서브메뉴
검색
Proactive System Optimal Routing of Network Traffic
Proactive System Optimal Routing of Network Traffic
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20250211152839
- ISBN
- 9798384450986
- DDC
- 628
- 저자명
- Ke, Zemian.
- 서명/저자
- Proactive System Optimal Routing of Network Traffic
- 발행사항
- [Sl] : Carnegie Mellon University, 2024
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- 형태사항
- 183 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 86-04, Section: A.
- 주기사항
- Advisor: Qian, Sean.
- 학위논문주기
- Thesis (Ph.D.)--Carnegie Mellon University, 2024.
- 초록/해제
- 요약On transportation networks, which are designed and constructed to serve people with efficient and reliable mobility, people are experiencing prolonged and unreliable travel time under expected recurrent congestion and unexpected events almost everywhere and every day. The defective mobility leads to significant economic losses and diminished quality of life. The core issue stems from the imbalance between the ever-growing travel demand and the constrained supply of transportation infrastructure. Additionally, the inherent uncertainties in demand and supply further complicate the management of transportation networks. This thesis addresses these challenges by proactively affecting the routing behaviors of a small portion of vehicles to improve network efficiency. The first question is how to fairly affect routing behaviors as system-optimal routes are generally not the shortest. We show the network state can be close to the system's optimal state by subsidizing a small portion vehicles (i.e., ride-hailing vehicles in our study) in a cost-effective way. The thesis investigates the impact of Transportation Network Companies (TNCs) on traffic networks and introduces the Optimal Ride-hailing Pricing (ORHP) scheme. ORHP is designed as a bi-level optimization problem where public agencies provide subsidies to TNCs, incentivizing them to adjust routing for their fleets in a way that improves overall network performance. This approach balances fleet cost minimization and service quality, leading to a win-win scenario for both public agencies and TNCs.Then, the thesis explores real-time system optimal routing under uncertainties. A novel reinforcement learning algorithm, TransRL, is developed to address the challenges posed by demand uncertainties and unknown system dynamics. TransRL integrates traditional physics-based traffic models with reinforcement learning. We also compare the traditional traffic model-based optimization and the model-free Reinforcement Learning (RL) to examine the suitability of these baseline methods under different uncertainty levels. More importantly, the experiment results indicate that the proposed TransRL outperforms existing methods and presents superior reliability and interpretable behaviors.To support informative traffic management, the next chapter of the thesis focuses on traffic prediction under recurrent and non-recurrent conditions. A Mixture of Experts (MoE) model is proposed to enhance prediction accuracy by separately modeling these distinct traffic patterns. By integrating multi-source data and employing specialized recurrent and non-recurrent expert models, the MoE framework significantly improves prediction performance. Furthermore, We provide valuable insights into the differences between recurrent and non-recurrent patterns by interpreting the predictions separately.Overall, this thesis presents a comprehensive approach to proactive traffic management, combining pricing strategies, real-time routing control, and advanced predictive modeling. The findings validate the potential of improving transportation mobility by real-time optimal routing even with practical constraints, uncertainties, and complexity of transportation systems.
- 일반주제명
- Transportation
- 기타저자
- Carnegie Mellon University Civil and Environmental Engineering
- 기본자료저록
- Dissertations Abstracts International. 86-04A.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008250123s2024 us c eng d■001000017164165
■00520250211152839
■006m o d
■007cr#unu||||||||
■020 ▼a9798384450986
■035 ▼a(MiAaPQ)AAI31562080
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a628
■1001 ▼aKe, Zemian.
■24510▼aProactive System Optimal Routing of Network Traffic
■260 ▼a[Sl]▼bCarnegie Mellon University▼c2024
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2024
■300 ▼a183 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 86-04, Section: A.
■500 ▼aAdvisor: Qian, Sean.
■5021 ▼aThesis (Ph.D.)--Carnegie Mellon University, 2024.
■520 ▼aOn transportation networks, which are designed and constructed to serve people with efficient and reliable mobility, people are experiencing prolonged and unreliable travel time under expected recurrent congestion and unexpected events almost everywhere and every day. The defective mobility leads to significant economic losses and diminished quality of life. The core issue stems from the imbalance between the ever-growing travel demand and the constrained supply of transportation infrastructure. Additionally, the inherent uncertainties in demand and supply further complicate the management of transportation networks. This thesis addresses these challenges by proactively affecting the routing behaviors of a small portion of vehicles to improve network efficiency. The first question is how to fairly affect routing behaviors as system-optimal routes are generally not the shortest. We show the network state can be close to the system's optimal state by subsidizing a small portion vehicles (i.e., ride-hailing vehicles in our study) in a cost-effective way. The thesis investigates the impact of Transportation Network Companies (TNCs) on traffic networks and introduces the Optimal Ride-hailing Pricing (ORHP) scheme. ORHP is designed as a bi-level optimization problem where public agencies provide subsidies to TNCs, incentivizing them to adjust routing for their fleets in a way that improves overall network performance. This approach balances fleet cost minimization and service quality, leading to a win-win scenario for both public agencies and TNCs.Then, the thesis explores real-time system optimal routing under uncertainties. A novel reinforcement learning algorithm, TransRL, is developed to address the challenges posed by demand uncertainties and unknown system dynamics. TransRL integrates traditional physics-based traffic models with reinforcement learning. We also compare the traditional traffic model-based optimization and the model-free Reinforcement Learning (RL) to examine the suitability of these baseline methods under different uncertainty levels. More importantly, the experiment results indicate that the proposed TransRL outperforms existing methods and presents superior reliability and interpretable behaviors.To support informative traffic management, the next chapter of the thesis focuses on traffic prediction under recurrent and non-recurrent conditions. A Mixture of Experts (MoE) model is proposed to enhance prediction accuracy by separately modeling these distinct traffic patterns. By integrating multi-source data and employing specialized recurrent and non-recurrent expert models, the MoE framework significantly improves prediction performance. Furthermore, We provide valuable insights into the differences between recurrent and non-recurrent patterns by interpreting the predictions separately.Overall, this thesis presents a comprehensive approach to proactive traffic management, combining pricing strategies, real-time routing control, and advanced predictive modeling. The findings validate the potential of improving transportation mobility by real-time optimal routing even with practical constraints, uncertainties, and complexity of transportation systems.
■590 ▼aSchool code: 0041.
■650 4▼aEnvironmental engineering
■650 4▼aTransportation
■653 ▼aTransportation networks
■653 ▼aTransportation Network Companies
■653 ▼aOptimal Ride-hailing Pricing
■690 ▼a0543
■690 ▼a0775
■690 ▼a0709
■71020▼aCarnegie Mellon University▼bCivil and Environmental Engineering.
■7730 ▼tDissertations Abstracts International▼g86-04A.
■790 ▼a0041
■791 ▼aPh.D.
■792 ▼a2024
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17164165▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
