Characterizing Travel Time Reliability and Passenger Path Choice in a Metro Network

Size: px

Start display at page:

Download "Characterizing Travel Time Reliability and Passenger Path Choice in a Metro Network"

Brandon Ward
6 years ago
Views:

1 Characterizing Travel Time Reliability and Passenger Path Choice in a Metro Network Lijun SUN Future Cities Laboratory, Singapore-ETH Centre lijun.sun@ivt.baug.ethz.ch National University of Singapore (NUS)

2 Background Service Reliability Travel time Train load Transfer convenience Service disruption Flow assignment

3 Background Service Reliability The reliability of metro systems is higher than other transit modes. However, travel time variability still shows accumulative effect. 400 Standard deviation (sec) Mean travel time (sec)

4 Background Most metro systems are closed environments, which only register transactions when passengers enter and leave the system Crucial questions: Predicting travel time (and reliability) Inferring route choices Inferring train load Identifying critical transfer location Building sophisticated flow assignment models All linked together

5 Background Data Operation log? Transfer demand? Link flow? Route choice? Trajectory? Smart card (Boarding station, Alighting station, Travel time)

6 Metro System T b T a

7 Metro System What we know? (Observed) Network configuration Boarding station B Alighting station A Travel time t What we do not know? (Unknown) Travel time on each link Reliability of each link Other time cost (waiting at platform, walking between fare gantry and platform) Route choice

8 Metro System Research question Given observations to infer unknowns observed Boarding station B Alighting station A Travel time t unknown o Travel time on each link o Reliability of each link o Other time cost o Route choice

9 Metro System Methodology: Bayesian inference P(unknown observed) P(observed unknown) x P(unknown) Likelihood x Prior observed Boarding station B Alighting station A Travel time t unknown o Travel time on each link o Reliability of each link o Other time cost o Route choice

10 Unknown Travel time on each link Travel time variation on each link (in-vehicle / transfer links) Assuming that link cost follows a normal distribution, which has a constant coefficient of variation (linear mean v.s. std) x c, c a a a 2 Assuming that all links are independent Then the travel time on a particular route follows t r ca, c ar r 2 2 a ar

11 Unknown Other time cost Assuming that other cost follows a normal distribution, with y Assumed to be consistent for all OD pairs m 2, y

12 Unknown Route choice Serangoon Jurong East

13 Unknown Which route to take? Using brute-force search Serangoon Jurong East

14 Unknown Which route to take? Using brute-force search

15 Unknown Which route to take? Using brute-force search

16 Unknown Which route to take? Using brute-force search

17 Unknown Which route to take? Using brute-force search

18 Unknown Which route to take? Using brute-force search

19 Unknown Which route to take? Using brute-force search

20 Unknown Which route to take? Using brute-force search

21 Unknown Which route to take? Using brute-force search

22 Unknown Which route to take? Using brute-force search

23 Unknown Which route to take? Using brute-force search More than 30 routes in total

24 Unknown Route choice (in general) Multinomial Logit (MNL) Representative Utility Choice probability k Parameters are unknown P i V i X ja k exp Vi exp k V j ik

25 Unknown Route choice (for the metro network) Multinomial Logit (MNL) Utility V 1 ca 2 c r ar \r ar in-vehicle time t transfer time t a

26 Unknown Route choice (for the metro network) Multinomial Logit (MNL) Utility V 1 ca 2 c Choice probability r ar \r ar in-vehicle time R w 1 f t w transfer time r, c, θ t a exp rr w For each OD pair w V r exp V r

27 Observed Observing travel time for OD pair The probability observing on route The probability observing on OD pair t tb t a w a, b t r t r c m, c ar t p,,, w t c θ m h t r fw r, c, θ, m rr a a y ar w w

28 Observed The likelihood of observation all smart card transactions (travel time) c,, θ, m T p Tw c,, θ, m ww ht rf w r c,, θ, m wwtt w rrw

