Assessing the Impact of Future Personalised Public Transport

Transcription

1 CASPT 2018 Extended Abstract Assessing the Imact of Future Personalised Public Transort Yu Jiang Avishai (Avi) Ceder Abstract The urose of this study is to assess the otential imact of future ersonalised ublic transort (PT) services allowing a traveller to communicate with an information rovider using secific ersonal references when undertaking a tri. Our methology is based on finding the K-shortest ath for travellers where the value of each link comrises the cost of the various weighted factors based on users references, and assigning travellers to a ath set according to service frequency and ath cost. Case studies were conducted using the Coenhagen Central Network. It was found that, for instance, future ersonalised PT, could reduce user travel cost by 5.91%. Keywords: Personalised Public Transort Smarthone Guidance K-shortest Path Transit Assignment. 1 Intruction Public transort (PT) is a key element in most major cities around the world. With the develoment of smarthones, real-time and readily available journey lanning information is becoming an integral art of the PT system. The information might not only be better to inform traveller s tri lan, but also allows a traveller communicating about his/her secific references when undertaking a tri. PT travellers are likely to have references over different features of a journey, that might even influence their travel decisions including mal choice and route choice. A number of factors/attributes have been found to be imortant in affecting Yu Jiang Transort DTU, Danmarks Tekniske Universitet (DTU), Lyngby, Denmark yujiang@dtu.dk Avisahi (Avi) Ceder Faculty of Civil and Environmental Engineering, Technion-Israel Institute of Technology, Haifa, Israel. And Deartment of Civil and Environmental Engineering, The University of Auckland Auckland, New Zealand a.ceder@auckland.ac.nz

2 assengers choice, such as the quality of a PT service, connectivity, fare cost, accessibility, and journey distance (e.g., Kingham et al. 2001, Galdames et al. 2011; Ceder and Teh, 2010). These factors or attributes could lay a major role in determining PT route choices (Ceder et al. 2013; Ceder, 2016; Grison et al. 2017). Moreover, assengers references over these attributes could vary deending on various factors, such as time of day, mo, schedule to follow, family issue, etc. Therefore, every tri a user makes has unique requirements. Accordingly, a welldesigned smarthone alication should allow for travellers to interact with the oerator in term of sending references and receiving tailored information. In the last few years and recently an increased attention was and is given for informative and to some extent ersonalised smarthone alications by the industry and researchers (e.g., Global Mass Transit 2014; Shaheen et al. 2016). Generally seaking, as er Chorus (2012), a ersonalised smarthone alication can be beneficial from two ersectives: (i) The services remember and learn from the travellers choice rofile and allow us to redict travellers mobility or to issue context-sensitive ersonal advice (Lathia et al. 2013; Bouhana et al. 2013; Arentze 2013); and (ii) The services consider travellers references overs different attributes (i.e., Peng and Huang, 2000; Zografos et al. 2009; Chorus et al. 2009). The focus of this study is ersective (ii) and the objectives are threefold. First, it rooses a methology based on the K-shortest ath to mel future ersonalised PT services, which consider travellers references of different PT attributes in the route guidance. Second, it develos a corresonding transit assignment mel to redict the usages for the future PT services. Finally, it alies the route-guidance and transit assignment mel to assess the otential imacts of future ersonalised PT services. 2 Methology In traditional smarthone alications, users are given a set of aths and might select to sort the aths using given attributes, such as travel time (i.e., Google ma), and ossibly with a boundary related to an attribute, such as the maximum number of transfers (i.e., Rejselanen A in Denmark). These as, to some extent, cater assengers references for different PT attributes, but cannot cature the trade-off between the various attributes and the stochasticity involved with assengers references across time, sace, weather, imortance, mo, etc. In contrast, a future ersonalised smarthone alication should allow for assengers to indicate, at the time of a need for a ride, their references of different attributes and in return to obtain a set of best-recommended routes/aths. To mel this, we roose to use a weighted travel cost incororating various PT attributes and to use a K-shortest ath algorithm to determine a set of aths adatable to users references. To attain this goal, assengers route choice behaviour is taken also into account by develoing a transit

