of the proposed unified framework for path travel time prediction. Three key contributions have been justified with corresponding case studies or numerical experiments in this thesis. Firstly, the case study conducts sensitivity tests using different sampling rates of AVI data. It reveals that the proposed unsupervised algorithm robustly surpasses the existing filtering algorithms without using ground truth for training. The second contribution is confirmed using multiple sources of traffic data gathered on an urban expressway in Hong Kong. It shows that the prediction accuracy of path travel times by vehicle class in the current time interval is significantly improved when a proper combination of data sources is selected for training. The proposed prediction model can output the multi-class path travel times with satisfactory performance. Lastly, the empirical tests illustrate that the proposed modeling framework, considering the weather information, achieves a higher accuracy of predicted path travel times in future time intervals. It outperforms the other benchmarks on a dataset collected in Hong Kong, a city with abundant rainfall throughout the year. These three significant contributions in this thesis are properly justified to support the proposed unified framework as a valuable platform for further research in the development of various ATIS. |
| Rights: | All rights reserved |
| Access: | open access |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/13562