Modelling passenger behaviour in mixed scheduleand frequency-based public transport systems

Abstract

Transport systems in metropolitan areas are on both the road and public transport side challenged on providing sufficient capacity for the increasing mobility needs. An increasing number of hours is wasted in congestion on the roads, and good public transport service is needed to provide sufficient capacity in the transport system. Public transport is not only seen as a way of increasing the mobility in metropolitan areas, but also as one of the important contributors to the transition for more sustainable mobility in urban areas. The public transport systems must thus be an attractive alternative to taking the car to attract more passengers, and facilitate the transition for a more sustainable transport system. The public transport systems are often complex with a mix of lines, where passengers for some services rely on a published detailed timetable (schedule-based lines), while they for other high frequency lines rely on the headway between consecutive services on the line (frequency-based lines). Due to the complexity of these systems, advanced models are needed to analyse the level of service provided to the passengers given the timetable of the network. Furthermore, the inputs for these models require analysis of various aspects of passenger travel behaviour based on reliable data sources. This PhD thesis concerns several aspects within modelling of public transport systems with a passenger oriented perspective. The thesis is split into three main parts; Part I, Assignment models for mixed schedule- and frequency-based public transport systems, presents novel methodological approaches for determining the level of service for passengers in public transport networks with both schedule- and frequency-based services; Part II, Route choice models for mixed schedule- and frequency-based public transport systems, focuses on passengers’ route choice preferences from origin to destination in these complex networks, based on revealed passenger route choice surveys. The third and final part of the thesis, Studies on public transport passenger behaviour based on smart card data, covers three analyses of passenger travel behaviour based on smart card data. Part I of the thesis covers the difficult task of assigning (predicting) passengers to routes from origin to destination in order to evaluate the level of service provided to the passengers. Specifically, two studies focus on the combination of schedule- and frequency-based services in public transport networks and how to assess the travel times on the attractive routes in such a network. The first paper develops a novel methodology to assign passengers in a mixed schedule- and frequency-based network. First, choice-sets with different possible routes are generated based on a heuristic, which requires that the passenger can reach the destination within a certain threshold using the specific route. A subsequent step distributes the passengers across the alternatives using a discrete choice model. The resulting flow distributions across alternatives are stable regarding the specification of a line as either schedule- or frequency-based. Compared to other models, this allows the modeller to make fewer assumptions on the actual schedule of a line, and eases the evaluation when several timetable scenarios need to be compared. The second paper proposes a method that uses Markov chains to identify the travel time distributions for different route choice alternatives, when stochastic running times of a line due to delays are taken into account. Given a set of attractive lines for passengers travelling from origin to destination, the methodology calculates the travel time distribution for different combinations of lines and thereby alternative routes through the network. Both schedule- and frequency-based lines can be part of the input to the model. By using Markov chains the probability of reaching a connecting service, can be analytically described, whereby the use of traditional demanding simulation models can be avoided. Several detailed analyses can be derived based on the resulting travel time distributions, which can become an important tool for timetable planners. While Part I of the thesis covers the evaluation of the level of service offered to the passengers, Part II covers the input to these models by investigating the route choice preferences of the passengers. Travellers evaluate the attractiveness of a route based on several features such as travel time and number of transfers, but the specific challenge concerning how passengers trade off for example routes with a high travel time vs. routes with lower travel time which include more transfers persist. These trade-offs are investigated in two papers using a dataset covering self-reported trips using public transport in the Greater Copenhagen area. In both papers a discrete choice model is the basis for the extraction of the passenger route choice preferences, and this is achieved by comparing the observed routes with a large set of alternative routes the passenger could have chosen. The third paper investigates the trade-offs passengers have to make when choosing between alternatives with different waiting times and in-vehicle times. Waiting time for schedule-based services, such as regional trains and local busses, are estimated separately from frequency-based services (metro and high-frequency busses) and this shows, that passengers have a higher nuisance for waiting for frequency-based services compared to waiting for schedule-based services with a known timetable. However, in general lower waiting times for frequency-based services makes the decision between alternatives with schedule- or frequency-based services almost the same. If the differences in parameters of waiting time are not accounted for in assignment models, there is a risk of creating a biased flow estimation, which can eventually lead to wrong conclusions in feasibility studies. The paper also investigates whether the marginal dis-utility of in-vehicle time varies across and within each sub-mode, i.e. metro, bus and trains. It is shown, that the marginal dis-utility for metro considerably increases for longer trips whereas the marginal dis-utility decreases for in-vehicle time in trains. The fourth paper focuses on the choice of transfer location in passenger route choice. The paper reviews existing literature on transfer attributes which affects passenger route choice, and selects three attributes found to be important for passenger route choice. The analysis shows that passengers