Job title:
Physics-Informed Machine Learning and Multi-regional Macroscopic Approach for Traffic Dynamics Modeling
Company:
Job description
Offer DescriptionMain host Laboratory: COSYS-GRETTIAMain location: Paris area, FranceDoctoral affiliation: UNIVERSITE GUSTAVE EIFFELPhD school: MATHEMATIQUES ET SCIENCES ET TECHNOLOGIES DE L’INFORMATION ET DE LA COMMUNICATION (MSTIC)BackgroundThe increasing pressure on urban transportation systems requires innovative solutions to enhance their efficiency. Traffic congestion, particularly in large urban areas, has led to the development of advanced macroscopic models for analyzing and predicting traffic dynamics [1]. One such model, the generalized bathtub model, provides a macroscopic approach to traffic dynamics by representing traffic flow with partial differential equations (PDEs). This model captures the evolution of traffic flow and density over time and space, offering a versatile and robust framework for traffic analysis [2]. The advantages of the generalized bathtub model include its ability to handle complex traffic dynamics while maintaining a high level of computational efficiency, making it well-suited for real-time traffic management applications. The generalized bathtub model has proven effective in handling traffic dynamics for a single region [3]. However, calibrating the parameters and defining network clusters for multi-regional networks to address the heterogeneity of traffic propagation at large-scale networks remains challenging. Furthermore, the literature has not yet extensively addressed multi-regional generalized bathtub models, highlighting a potential area for improvement and research. Although [3] and [4] have addressed the heterogeneity of traffic demand, the literature remains silent on addressing the heterogeneity of supply in multi-regional generalized bathtub models.Physics-Informed Machine Learning (PIML) models, which combine knowledge of physical systems with machine learning techniques, have shown great promise in improving the accuracy and efficiency of dynamic traffic modeling [5]. Few recent studies consider PIML for traffic estimation [6] and parameter control [7]. For example, [7] proposed a multi-agent deep reinforcement learning approach for scalable multi-region perimeter metering control, which demonstrated the effectiveness of integrating clustering methods in traffic management strategies. This provides motivation for extending dynamic macroscopic models, in particular generalized bathtub model, to multi-regional networks by integrating clustering methods, which allows for a more comprehensive understanding of traffic patterns and enables better traffic management strategies.ObjectivesThe main goal of this research is to develop a hybrid model that combines dynamic clustering and PIML for traffic dynamics modeling and estimation in multi-regional macroscopic networks. This novel approach aims to improve the accuracy and computational efficiency of traffic simulations, offering a robust tool for traffic planners and policymakers to design more efficient and sustainable transportation systems. Consequently, this Ph.D. thesis will try to address the following research questions:
- How can we perform the segmentation process over the network to have a good structure for parameter estimation in macroscopic models?
- What is the trade-off between computation time and accuracy of the model (number of clusters vs objective function of calibration)?
- How can we build a PIML model for multi-regional networks, including the incorporation of PDEs in the machine learning process?
- Which preprocessing steps are necessary for macroscopic models, and how can we validate them?
MethodologyTo accomplish the research objectives, the proposed plan entails the following primary steps: (i) Review and compare existing traffic dynamic models and Physics-Informed Machine Learning approaches; (ii) Modify existing methodologies or develop a new hybrid model incorporating dynamic clustering and PIML for multi-regional network analysis; (iii) Evaluate and validate the newly developed model in terms of accuracy, computational efficiency, and applicability in various traffic scenarios.Existing toolsGRETTIA has an extensive background in traffic dynamics modeling and macroscopic approaches [3][4], including the application of Machine Learning (ML) models to traffic. This research will build upon the group’s prior work in clustering methods [8] and macroscopic models and extend it by incorporating dynamic clustering and multi-regional analysis. The group possesses recognized expertise in data-driven approaches applied to urban mobility [9]. In addition, Prof. Jean-Patrick Lebacque will participate in the supervision of this Ph.D. thesis as an expert in dynamic traffic models. This thesis aims to leverage the group’s advancements to propose a novel methodological framework that integrates operational research, machine learning, and dynamic clustering for PIML in multi-regional macroscopic approaches.In particular, the research will draw from the work of [3] on generalized bathtub models for traffic dynamics and the research conducted on clustering approaches for dealing with contiguity constraints. By integrating these diverse areas of expertise, the thesis will develop a comprehensive solution for traffic dynamics modeling using PIML and dynamic clustering.References[1] Ayed, I., de Bézenac, E., Pajot, A., & Gallinari, P. (2020). Modelling spatiotemporal dynamics from Earth observation data with neural differential equations. Journal of Artificial Intelligence Research, 70, 1-28.[2] Jin, W. L. (2020). Generalized bathtub model of network trip flows. Transportation Research Part B: Methodological, 136, 138-157.[3] Ameli, M., Faradonbeh, M. S. S., Lebacque, J. P., Abouee-Mehrizi, H., & Leclercq, L. (2022). Departure time choice models in urban transportation systems based on mean field games. Transportation Science, 56(6), 1483-1504.[4] Lebacque, J. P., Ameli, M., & Leclercq, L. (2022, June). Stochastic departure time user equilibrium with heterogeneous trip profile. In The 10th symposium of the European Association for Research in Transportation (hEART).[5] Pereira, M., Lang, A., & Kulcsár, B. (2022). Short-term traffic prediction using physics-aware neural networks. Transportation research part C: emerging technologies, 142, 103772.[6] Zhou, D., & Gayah, V. V. (2021). Scalable multi-region perimeter metering control for urban networks: A multi-agent deep reinforcement learning approach. Transportation Research Part C: Emerging Technologies, 127, 103066.[7] Usama, M., Ma, R., Hart, J., & Wojcik, M. (2022). Physics-Informed Neural Networks (PINNs)-Based Traffic State Estimation: An Application to Traffic Network. Algorithms, 15(12), 447.[8] Côme, E. (2023), Bayesian contiguity constrained clustering, spanning trees and dendrograms, arxiv:2302.12546 stat.CO[9] Laharotte, P-A, Billot, R., Côme, E., Oukhellou, L., Nantes, A., El Faouzi, N (2014) Spatiotemporal analysis of bluetooth data: Application to a large urban network IEEE Transactions on Intelligent Transportation Systems, 16(3), 1439-1448Funding category: Contrat doctoralPHD Country: FranceWhere to apply WebsiteRequirementsSpecific RequirementsThe candidate must:
1) Have a Master 2 or equivalent in Machine Learning, AI, Applied mathematics, mathematics, transportation engineering, computer science, operations research or other field strongly related to Applied mathematics.
2) Have excellent analytical and communication skills in written and spoken English.
3) Be able to work independently and take responsibility for the progress and quality of the project.
4) Have experience in traffic data collection, statistical data analysis and exploration, and geospatial data analysis.
5) Have excellent programming skills.Additional InformationWork Location(s)Number of offers available 1 Company/Institute Université Gustave Eiffel / IFSTTAR – Site de Marne-la-Vallée Country France City champs sur marne GeofieldContact WebsiteSTATUS: EXPIREDShare this page
Expected salary
Location
Champs-sur-Marne, Seine-et-Marne
Job date
Sat, 08 Jun 2024 07:26:36 GMT
To help us track our recruitment effort, please indicate in your email/cover letter where (vacanciesineu.com) you saw this job posting.
