The design of Demand-Adaptive public transportation Systems: Meta-Schedules

Transcription

1 The design of Demand-Adaptive public transportation Systems: Meta-Schedules Gabriel Teodor Crainic Fausto Errico ESG, UQAM and CIRRELT, Montreal Federico Malucelli DEI, Politecnico di Milano Maddalena Nonato Università di Ferrara JOURNE SCIENTIFIQUE Chaire de recherche industrielle CRSNG en management logistique ESG UQAM The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.1/29

2 Outline 1. Demand Responsive transit Systems (e.g., DAR) 2. Demand Adaptive transit Systems (DAS) 3. Scheduling issues in DAS, DAR, and Traditional transit 4. The Meta-Schedule problem single segment subproblem - sampling-based technique sewing segments together 5. Conclusions The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.2/29

3 Flexibility in Transit Systems Offer competitive transportation services Capture additional demand Better address population needs Cover larger/additional areas Sustainability Reduce operating costs Increase resource utilization Integration with traditional transportation systems User point of view Management point of view The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.3/29

4 Dial a Ride Systems Users ask for personalized rides (door-to-door service) But are served collectively Similar to a collective taxi service Initially devised to meet needs of users with reduced mobility Extended (somewhat) to deal with low demand areas or periods Quite expensive compared to regular service Frequency? Regularity? The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.4/29

5 Demand Adaptive System (DAS) Combine DAR flexibility to traditional system regularity and low-cost Compulsory stops with time windows regular line Optional stops on request (active users) Segments: Set of optional stops between two consecutive compulsory stops Users can still access the service at compulsory stops (passive users) The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.5/29

6 A DAS line Passive users only The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.6/29

7 A DAS line Active and passive users The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.7/29

8 Integrated System DAS as feeder lines The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.8/29

9 Multiple Lines Transfer points among flexible (and traditional) lines (compulsory stops) The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.9/29

10 Schedules: Traditional Transit and DAR Traditional Transit Passing time at each stop of the line Designed for medium-term periods (six months, one year) Users plan their trips based on published schedules Tactical planning decision (line definition: higher level ) DAR Particular to each vehicle tour Varying according to the actual requests Operational planning activity The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.10/29

11 Scheduling Issues for DAS Combines planning phases of traditional transit and DAR Two schedules are built Meta-Schedule: Similar to Traditional Transit Basic line definition: time windows at compulsory stops Users plan trips based on the published Meta-Schedule Tactical planning decision (compulsory stops, segments, frequencies at higher level) Each departure schedule: Similar to DAR Defines the actual vehicle itinerary Varies according to actual requests Must be compatible with the Master Schedule Operational planning activity The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.11/29

12 Basic DAS Design Problem Input The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.12/29

13 Basic DAS Design Problem Input Region to serve The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.12/29

14 Basic DAS Design Problem Input Region to serve Demand The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.12/29

15 Basic DAS Design Problem Input Region to serve Demand Decisions The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.12/29

16 Basic DAS Design Problem Input Region to serve Demand Decisions Compulsory stops The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.12/29

17 Basic DAS Design Problem Input Region to serve Demand Decisions Compulsory stops Sequencing The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.12/29

18 Basic DAS Design Problem Input Region to serve Demand Decisions Compulsory stops Sequencing Segments The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.12/29

19 Basic DAS Design Problem Input Region to serve Demand Decisions Compulsory stops Sequencing Segments Time windows The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.12/29

20 Basic DAS Design Problem Input Region to serve Demand Decisions Compulsory stops Sequencing Segments Time windows Goals Low routing costs Hight level of service The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.12/29

21 Time Windows The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.13/29

22 Time Windows Policy Given a compulsory stop and its time window [a, b] The Vehicle Must leave the compulsory stop within the time interval [a, b] May arrive before a a b The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.13/29

23 Time Windows Consequences Passive users must be at the compulsory stop not later than a Passengers on the bus may experience idle times Time windows represent bounds on user travel time a b The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.13/29

24 Time Windows Main Features Distance Width distance width width time The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.13/29

25 Time Windows Main Features Distance Time to serve the optional stop Idle times at compulsory stops Width distance width width time The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.13/29

26 Time Windows Main Features Distance Time to serve the optional stop Idle times at compulsory stops Width Flexibility Long waiting times for passive users distance width width time The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.13/29

27 Meta-Schedule: problem description Input Topological design Demand for transportation Policy for time windows The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.14/29

28 Meta-Schedule: problem description Input Topological design Demand for transportation Policy for time windows Output A time window for each compulsory stop The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.14/29

29 Meta-Schedule: problem description Input Topological design Demand for transportation Policy for time windows Output A time window for each compulsory stop Conflicting goals Provide sufficient time to serve optional stops Minimize total maximum time Small time windows Minimize idle times... The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.14/29