prefer routes, which includes a shop available at any of the transfer stations visited during the trip, thus indicating the preference for being able to do smaller grocery shopping en-route. Passengers also prefer escalators over regular stairs, and prefer that transfers should be easy to navigate through. Using these attributes, it is possible to disentangle the transfer penalty for stations with different characteristics. The best possible transfer thereby has a penalty equivalent to spending 5.4 minutes extra in a bus, whereas the worst possible transfer is comparable to spending 12.1 minutes in a bus. The results have important policy implications for evaluation of different station designs and how the resulting passenger flows will be, if stations are upgraded or redesigned. Such investments can turn out to be more cost effective than track upgrades or other improvements of the railway, while still providing a better level-of-service to passengers. The final part of the thesis, Part III, covers three analyses based on smart card data. Smart card data is available in rich numbers from automatic fare collection systems and is becoming an increasingly important tool for analysing passengers’ travel behaviour. This thesis uses smart card data with different degrees of detail, and the studies span from analysing the individual mobility to more aggregated analysis, where smart card data covers the heterogeneity of passenger behaviour. The fifth paper investigates individual mobility over a long time period based on data from the Danish smart card, Rejsekort. The study analyses travel behaviour before, during and after a three month track closure on a suburban rail line in the Greater Copenhagen area, where replacement busses served the line resulting in significantly increased travel times. Passengers are clustered based on their travel behaviour before and after the track closure. A similar track section is used as comparison to the changes in ridership at the suspended track section, as the individual passenger travel behaviour changes considerably over time due to changes in individual employment and general seasonal trends. By comparing the changes in travel behaviour for passengers travelling frequently before the disruption on either the affected or reference line, no apparent difference is seen for the period after normal operations resumed. However, the total ridership on the affected line decreased compared to the reference line, and a comparison of the changes in passenger travel for the different groups, suggests that the deficit is a result of less attraction of new passengers on the affected line. By analysing the daily travel patterns for the group who commuted on the affected line before the disruption, it is found that 17% of the passengers almost entirely stopped using public transport during the disruption, but returned to a regular usage of public transport after the normal operations resumed. This indicates, that at least some passengers favor public transport and are not forced to use the public transportsystem. Data from Rejsekort is also used in the sixth paper, but on a more detailed level. The paper fuses smart card data and automatic vehicle location data to estimate the walking time used from alighting a bus until the passenger taps in at a train platform. This walking time is essential to know, as it is used in timetabling and synchronisation of busses and trains. Using the raw observed times from the data fusion leads to significantly overestimated walking times, as some passengers are doing activities during their transfer. Therefore, a hierarchical Bayesian mixture model is used to isolate the passengers doing activities during the transfer from the passengers walking directly. The results show that the model is able to accurately replicate the observed walking times and estimate the walking time necessary to walk from bus stop to train platform. The study establishes a more data-driven procedure for estimation of walking times at transfers, and is applied to 129 stations in the Eastern part of Denmark. Tests show that the share of passengers doing activities during their transfers increases with the number of shops available near the transfer station. Whereas the two preceding papers focus on the use of data from Rejsekort, the seventh and final paper utilises an extensive smart card dataset from Hong Kong. The number of passengers in the Metro in Hong Kong exceeds the available capacity during the peak hours, and the paper describes and analyses unusual path choice behaviour that stems from the excessive crowding. Under the excessive crowding situations passengers can be observed to do reverse routing, namely choose to transfer at a station further down a line in order to travel backwards and pass the station where passengers would usually transfer in uncrowded conditions. Such reverse routing can increase the travel time reliability and also increase the chance for the passenger to get a seat or better standing position in the train. However, based on the analysis, no final conclusions can be made on the share of passengers using this option of reverse routing. However, the results indicate that passengers travelling furthest after transferring have a slightly different behaviour, which could stem from a higher degree of reverse routing. It can also be substantiated by the finding in the third paper, that the marginal value of time is increasing for passengers using the metro. A short paragraph in the paper also considers whether such unusual route choices are occurring in the Danish Metro, but based on analysis of data from Rejsekort, this can quickly be ruled out to be the case. In summary, this PhD thesis has contributed to i) new methodologies to assign passengers to routes for detailed and analytical evaluation of the level of service provided to passengers in mixed schedule- and frequency-based public transport system, ii) revealing and quantifying of the significant dis-utility of transfers in public transport route choice in combination with detailed analysis of the important characteristics of station attributes, and iii) develop two novel methods using data from Rejsekort for analysing both longterm travel behaviour and walking times at transfers, and iv) investigate the effects of crowding on passenger path choice in congested metro systems. Overall the thesis covers a broad span of public transport modelling and contributes to already existing knowledge in the domain. Several new methodologies are developed, especially on the use of smart card data, and these can be used for further research within the domain.