Loading...

A Neural Network Model Considering Travel Flow Stability and Graph Features of the Network for Origin-Destination Matrix Prediction under Limited Data Conditions

Alitabar Malekshah, Reza | 2025

167 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 58230 (09)
  4. University: Sharif University of Technology
  5. Department: Civil Engineering
  6. Advisor(s): Amini, Zahra
  7. Abstract:
  8. The origin-destination (OD) matrix is a fundamental component in urban transportation modeling and planning, representing the volume of trips between different zones within a city. Accurately estimating the OD matrix plays a critical role in travel demand analysis and traffic management systems. The gravity model is one of the most commonly used methods for OD matrix estimation. This study aims to use location information of urban zones, including trip production and attraction rates—key inputs of the gravity model—and proposes an alternative method to estimate the OD matrix using a Neural Network model. The proposed Neural Network model is compared with the gravity model in terms of accuracy and performance in estimating the OD matrix. Two models are designed for evaluation: the first model estimates the OD matrix based solely on travel demand data, while the second model incorporates network-based features extracted using graph-based methods. The results show that graph-based Neural Network models, without relying on cost-based metrics such as travel time and distance, can capture the structural characteristics of the network and yield accurate predictions across the entire network. Furthermore, comparison with the gravity model indicates that the proposed method achieves better accuracy and performance, especially in high-density areas. These findings highlight the importance of including network structure features and graph-based metrics to enhance estimation accuracy. Furthermore, the findings highlight the importance of customizing loss functions to align with specific predictive objectives, such as forecasting travel demand for low-demand origin-destination pairs
  9. Keywords:
  10. Origin-Destination Matrix ; Urban Transportation Network ; Shortest Path ; Centrality Metrics ; Neural Network ; Graph Indices ; Physics-Based Model ; Origin-Destination Matrix Prediction

 Digital Object List