Full metadata record
| DC Field | Value | Language |
|---|---|---|
| dc.contributor | Department of Electrical and Electronic Engineering | en_US |
| dc.contributor.advisor | Chung, Edward (EEE) | en_US |
| dc.contributor.advisor | Gu, Weihua (EEE) | en_US |
| dc.creator | Zou, Xiexin | - |
| dc.identifier.uri | https://theses.lib.polyu.edu.hk/handle/200/13088 | - |
| dc.language | English | en_US |
| dc.publisher | Hong Kong Polytechnic University | en_US |
| dc.rights | All rights reserved | en_US |
| dc.title | Traffic prediction and transfer learning of transport system performance using deep learning | en_US |
| dcterms.abstract | Forecasting traffic volume is crucial for effective traffic control, as it allows traffic flow to be regulated in advance, thereby alleviating congestion. One highly cost-effective means of regulating traffic flow is traffic signal control at intersections. The optimal timing plan can be obtained by comparing intersection performance under different signal plans to minimize vehicle delay and relieve traffic congestion. | en_US |
| dcterms.abstract | Network-level traffic volume prediction and detector failure have rarely been considered in the literature. Chapter 2 proposes a framework based on long short-term memory and the multilayer perceptron that can predict network-level traffic volumes even with detector failure. The proposed method utilizes a deep profile model to learn the detector’s signature and group detectors with similar signatures. A deep predictive model is proposed for each cluster, and failed detectors can refer to similar ones within the cluster. The validation using three cities’ Japan Road Traffic Information Center data demonstrates the method’s effectiveness during typical and atypical days (the COVID-19 lockdown period and the 2021 Tokyo Olympics). Further, it considers detector reliability: the increase in mean absolute error is less than 1 veh/5 min when the probability of detector failure increases to 20%. | en_US |
| dcterms.abstract | While the proposed method in Chapter 2 performs well when data is sufficient, building deep models for extracting detector signatures and prediction is challenging when data is limited. To address this, Chapter 3 proposes a method that also utilizes deep models and clustering algorithms but introduces few-shot learning and transfer learning for learning when data is scarce. Validation using the PeMS dataset demonstrates the method’s ability to provide accurate traffic flow, speed, and occupancy forecasts while accounting for detector reliability. | en_US |
| dcterms.abstract | After knowing traffic volumes, regulating traffic flow through traffic signals at intersections is a proactive approach to alleviate traffic congestion. However, to achieve optimal and effective signal timing, it is essential to have a reliable performance indicator that can evaluate the effectiveness of different signal timing plans based on the volumes of each movement. Chapter 4 presents a deep learning-based method to estimate the average approach delay at isolated signalized intersections. The versatile model also can predict various performance variables, such as degree of saturation(DoS). Considering the correlation between delay and DoS, feature integration is applied to integrate learned DoS features to improve delay prediction. Moreover, transfer learning is introduced to improve delay prediction, making the proposed DL model suitable for data-rich and data-scarce intersections. The proposed method is validated on intersections generated by SIDRA and SUMO, and the results demonstrate that the proposed method achieves high accuracy in predicting delay. Additionally, the proposed method offers fast intersection performance feedback without requiring vehicle trajectory information and can serve as a surrogate model of a micro-simulator for signal timing optimization tasks. | en_US |
| dcterms.abstract | This thesis presents deep learning methods for traffic prediction and transport system performance. The proposed approach can accurately predict network-level traffic volume and quantitatively assess intersection performance. The performance value obtained guides the signal control center in determining optimal signal timing for different traffic volumes, improving traffic flow efficiency. | en_US |
| dcterms.extent | xii, 103 pages : color illustrations | en_US |
| dcterms.isPartOf | PolyU Electronic Theses | en_US |
| dcterms.issued | 2024 | en_US |
| dcterms.educationalLevel | Ph.D. | en_US |
| dcterms.educationalLevel | All Doctorate | en_US |
| dcterms.LCSH | Traffic estimation -- Data processing | en_US |
| dcterms.LCSH | Roads -- Interchanges and intersections | en_US |
| dcterms.LCSH | Traffic signs and signals | en_US |
| dcterms.LCSH | Deep learning (Machine learning) | en_US |
| dcterms.LCSH | Hong Kong Polytechnic University -- Dissertations | en_US |
| dcterms.accessRights | open access | en_US |
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/200/13088