Travel time variability (TTV) measures the extent of unpredictability in travel times faced by travellers. Unpredictability can arise due to day-to-day fluctuations in traffic demand, traffic incidents affecting capacity or weather conditions.
This project describes the development of a prototype model to predict travel time variability on Danish motorways. In this project TTV is measured as the standard deviation of travel times over all typical weekdays in the period of analysis.
The model is intended as a post-processing module to be applied after a traffic model has been used to predict travel demand on a road network. Its output (the level of TTV) does not feed back into the traffic model to account for the behavioural response to TTV. The model is thus an example of a ‘Method 1’ approach in the terminology of de Jong and Bliemer (2015), who recommend using this simple type of model in the short run. Such approaches have been implemented in the Netherlands, the UK, the US and Sweden. In the longer run, they recommend working towards implementing the prediction of TTV and the behavioural response to TTV into the traffic models.
The prediction model is based on a statistical model of the relationship between observed travel times and traffic flows. Both travel time and traffic flow are dynamic processes that evolve over the day and affect each other. Our aim has been a simple model which takes into account the dynamic relationship between the two and avoids the potential endogeneity issues related to this relationship. In this respect, our model constitutes a significant methodological improvement compared to the traditional speed-flow curves.
The prediction model enables us to predict mean travel time and TTV (measured as the standard deviation of travel time over different days) for a given traffic scenario, and to compute the travel costs associated with both, assuming a known monetary value of TTV.
In the project framework there are presented four simple case scenarios to illustrate the use of the method. In the scenarios, we apply a reliability ratio of one, i.e. the value of one minute’s standard deviation equals the value of one minute’s mean travel time, cf. the recommendations in DTU Transport, 2008. In most scenarios, we find that the change in the costs of TTV makes a significant contribution to the overall change in travel costs.
The prediction model is estimated on data from the Køge Bugt Motorway, a congested Danish motorway through the south-western suburban area of Copenhagen towards the city centre. In principle, the model is general enough to cover other roads, as it is highly simple. However, we recommend extending it by re-estimating its parameters on data from other roads when the necessary data become available. Moreover, we recommend further developing the model to take spillback effects into account.