29 Observed Prior knowledge Mean link travel time follows normal distribution ca 2,1 Travel time between stations / transfer time: around 2 minutes Other cost follows a normal distribution m 4,1 Waiting time plus access/egress cost: around 4 minutes in total

30 Observed Prior knowledge Parameters for MNL: we do not have any information In the literature: Raveau S, Guo Z, Muñoz JC, & Wilson NHM (2014) A behavioural comparison of route choice on metro networks: Time, transfers, crowding, topology and socio-demographics. Transportation Research Part A: Policy and Practice 66:

31 Observed Prior knowledge Parameters for MNL: we do not have any information We take uniform priors θ 4,0 Coefficient of variation We take a uniform prior 0,1

32 Posterior c,, θ, m T Likelihood x Prior pt r f w r c, θ c;2,1 m;4,1 wwttw rrw cc

33 Solution Algorithm MCMC (Markov Chain Monte Carlo) Variable-at-a-time Metropolis sampling scheme δ c,, c,,,, m,, 1 N N 4

34 Solution Algorithm MCMC (Markov Chain Monte Carlo) Variable-at-a-time Metropolis sampling scheme δ c,, c,,,, m,, 1 N N 4 STEP (0) Specify initial sample δ 0 c ,, cn,, 1, 2, m Set t=0

35 Solution Algorithm STEP (1) At step t, sample in turn t i (for i = 1: N+4) Calculate * t * t p T,, * t i i i i i, i min1, t t t t p T, i i i, i t t1 t1 t t,,,,, N Where i i 1 i1 is the latest updated variables except 1 4 t i t1 * Accept with probability Otherwise, set i i t1 t i i *, i t i

36 Solution Algorithm STEP (2) t T t t1 If : set Go to STEP (1) Else: Stop iteration

37 Numerical Example: MRT Network in Singapore MCMC provides distribution of unknown parameters rather than one value Burn-in: 5000 steps Effective sample: ( draws in total) Standard deviation for Gaussian random walk Metropolis proposals t 2 ~ 0, i i * i i i

38 Numerical Example MCMC provides distribution of unknown parameters rather than one value 30.8%, % Index %

39 Numerical Example In this example 15 Posterior Prior Our prior knowledge is inaccurate the large number of travel time observations has corrected it P (c1) 10 5 Transfer Bouna Vista Stn P(c107) c 107 Posterior Prior c 1 Travel time from EW1 to EW2

40 Numerical Example Flow assignment based on route choice model Direction 1 Direction 2

41 Conclusion An integrated statistical model on travel time reliability and route choice behavior A metro network in which only travel time is observed Bayesian inference framework to formulate posterior probability Given the high-dimension of parameters, variable-at-a-time Metropolis sampling algorithm is applied to obtain posterior distribution.

42 Conclusion An integrated statistical model on travel time reliability and route choice behavior A metro network in which only travel time is observed Bayesian inference framework to formulate posterior probability Given the high-dimension of parameters, variable-at-a-time Metropolis sampling algorithm is applied to obtain posterior distribution. With this framework, we characterized travel time and its variation on each link. Meanwhile, we also identified contribution of different factors in determining passenger route choice behavior/movement.

43 Discussion & Outlook Most metro systems are closed environments, which only register transactions when passengers enter and leave the system; as a result, route choice (interchange/transfer) and service reliability are not captured in smart card data Our framework does not require an specific route choice model; thus, it can be applied on a more sophisticated model which takes more factors into account, helping us to further understand passenger behavior and build advanced flow assignment models, and further infer individual train load Although Singapore s network is simple, this framework shows great potential in applying on more complex metro networks, such as London Underground Identifying critical/crowding location/facility in metro network

44 Thank you! Lijun SUN Future Cities Laboratory, Singapore-ETH Centre

ANALYSING ROUTE CHOICE DECISIONS ON METRO NETWORKS

ANALYSING ROUTE CHOICE DECISIONS ON METRO NETWORKS ABSTRACT Sebastián Raveau Department of Transport Engineering and Logistics Pontificia Universidad Católica de Chile Avenida Vicuña Mackenna 4860, Santiago,