3 assignment mel based on the K-shortest ath concet. The overall framework of our analytical aroach is illustrated in Figure 1. From a user s ersective, a traveller could assess various PT-service information rovided by the oerator and secify the relative imortance of different attributes of his/her route choice decision made via a smarthone. Given these arameters, a athfinding comonent returns K-shortest ath otions to be dislayed on the traveller s smarthone alication. Considering assengers real-time demand and the comutational algorithmic effort, we suggest comuting the K-shortest ath for all OD airs offline and storing the resultant ath sets in a database. For the offline comuting, we could set all attributes equally imortant or calibrate these arameters from survey data. Smarthone A Relative imortance of different attributes Transit Services Path Finding Personalised routes Transit Assignment K-shortest Path Fig. 1 Framework In real-time scenarios, the inut of references is used to udate the weighted costs of the K-shortest ath and obtain a new order of best recommended. At the same time, the set of best aths is utilised in a transit assignment mel to redict the usages of PT services uon which the oerator could imlement oerational tactics for imroving the level of service rovided. The mathematical formulations encasulated in the analysis framework are briefly elaborated as follows. m m dij = α cij, ij (1) m π = δij dij,, (2) ij

4 ( θπ ) ( θπ ) fl ex x = g,, Pk f ex l L l Ω Equation (1) defines the weighted link cost d ij, where (3) m c ij is attribute m associated m with link ij and α is the weighting arameter alied to attribute m based on users references. Equation (2) comutes the weighted travel cost of ath connecting nes o and d. Equation (3) deicts the transit assignment mel given the K-shortest ath set, where x denotes the number of assengers travelling on ath ; f P k reresents the frequency of the first PT service associated with ath ; L denotes the set of transit lines dearting from ne o; Ω is the set of aths that starting with the same PT service; g is the travel demand between nes o and d. The assignment mel assumes that each assenger has a set Ω of attractive aths (based on the K-shortest ath rocedure s outcome) and boards the first PT vehicle associated with these aths. In case the first PT vehicle is utilised by more than one aths, the assenger is then assigned to a ath using logit mel. That is, this assenger will use the first arrived PT vehicle, but will use different transfers than the case in which the vehicle won t be utilised by other aths. In the K-shortest ath algorithm, the weighted factors are associated with an individual user, who uses a smarthone alication to set his/her references. However, for the assignment mel, an oerator is more interested in the aggregated flow distribution instead of disaggregated route choice. Thus, in line with Nielsen (2000), we adot a simulation rocedure to generate various assengers references to comute average flows. 3 Case Study The methology develoed was alied for the Coenhagen Central Area shown in Figure 2. The Greater Coenhagen is divided into 99 zones. The studied area contains Coenhagen s Zones 1, 2, 3, and half of 4, which cover most of central Coenhagen city. All PT services oerated between 7 am and 9 am are included. The frequencies of the PT services were aroximated based on the latest-ossible timetable information of the fall of The demand data includes all tris between 7 am and 9 am. In short, there are 278 PT lines, 397 stos, and OD airs. The rogram was ced in C# and run on a lato with Intel(R) Core(TM) i5-3380m The first exeriment was designed to examine the significance of using a ersonalised smarthone alication. Two exeriments were constructed. In the first exeriment assengers were assumed to select the shortest ath without considering their references. In the second exeriment it was assumed that these assengers indicate l

5 their references using a smarthone alication, and that the alication dislays the K-shortest ath based on a weighted ath cost. The first exeriment resembles the usage of a traditional alication; the second exeriment reresents a future ersonalised alication. Without a loss of generality, references were randomly generated for each OD air and set to be identical for the two exeriments. For simlicity the analysis focuses on the weighted cost of the 1 st ath of the recommended ath set. Our results reveal that, in an average sense, assengers of the second exeriment could reduce their weighted travel cost by 5.91%. Fig.2 Coenhagen Central Network (Zones 1, 2, 3 and 4) The second exeriment was designed to test the robustness of the methology using 1000 simulation runs. The results show that the average cost reduction of all 1000 runs is 5.94%. The maximum, minimum, and the standard deviation of the cost reduction are 6.19%, 5.64% and 0.08%, resectively. These results confirm that the roosed methology is robust, and its erformance is stable in terms of reducing the weighted ath cost for a large network. 4 Conclusions and Future Research This study develoed a methology to mel future ersonalised PT services based on finding the K-shortest ath of travellers where the value of each link is comrised of the cost of the various weighted factors reresenting the users references. Meanwhile, a logit based transit assignment mel was roosed to redict travellers route choice. The melling framework was alied to assess the imact of the ersonalised PT services in the Coenhagen Central area. It demonstrates that assengers could reduce their weighted travel cost by 5.91%, and the ersonalised PT is robust in reducing the weighted travel cost.

