Slide background

SINGLE_NEW

Νέα και Ανακοινώσεις

Ομιλία από τον καθηγητή Mehdi Keyvan Ekbatani, University of Canterbury, New Zealand

Τρίτη 6 Ιουνίου 2023, στις 10:30, στην αίθουσα Γ3.1.14

 

Στο πλαίσιο του Workshop “Lane-free Traffic”, που οργανώνεται από το Εργαστήριο Δυναμικών Συστημάτων και Προσομοίωση (ΜΠΔ) με την συμμετοχή του Technical University of Munich, θα δοθεί, ανοιχτή στο κοινό, ομιλία από τον καθηγητή Mehdi Keyvan Ekbatani, Civil & Natural Resources Engineering at the University of Canterbury, την Τρίτη 6 Ιουνίου 2023 στις 10:30 στην αίθουσα Γ3.1.14.

Title: Data-driven Approaches for Traffic Monitoring, Modelling and Management in Urban and Motorway Networks

Abstract: The emergence of novel data-driven methods has brought new perspectives in the field of data analysis. Data-driven methods have shown promising results in various fields of transportation engineering, such as a network-wide traffic forecast, short-term traffic, and driving behaviour modeling. The implementation of any comprehensive surveillance method for dynamic estimation of traffic characteristics in a large-scale urban network requires a considerable amount of information. While the application of stationary sensors (e.g. inductive loop detectors, microwave detectors, and plate recognition cameras) for collecting traffic data is always limited by coverage area and high costs, the rapid evolution in information technology has brought new opportunities for access to various sources of data. In view of high computational requirements and technical shortcomings, online estimation of network-wide traffic dynamics (the so-called Network Fundamental Diagram) by floating car data (FCD) faces difficulties in terms of real-time feasibility. Moreover, monitoring traffic network conditions with stationary measurements (e.g. loop detectors) is only possible through a small subset of links. Hence, the possibility of estimating accurate NMFD at a lower cost through a limited number of loop detectors in the network (i.e. reduced operational NMFD estimation) has been under the spotlight of the research community. However, the application of the proposed methods seeking to find optimal locations of loop detectors faces obstacles because of the reliance on specific assumptions such as using priory known loop detector data (LDD) or overlooking the role played by the unloading period (i.e. network recovery).

Considering the potential impact of distinctive characteristics like network properties (e.g. topological features) and traffic characteristics on the shape of NMFD, a statistical solution algorithm is deployed to estimate operational NMFD. The algorithm concerns both loading and unloading periods simultaneously and sheds light on the influential features of each period. It is been shown that the proposed method works efficiently even with the presence of a limited number of loop detectors across the network (i.e. 1% loop detectors' coverage rate). Furthermore, the robustness of the proposed method against the network stochasticity and also the feasibility of the model in low penetration rates of FCD (e.g. 3%) are other indicators of the model's applicability to real-world scenarios.

Network-wide perimeter control strategies have been shown promise in recent years. These perimeter control strategies are mostly based on networks with fixed boundaries. However, fixed partitions may not exploit the full potential of control performance when traffic condition dynamically changes. In this study, a dynamic partitioning has been integrated into the prediction model to formulate an optimization, in which optimal traffic performance is achieved by iteratively finding the appropriate partitions as well as the control decision variables. Moreover, a Deep Learning (DL) based estimator is incorporated into the framework to address the hefty computational burden caused by the nesting of optimization loops. In the hybrid scheme, the DL-based prediction replaces the model-based prediction if a confidence condition can be fulfilled.

In the era of big data, the connected vehicle (CV) can collect massive naturalistic driving data of itself and surrounding vehicles. The CV data are available at a larger scale than ever before, which can be leveraged to extract detailed and useful information on Lane Chane behaviors. Meanwhile, data-driven methods have been increasingly used and promoted the development of LC studying. The model performances are improved steadily with the advancement of approaches, especially the latest deep learning (DL) algorithms. The emerging CV data and advanced DL methods provide new opportunities in LC research. This study aims at using CV data and DL algorithms for LC modeling.

Bio: A/Prof. Mehdi Keyvan-Ekbatani’s main research area is Traffic Flow Theory, Urban Traffic Control, Driving Behaviour Modelling and Intelligent Transportation Systems.  He completed his graduate studies in the field of transportation engineering at the Sharif University of Technology, Iran, in 2010. He received the Ph.D. degree in urban traffic control from the Dynamic Systems and Simulation Laboratory (DSSL), Technical University of Crete, Greece. He was a Post-Doctoral Researcher with the Department of Transport and Planning, Delft University of Technology, The Netherlands, from 2014 to 2017. Since 2017 he is with the University of Canterbury (UC), New Zealand. He established the Complex Transport Systems Laboratory (CTSLAB) at UC. He was a recipient of the Best Paper Award at the European Transportation Research Arena 2012 and the 2014 IEEE ITSS Best Ph.D. Dissertation Award. In 2021, he received the Best Emerging Researcher Award from the Faculty of Engineering, University of Canterbury.

 

© Σχολή Μηχανικών Παραγωγής & Διοίκησης  Attribution-NonCommercial-ShareAlike 4.0 International Licence