Makroskopik Ağ Sistemlerinde Veri Odaklı Trafik Akışı Modellemesi
No Thumbnail Available
Date
2025
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Open Access Color
OpenAIRE Downloads
OpenAIRE Views
Abstract
Kentsel trafik sıkışıklığı, günümüz şehirleri için süregelen, karmaşık ve yüksek maliyetli bir problemdir. Artan seyahat süreleri, çevresel bozulma, enerji israfı ve ekonomik kayıplar bu problemin doğrudan sonuçları arasında yer almaktadır. Bu sorunlarla etkili şekilde başa çıkabilmek yalnızca altyapı yatırımlarıyla değil; aynı zamanda ulaşım politikaları, trafik yönetimi ve kontrol sistemlerinin bilimsel temellerle tasarlanmasıyla mümkündür. Bu kapsamda, trafiğin zaman ve mekân içinde nasıl evrildiğine dair sistematik ve ölçeklenebilir bir anlayış geliştirmek kritik önem taşır. Gerçek dünyada yapılacak deneyler genellikle maliyetli, zaman alıcı ve bozucudur. Bu nedenle, kentsel trafik sistemlerinin modellenmesi; alternatif senaryoların test edilmesi, politika etkilerinin değerlendirilmesi ve uzun vadeli sonuçların öngörülebilmesi açısından vazgeçilmez bir araçtır. Bununla birlikte, mevcut trafik modelleme yaklaşımları önemli sınırlılıklar taşır. Mikroskobik modeller bireysel araç davranışlarını yüksek ayrıntıyla temsil etse de, büyük ağlarda hesaplama açısından verimsizdir ve yoğun kalibrasyon verisi gerektirir. Makroskobik modeller ise daha hesaplıdır; ancak sabit başlangıç-varış (OD) akışları, homojen yol davranışları ve sürekli akış varsayımları gibi sadeleştirici kabuller içerir. Bu da onları karmaşık ve heterojen şehir yapıları için yetersiz kılar. Bu tez, trafik akışını yönlü bir ağda ayrık zamanlı yük alışverişiyle temsil eden veri odaklı bir makroskobik model önermektedir. Yol türlerine özgü akış dinamikleri, ağ topolojisi ve gözlemlenen trafik yoğunlukları modele entegre edilerek darboğazlar, geri tepme ve yük yeniden dağılımı gibi olgular temsil edilmektedir. Model parametreleri, evrimsel optimizasyon yoluyla, örtük talep varsayımı olmadan veriden öğrenilmektedir. Model, klasik Hücresel İletim Modeli (CTM) ile karşılaştırılmış; SUMO simülasyonları ve İstanbul, Londra ile New York verileri üzerinde üstünlük göstermiştir.
Urban traffic congestion remains a major challenge, causing increased travel times, environmental degradation, and economic losses. Addressing these issues requires accurate models that capture how traffic evolves across space and time. Since real-world experimentation is costly and disruptive, modeling offers a controlled and replicable way to test infrastructure interventions, transportation policies, and adaptive control strategies. However, existing modeling approaches face a trade-off between behavioral realism and computational scalability. Microscopic models simulate individual vehicles in detail but are computationally infeasible for large urban networks. In contrast, macroscopic models offer efficiency but often rely on simplifying assumptions—such as fixed origin–destination flows, homogeneous road behavior, or continuous flow dynamics—that reduce their realism in complex urban settings. This thesis presents a data-driven macroscopic traffic model that simulates congestion as a discrete-time load exchange process on a flow network. It captures nonlinear phenomena—such as bottlenecks, spillbacks, and adaptive load redistribution—by incorporating road-type–specific flow rules, observed traffic densities, and network topology. By accounting for the heterogeneous behavior of road classes like arterials, collectors, and residential streets, the model more accurately reflects real-world dynamics. Due to the system's nonlinear and adaptive nature, analytical solutions are intractable; instead, parameters are calibrated through evolutionary optimization using both synthetic and real data. To evaluate performance, the model is benchmarked against the classical Cell Transmission Model (CTM). Synthetic networks generated via the SUMO simulator are used to assess scalability and generalization. Real-world traffic datasets from Istanbul, London, and New York test predictive accuracy under diverse topologies and demand conditions. Results show that the proposed model consistently outperforms CTM in both accuracy and adaptability, offering a scalable and empirically grounded alternative for urban traffic modeling.
Urban traffic congestion remains a major challenge, causing increased travel times, environmental degradation, and economic losses. Addressing these issues requires accurate models that capture how traffic evolves across space and time. Since real-world experimentation is costly and disruptive, modeling offers a controlled and replicable way to test infrastructure interventions, transportation policies, and adaptive control strategies. However, existing modeling approaches face a trade-off between behavioral realism and computational scalability. Microscopic models simulate individual vehicles in detail but are computationally infeasible for large urban networks. In contrast, macroscopic models offer efficiency but often rely on simplifying assumptions—such as fixed origin–destination flows, homogeneous road behavior, or continuous flow dynamics—that reduce their realism in complex urban settings. This thesis presents a data-driven macroscopic traffic model that simulates congestion as a discrete-time load exchange process on a flow network. It captures nonlinear phenomena—such as bottlenecks, spillbacks, and adaptive load redistribution—by incorporating road-type–specific flow rules, observed traffic densities, and network topology. By accounting for the heterogeneous behavior of road classes like arterials, collectors, and residential streets, the model more accurately reflects real-world dynamics. Due to the system's nonlinear and adaptive nature, analytical solutions are intractable; instead, parameters are calibrated through evolutionary optimization using both synthetic and real data. To evaluate performance, the model is benchmarked against the classical Cell Transmission Model (CTM). Synthetic networks generated via the SUMO simulator are used to assess scalability and generalization. Real-world traffic datasets from Istanbul, London, and New York test predictive accuracy under diverse topologies and demand conditions. Results show that the proposed model consistently outperforms CTM in both accuracy and adaptability, offering a scalable and empirically grounded alternative for urban traffic modeling.
Description
Keywords
Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve Kontrol, Trafik, Ağ Trafiği, Doğrusal Olmayan Dinamik Sistemler ve Yöntemler, Sistem Simülasyonu, Şehiriçi Trafiği, Computer Engineering and Computer Science and Control, Traffic, Network Traffic, Nonlinear Dynamical Systems and Methods, Urban Traffic
Turkish CoHE Thesis Center URL
Fields of Science
Citation
WoS Q
Scopus Q
Source
Volume
Issue
Start Page
End Page
121
Collections
Google Scholar™
Sustainable Development Goals
5
GENDER EQUALITY

8
DECENT WORK AND ECONOMIC GROWTH

9
INDUSTRY, INNOVATION AND INFRASTRUCTURE

10
REDUCED INEQUALITIES

11
SUSTAINABLE CITIES AND COMMUNITIES
