Assessing the Value of Model Calibration for Signalized Intersection

Abstract

The Highway Capacity Manual (HCM) provides the most widely used methodologies for evaluating the quality of service on highway and street facilities. However, many users contend that the HCM models they are using are not accurate in emulating some real world conditions. In this thesis, the HCM control delay model for signalized intersections is assessed to identify where and when such deficiencies occur.

Four signalized intersections within the Chicago central business district (CBD) area are selected to assess the accuracy of the HCM control delay model. The lane groups studied include three through lane groups and one permissive left-turn lane group. Input data (e.g. lane group volumes, proportion of vehicles arriving on green, effective green time, etc.) for the HCM model and empirical data (e.g. actual measured control delay) were collected cycle by cycle in the field. Saturation flow rates estimated from (a) the HCM default parameter values, (b) field calibration and (c) statistical optimization are entered into the HCM control delay model, respectively, to calculate control delays. Then, the control delays calculated from the model are contrasted to those measured from the field.

For the through lane groups, the analysis indicates that the control delay calculated using the HCM default parameter values overestimates field delay in most of cases, and it may not accurately reflect the lane groups' performance. However, when field calibrated and/or optimized values are used, the control delay estimated using the model is close to that obtained from field measurements, and this indicates that the HCM control delay model is reliable. For the permissive left-turn lane group, no matter which values of saturation flow rate are used, the control delay estimated using the HCM model is not comparable with that obtained from field measurements due to the small sample size of the number of vehicles arriving per cycle, and other factors apparently not reflected in the delay model.