6 Given the technology available, the high use of smarthone devices, and the otential for user references to be taken into account, it is recommended that further research consider the following oints: Alication of the methology with a stochastic shortest and K-shortest ath algorithm to understand the variance in PT tris that occur in reality and incororate assengers risk-aversion attitude in the transit assignment mel (Jiang and Szeto, 2016). Extend the otions rovided to users to cater for more real-time elements such as assengers load rofiles and network congestion. Undertake a case study using more data to ensure the adequacy of the methology. Acknowledgements: This work is artly funded by the Innovation Fund Denmark (IFD) under File No References Arentze, T.A. (2013). Adative ersonalised travel information systems: A Bayesian meth to learn users' ersonal references in multimal transort networks. IEEE Transactions on intelligent transortation systems, 14(4), Bouhana, A., Fekih, A., Abed, M., Chabchoub, H. (2013). An integrated case-based reasoning aroach for ersonalised itinerary search in multimal transortation systems. Transortation Research Part C, 31, Ceder, A. (2016). Public Transit Planning and Oeration: Meling, Practice and Behavior, Second edition, CRC Press, Boca Raton, USA. Ceder, A., Teh., C.S. (2010). Comaring Public-Transort Connectivity Measures among Major Cities (in New Zealand). Transortation Research Record: Journal of the Transortation Research Board, No. 2143, Ceder, A., Chowdhury, S., Taghiouran, N., Olsen, J. (2013). Melling ublictransort users' behaviour at connection oint. Transort Policy, 27, Chorus, C.G. (2012). Travel Information: Time to Dro the Labels?. IEEE Transactions on Intelligent Transortation Systems, 13(3), Chorus, C.G., Arentze, T.A., Timmermans, H.J. (2009). Traveler comliance with advice: A Bayesian utilitarian ersective. Transortation Research Part E, 45(3), Galdames, C., Tudela, A., Carrasco, J.A. (2011). Exloring the role of sychological factors in me choice mels by a latent variables aroach. Transortation Research Record: Journal of the Transortation Research Board, 2230, Global Mass Transit reort, Smarthone Allications in Transit Services: Growing Poularity. Retrieved from htt:// 30 June Grison, E., Burkhardt, J.M., Gyselinck, V. (2017). How do users choose their routes in ublic transort? The effect of individual rofile and contextual factors. Transortation Research Part F, 51,

7 Jiang, Y., Szeto, W.Y. (2016). Reliability-based stochastic transit assignment: Formulations and caacity aradox. Transortation Research Part B, 93, Kingham, S., Dickinson, J., Cosey, S. (2001). Travelling to work: will eole move out of their cars. Transort olicy, 8(2), Lathia, N., Smith, C., Froehlich, J., Cara, L. (2013). Individuals among commuters: Building ersonalised transort information services from fare collection systems. Pervasive and Mobile Comuting, 9(5), Nielsen, O.A. (2000). A stochastic transit assignment mel considering differences in assengers utility functions. Transortation Research Part B, 34(5), 2000, Peng, Z.R., Huang, R. (2000). Design and develoment of interactive tri lanning for web-based transit information systems. Transortation Research Part C, 8(1), Shaheen, S., Martin, E., Cohen, A., Musunuri, A., Bhattacharyya, A. (2016). Mobile As and Transortation: A Review of Smarthone As and a Study of User Resonse to Multimal Traveler Information. Final reort, ITS of UC Berkeley, California Deartment of Transortation (Caltrans), October. Zografos, K.G., Androutsooulos, K.N., Sitadakis, V. (2009). Design and assessment of an online assenger information system for integrated multimal tri lanning. IEEE Transactions on Intelligent Transortation Systems, 10(2),