Modeling Travel Time Variability on Urban Links in London
S Hasan, M E Ben-Akiva, Massachusetts Institute of Technology, US; C Choudhury, Massachusetts Institute of Technology/BUET, US; A Emmonds, Transport for London, UK
This research focuses on econometric analysis of data available from different sources to investigate the causes of traffic congestion in Central London and identify the key factors contributing to travel time variability.
Recent traffic surveys in London show a decline in traffic demand levels and perversely a decline in speeds and increase in congestion. These are likely to be attributed to factors that affect physical capacity of the road (e.g. introduction of bus lanes, cycle lanes, advanced stop lines etc.) as well as factors affecting the effective capacity (e.g. introduction of bus signal priority, pedestrian phases, additional pedestrian phases etc.). Other potential causes of congestion that have been identified include introduction of traffic calming measures and increase in number of safety/speed cameras etc.
This research focuses on econometric analysis of data available from different sources to investigate the causes of traffic congestion in Central London and identify the key factors contributing to travel time variability. A two-step procedure has been used where the first step estimates regression models using data for a single day. The second step predicts travel times on a different day(s) by applying the coefficients from the first step. This procedure significantly differs from previous researches on this area (e.g. DETR 1999, Robinson 2005, Krishnamoorthy 2008 etc.) which were limited in scope due to the deductive nature of their approaches where travel times are not directly observed; instead they are estimated indirectly from other sources of information (i.e. flow from detectors).
Several linear and non-linear regression models have been developed to determine the relative contribution of different factors on variation of travel time with individual travel times being dependent variables and factors that influence traffic behavior being independent variables. A number of candidate factors were identified, such as vehicle-specific dummy variables, traffic density level, vehicle type specific density, e.g., buses, taxis, heavy goods vehicles, light goods vehicles etc. and incident information.
The explanatory variables in our regression models and individual vehicle travel times come from the Automatic Number Plate Recognition (ANPR) data. The explanatory variables related to incidents come from the London Traffic Incident System (LTIS) database. The initial analysis was done with three different links with varying characteristics. We analyzed the contributions of different factors on recurrent delay and non-recurrent delay for these links.
The preliminary results indicate that non-linear regression models can explain travel time variability of an urban link with multiple signalized intersections accounting for different traffic related variables. Further non-recurrent delays can be modeled satisfactorily by incorporating incident information in the model structure. Spatial and temporal transferability of the modeling parameters have also been tested. Though temporal transferability results of the models are satisfactory, spatial transferability of the model parameters need further investigations.
The research results provide useful insights regarding the causes of traffic congestion in London. One of the findings is that different vehicle type (i.e. heavy goods vehicle (HGV), bus, taxi, car etc.) has significantly different free flow travel time. The travel time also depends on the overall density of the road and the relationship between travel time and traffic density are found piecewise linear. There are also some evidences that vehicle-type-specific density (taxi density, HGV density) plays a role in travel time.
However, several critical data and modeling issues have been identified in the preliminary model analysis. For example, comparison of ANPR data against Automatic Traffic Count data (where available) indicated that ANPR data which under-represent true traffic leading to measurement errors in the independent variables and resulting biased coefficient estimates of the parameters. On the other hand, some of the important variables which are required for detailed traffic behavior modeling and essential for understanding the complex traffic dynamics are missing in the current models. Such variables include queue length and turning movements at each intersection, cross-traffic volume, signal timing etc. A simulation based approach has been formulated to generate the missing values and will be used in a subsequent phase of this research.
This research is based upon work supported by Transport for London (TfL) under the MIT-TfL joint research program. Any opinions, findings and conclusions or recommendations expressed in this publication are those of the authors and do not necessarily reflect the views of TfL.
DETR (1999), The Appraisal of Measures to reduce Travel Time Variability (TAMTTV).
Krishnamoorthy, R. K. (2008), Travel time estimation and forecasting on urban roads, PhD thesis, Centre for Transport Studies, Imperial College London.
Robinson, S. (2005), The development and application of an urban link travel time model using data derived from inductive loop detectors, PhD thesis, Centre for Transport Studies, Imperial College London.
Association for European Transport