Efficient OD matrix estimation based on metamodel for nonlinear assignment function

In this work, the problem of dynamic origin-destination (OD) matrix estimation, using traffic observations and its assimilation into traffic assignment models, is addressed. In the past decades, a rich body of literature, has been devoted to development of the heuristic methods to reduce high computational effort of traffic simulation without significant trade-off of result’s accuracy. In this paper we propose a metamodel as a function that maps the nonlinear relationship between traffic observations and OD flows. Derived metamodel framework is consistent with all the estimated demands through iteration steps, which results in non-constant, demand dependent, assignment matrix. Further, metamodel requires the same amount of traffic simulations as the classic approach, to maintain a low computational costs. In addition to proposed metamodel, a stochastic gradient method, that is more efficient to escape from local minima, is applied to solve the formulated dynamic OD matrix estimation problem. The performance of the proposed metamodel and solution approach has been evaluated for four OD estimation methods formulation. Least square OD matrix estimation model solved with gradient descent algorithm has been selected as a complementary method, widely used in practice, whose limitations and assumptions we aimed to solve. Numerical experiments are performed on real network, (Vitoria, Basque Country, Spain) with real data and two historical OD patterns to evaluate the performance of the proposed solution approach. Experimental results show that the proposed stochastic gradient metamodel method demonstrates stable decrease of the cost function with the lowest value through iteration steps. The experiments indicate that the proposed method based on metamodel yields to robust performance for both quality of the initial OD matrices, compared to benchmark method.

• Record URL:
  https://atrf.info/papers/2018/index.aspx
• Authors:
  • Masip, D
  • Djukic, T
  • Breen, M
  • Casas, J
• Conference:
  • Australasian Transport Research Forum (ATRF), 40th, 2018, Darwin, Northern Territory, Australia
• Publication Date: 2018-12

Language

• English

Media Info

• Pagination: 15p
• Monograph Title: 40th Australasian Transport Research Forum (ATRF), Darwin, October 30th - November 1st, 2018

Subject/Index Terms

• TRT Terms: Case studies; Evaluation; Highway traffic control; Mathematical models; Methodology; Origin and destination; Traffic flow
• Geographic Terms: Spain
• ATRI Terms: Case study; Evaluation; Methodology; Modelling; Origin destination traffic; Traffic flow; Traffic management
• Subject Areas: Operations and Traffic Management; Planning and Forecasting;

Filing Info

• Accession Number: 01696332
• Record Type: Publication
• Source Agency: ARRB
• Files: ITRD, ATRI
• Created Date: Feb 26 2019 2:41PM