Forecasting Day to Day Variability on the UK Motorway Network
S Sirivadidurage, A Gordon, C White, Mott MacDonald Limited, UK; D Watling, ITS, University of Leeds, UK
This abstract describes how day to day variability functions were calibrated for three and four lane motorways with hard shoulder, mandatory variable speed limits and three lane motorways with hard-shoulder running in the UK.
?Reliable journeys? is one of three key objectives of the UK Highways Agency who are responsible for the motorway and trunk road network, as well as being a central theme of the Eddington report and a major international topic around the world. In order to measure and predict reliability, we need to understand variability in travel times. Specifically, day to day variability (DTDV) refers to variations in journey time due to unpredictable changes in demand and random fluctuations in capacity. In other words this is what remains after accounting for all predictable variations (time of day effects, day type effects and seasonal effects) and variability due to incidents. The UK Department for Transport (DfT), recognises that DTDV should be taken into account when appraising potential benefits of transport schemes or policies. Managed motorway systems such as mandatory variable speed limits and hard shoulder running can provide significant DTDV benefits. DfT commissioned Mott MacDonald to carry out a project to calibrate functions for predicting the DTDV on several road types and to incorporate these into INcident Cost benefit Assessment (INCA) software. INCA calculates delays and travel time variability costs relating to incidents, and the benefits that may arise from remedial measures to reduce their impact. In addition, the transferable DTDV functions that were estimated have the potential to be exploited in other future modelling studies and packages, where forecasts of travel time variability and its reliability impacts are required.
This paper describes how DTDV functions were calibrated for several motorway link types. These include three and four lane motorways with hard shoulder, mandatory variable speed limits and three lane motorways with hard-shoulder running.
The source of data for this work was the Highways Agency Traffic Information System (HATRIS). The HATRIS database provides network monitoring information for the motorway and trunk road network from inductive loop sensors (MIDAS), automatic number plate recognition and matching and GPS tracking. The data stored for each link in the network includes average speed, average journey time, flow, data quality indicator, for every 15 minute time period of every day for each link. Variables were calculated by retaining the distinction by day type and time periods in order to ensure that seasonal, day type and within day effects are excluded from the calculation of DTDV. There were 2016 data points for each link per year before filtering and most road types had data for more than a year. Therefore it was possible to conduct a more rigorous analysis than previously due to the large amount of data available. Any HATRIS data with low quality and those affected by incidents was excluded from the subsequent analysis.
After a detailed investigation of different combinations of dependent and independent variables, it was found that the best functional form to represent DTDV was to express the standard deviation of average journey time per km as a cubic function of the mean journey time per km. The method of Analysis of Covariance (ANCOVA) was first used to investigate the significance of road type in explaining the level of DTDV. This showed that, in each case, fitting a separate function for each road type gives a significant improvement in the overall fit. Coefficients for the fitted functions were thus calibrated separately for each road type using weighted least squared (WLS) regression. Estimated coefficients for all road types were highly significant at the 95% level. The fitted functions showed R-squared values ranging from 0.42-0.94. For these functions the weighted residuals were calculated and then standardised. They showed no discernable trend, were homoscedastic and followed a normal distribution, which indicate that fitted functions satisfy linear regression properties. Therefore the fitted functions are considered appropriate.
All these DTDV functions are implemented in the INCA software. Therefore INCA now calculates DTDV in addition to incident delay and incident variability. This version of INCA has been used recently to appraise incident and DTDV impacts of several managed motorway schemes in the UK.
Association for European Transport