30 Problem Setting and Assumptions The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.15/29

31 Problem Setting and Assumptions Demand Probability distributions of requests for o/d pairs Probability of at least one request derived for each optional stop Serve the set of requests Serve the set of requested stops The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.15/29

32 Problem Setting and Assumptions Demand Probability distributions of requests for o/d pairs Probability of at least one request derived for each optional stop Serve the set of requests Serve the set of requested stops Time windows We consider time windows with common and fixed width δ Two possible easy extensions Fixed but different width Maximum width The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.15/29

33 Formaly Input A sequence of compulsory stops A set of optional stops partitioned into segments Travel time c ij for the pair of stops (i, j) in a segment Probability p i of being requested for service for optional stop i The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.16/29

34 Formaly Input A sequence of compulsory stops A set of optional stops partitioned into segments Travel time c ij for the pair of stops (i, j) in a segment Probability p i of being requested for service for optional stop i Output Time windows [a h, b h ] for compulsory stop f h, b h a h = δ It reduces to finding a sequence {b 0, b 1,...,b n } The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.16/29

35 Formaly Input A sequence of compulsory stops A set of optional stops partitioned into segments Travel time c ij for the pair of stops (i, j) in a segment Probability p i of being requested for service for optional stop i Output Time windows [a h, b h ] for compulsory stop f h, b h a h = δ It reduces to finding a sequence {b 0, b 1,...,b n } Goals Serve any requested optional stop with probability P ǫ Minimize b n The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.16/29

36 Subproblem: Single Segment f h 1 f h Input A segment The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.17/29

37 Subproblem: Single Segment f h 1 p i f h Input A segment Probability p i for optional stop i of being active The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.17/29

38 Subproblem: Single Segment f h 1 p i f h L h 1 Input A segment Probability p i for optional stop i of being active The departure time L h 1 at compulsory f h 1 The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.17/29

39 Subproblem: Single Segment f h 1 p i f h L h 1 Sh p Sh To any subset S h N h of optional stops is associated Probability p Sh of being active The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.17/29

40 Subproblem: Single Segment f h 1 p i f h L h 1 Sh p Sh T h = H h + L h 1 To any subset S h N h of optional stops is associated Probability p Sh of being active Service Time H h (active set) Hamiltonian Path? Arrival Time T h at the second compulsory f h The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.17/29

41 Subproblem: Single Segment f h 1 p i f h L h 1 Sh p Sh T h = H h + L h 1 To any subset S h N h of optional stops is associated Probability p Sh of being active Service Time H h (active set) Hamiltonian Path? Arrival Time T h at the second compulsory f h H h and T h are random variables The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.17/29

42 Subproblem: Single Segment f h 1 p i f h L h 1 S h (ω) T h (ω) = H h (ω) + L h 1 Goal The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.17/29

43 Subproblem: Single Segment f h 1 p i f h L h 1 S h (ω) T h (ω) = H h (ω) + L h 1 Goal Find the smallest value b h such that b h T h (ω) with probability 1 ǫ (to guarantee good service) The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.17/29

44 Subproblem: Single Segment f h 1 p i f h L h 1 S h (ω) T h (ω) = H h (ω) + L h 1 Goal Find the smallest value b h such that b h T h (ω) with probability 1 ǫ (to guarantee good service) We take b h = T 1 ǫ h P{H h (w) H 1 ǫ h = H 1 ǫ h + L h 1, with T 1 ǫ h, H 1 ǫ h } 1 ǫ and P{T h (w) T 1 ǫ h defined as } 1 ǫ The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.17/29

45 Subproblem: Single Segment f h 1 p i f h L h 1 S h (ω) T h (ω) = H h (ω) + L h 1 Crucial Point The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.17/29

46 Subproblem: Single Segment f h 1 p i f h L h 1 S h (ω) T h (ω) = H h (ω) + L h 1 Crucial Point How do we compute T 1 ǫ h or H 1 ǫ h (they differ by a constant)? The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.17/29

47 Subproblem: Single Segment f h 1 p i f h L h 1 S h (ω) T h (ω) = H h (ω) + L h 1 Crucial Point How do we compute T 1 ǫ h or H 1 ǫ h (they differ by a constant)? Cumulative Distribution Function (CDF) and Probability Mass Function (PMF) for H h (ω) and T h (ω) are needed The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.17/29

48 Subproblem: Single Segment f h 1 p i f h L h 1 S h (ω) T h (ω) = H h (ω) + L h 1 Crucial Point How do we compute T 1 ǫ h or H 1 ǫ h (they differ by a constant)? Cumulative Distribution Function (CDF) and Probability Mass Function (PMF) for H h (ω) and T h (ω) are needed Requires 2 N h service-time evaluations! The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.17/29

