A framework is described for the examination of mathematical models, where model calibration and evaluation is based on the ability of a model to generate a set of observed outputs from a supplied set of observed inputs. With refinement and development, the framework may lead to general structures for the more rigorous evaluation of models. An application of the proposed framework to a local area traffic assignment model is described. The application facilitates the description of the framework and some of its technical components. The traffic model was tested with peak period traffic data collected in an inner suburban area of Melbourne, Australia, during 1974. On the basis of the tests conducted, this model seems a potentially useful tool for use in small-area traffic analysis. The suggested evaluation framework has its roots in previous work in the areas of transportation, hydrology and mathematical statistics. A series of levels for model evaluation are defined. For models calibrated on their ability to reproduce observed phenomena, conclusions about model usefulness and applicability need to be made with some caution. It is necessary to investigate goodness-of-fit of model outputs, and the robustness of parameter estimates, with data independent of that on which the model was calibrated. A possible scheme, known as "dual-sample testing", for this investigation is outlined. A number of possible tests of model outputs and possible systematic errors are also discussed. (a) (TRRL)

  • Availability:
  • Corporate Authors:

    Pergamon Press, Incorporated

    Headington Hill Hall
    Oxford OX30BW,    
  • Authors:
  • Publication Date: 1979-6

Media Info

Subject/Index Terms

Filing Info

  • Accession Number: 00310409
  • Record Type: Publication
  • Source Agency: Transport Research Laboratory
  • Report/Paper Numbers: SERIAL
  • Files: ITRD, TRIS, ATRI
  • Created Date: Jun 26 1980 12:00AM