Public transport travel time and its variability
2017-03-22T01:26:47Z (GMT) by
Executive Summary Public transport agencies around the world are constantly trying to improve the performance of their service, and to provide passengers with a more reliable service. Two major measures to evaluate the performance of a transit system include travel time and travel time variability. Information on these two measures provides operators with a capacity to identify the problematic locations in a transport system and improve operating plans. Likewise, users can benefit through the provision of information on future travel times. They can use this information when selecting departure times since that information enables them to select departure times to minimize waiting times and to ensure efficient transfers between alternative transport modes. This thesis is focussed on enhancing understanding of public transport travel time and its variability. Through a literature review, three main areas emerged as knowledge gaps: (1) exploring travel time variability and its causes, (2) prediction of travel time and travel time variability, and (3) determination of an optimal schedule design to increase service reliability, and reduce travel time variability. Travel time variability and its causes: Travel time variability deteriorates service reliability, yet it is not well-researched in the transport literature. This is partly due to the lack of comprehensive data sets on bus travel times. While this problem is now being addressed through the uptake of Global Positioning System (GPS)-based tracking systems, methodologies to adopt these data sets to explore travel time variability are limited. The first part of this thesis addresses this issue by investigating bus travel time variability using a GPS data set for a bus route in Melbourne, Australia. It explores the nature and shape of travel time distributions for different departure time windows at different times of the day. The results show that in narrower departure time windows, travel time distributions are best characterized by normal distributions. For wider departure time windows, peak-hour travel times follow normal distributions, while off-peak travel times follow lognormal distributions. The research also shows how GPS data can be used to identify the causes of bus travel time variability. To this end, factors contributing to travel time variability are investigated at two levels: the route level and the route section level. In the route level analysis, temporal variables were used as proxies to real contributors. ‘Time of day’ (different 15 minute intervals during a day) is found to have the highest impact on travel time variability in all periods of day. ‘Day of week’ (different weekdays) is shown to have the greatest effect on inter peak travel times, whereas its effect is the least in morning peak period. ‘Month of year’ (different months in a year) shows the greatest impact on morning peak travel times, and the lowest influence on off-peak travel times. Peak hour travel times are also considerably different in summer and school holidays. Rain was not found to have a significant effect on travel time for any period during the day. This result could be related to the low number of rainfall observations in the data which was available for this study. In the route section level analysis, causes of travel time variability are explored by comparing the variability values across different route sections. Travel time records of different route sections are aggregated into different 15 minute intervals, and then used to calculate variability values for each 15 minute window. The variability is analysed through a linear regression analysis. It is found that section length, surrounding land use, number of bus stops, and number of signalized intersections influence day-to-day travel time variability. The arrival time of buses relative to the scheduled arrival time is also found to be significant in explaining the variation of travel time values. Variability is found to be higher in the morning peaks, and lowest in the off-peak periods. The analysis completed in this part of the research suffers from the limited depth of the explanatory variables adopted for the regression analysis caused by the lack of data notably ridership and traffic flow data. Problems of this type are quite common for practitioners and researchers alike and the use of proxy variables as adopted in this analysis might be a useful example to those facing data limitations. Nevertheless, there is clearly a scope to research travel time variability causes with a more comprehensive range of explanatory variables. Prediction of travel time and travel time variability: Despite the important effect of traffic flow on bus travel time, previous research has not considered a traffic measure making the predictions of bus travel time unresponsive to the dynamic changes in traffic congestion. In addition, existing methodologies are almost exclusively concerned with predicting average travel time for a given set of input values. Predicting travel time variability has not received sufficient attention in previous research. On the basis of data collected from a bus route in Melbourne, Australia, this thesis employs an artificial neural network modelling technique to predict bus average travel time. To this end, a set of input variables are used including real world traffic flow data collected by the Sydney Coordinated Adaptive Traffic Systems (SCATS) loop detectors. Since collection of traffic flow data might not be an easy task in other cities in the world, the thesis also explores the value that traffic flow data makes to the accuracy of travel time predictions compared to when either temporal variables or scheduled travel times (as adopted by a number of previous studies) are the base for prediction. While the use of scheduled travel times results in the poorest prediction performance, incorporating traffic flow data yields minor improvements in prediction accuracy compared to when temporal variables are used. Traditionally, neural networks give rise to a prediction point (average of the dependent variable) when presented with a set of input variables. However, this thesis develops a capability to provide a range (rather than a prediction point) when a certain set of input values is given. The method is adopted to predict bus travel time variability and demonstrates a promising ability to provide a range which has a high probability of including the actual travel time. Determining an optimal schedule design: A portion of variability in public transport travel time is caused by holding strategies applied at predefined ‘timing point’ stops to improve reliability. Early running buses are held until the scheduled departure times, and buses running late (relative to the schedule) will depart immediately after serving passengers. Scheduled departure times are defined by adding a ‘slack time’ to the mean bus arrival time to ensure the on-time departure of a certain portion of late running buses. Determining the location of timing points and the amount of the associated slack times is a key step in bus route planning. The literature refers to this as the transit schedule design problem. This research shows that the transit schedule design problem can be treated as an optimization problem, where the objective is to minimize a generalized cost function and decision variables are the location of timing points and slack times. Early studies considering timing point/slack time selection as an optimization problem have primarily dealt with simplified cases where only a single bus run or a single predefined timing point is considered. This thesis develops an optimization method to solve the schedule design problem to determine an optimal set of timing points and slack times when multiple bus runs are considered. The algorithm, which is based on the Ant Colony optimization technique, demonstrates its ability to solve the problem in a manageable time. The research demonstrates the potential of the algorithm to serve as an efficient tool for bus route planning applications.