49 Subproblem: Single Segment f h 1 p i f h L h 1 S h (ω) T h (ω) = H h (ω) + L h 1 Crucial Point How do we compute T 1 ǫ h or H 1 ǫ h (they differ by a constant)? Cumulative Distribution Function (CDF) and Probability Mass Function (PMF) for H h (ω) and T h (ω) are needed Requires 2 N h service-time evaluations! We estimate the PMF by sampling The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.17/29

50 Sampling Algorithm Take {r 1, r 2,...,r l } random samples of relatively small cardinality Compute the PMF k, CDF k, and b k h for each sample r k Compute the mean value and standard deviation of b k h If the standard deviation is small, i.e.,the solution is precise, stop Otherwise, increment the cardinality of the samples and iterate The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.18/29

51 Sampling Algorithm Take {r 1, r 2,...,r l } random samples of relatively small cardinality Compute the PMF k, CDF k, and b k h for each sample r k Compute the mean value and standard deviation of b k h If the standard deviation is small, i.e.,the solution is precise, stop Otherwise, increment the cardinality of the samples and iterate Observations The algorithm is very simple. But No guarantee of unbiased solution The dimension of the sample may become critical The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.18/29

52 Experimentation Focus on the estimation of the service time for a single segment Square-shaped segments, side length 300 Compulsory stops at the extremities of one diagonal Optional stops are uniformly distributed on the square To each optional stop is associated a positive probability, 50% in average P ǫ = 0.95 Hamiltonian Path computed by our basic B&C code The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.19/29

53 Results A20 Longest 1-Path Hamiltonian Path Real b h (ε) Dim. nsamples dimsamples bh (ε) Std-dev/ b h (ε) Time(sec.) The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.20/29

54 Results A30 Longest 1-Path Hamiltonian Path Real b h (ε) Dim. nsamples dimsamples bh (ε) Std-dev/ b h (ε) Time(sec.) The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.21/29

55 Results A40 Longest 1-Path Hamiltonian Path Real b h (ε) Dim. nsamples dimsamples bh (ε) Std-dev/ b h (ε) Time(sec.) The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.22/29

56 Results A50 Longest 1-Path Hamiltonian Path Real b h (ε) Dim. nsamples dimsamples bh (ε) Std-dev/ b h (ε) Time(sec.) The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.23/29

57 Conclusions (1) The algorithm for estimating time windows is fast The number of samplings is not very important The cardinality of samples is important. Around 200 Method successful in solving problems with high dimension The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.24/29

58 Sewing Segments We introduce a new random variable The vehicle departure time L h (ω) at compulsory stop f h where a h L h (ω) b h T h (ω) = L h 1 (ω) + H h (ω) needed to compute [a h, b h ] T h (ω) and L h (ω) are linked by the relation L h (ω) = T h (ω) if ω a h T h (ω) b h ; a h if ω T h (ω) < a h ; b h if ω T h (ω) > b h. Observe that L h 1 (ω) and H h (ω) are independent T h (ω) can be computed by convoluting L h 1 (ω) and H h (ω) The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.25/29

59 Complete Scheme Input Sequence of segments G = 1,2,...,n G h Service probability P ǫ = (1 ǫ) n The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.26/29

60 Complete Scheme Input Sequence of segments G = 1,2,...,n G h Service probability P ǫ = (1 ǫ) n Algorithm 1. For each segment G h, h {1, 2,...,n} (a) Compute PMF and CDF of L h 1 (ω) (b) Compute PMF and CDF of H h (ω) (c) Compute PMF and CDF of T h (ω) as the convolution of the PMFs of L h 1 and H h (d) Compute T 1 ǫ h and set b h = T 1 ǫ h. The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.26/29

61 Complete Scheme Input Sequence of segments G = 1,2,...,n G h Service probability P ǫ = (1 ǫ) n Algorithm 1. For each segment G h, h {1, 2,...,n} (a) Compute PMF and CDF of L h 1 (ω) (b) Compute PMF and CDF of H h (ω) (c) Compute PMF and CDF of T h (ω) as the convolution of the PMFs of L h 1 and H h (d) Compute T 1 ǫ h Output and set b h = T 1 ǫ h. The best sequence {b 1, b 2,...,b n } Such that any randomly requested optional stop is served with probability P ǫ The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.26/29

62 Experimentation Goal evaluate the quality of the produced master schedules Given several: topological DAS line design origin destination demand distributions Compute several master schedules varying: time window width δ probabilities ǫ Simulate operations monitoring: percentage of rejected requests passive user waiting times idle times at compulsory... The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.27/29

63 Results ǫ δ b 4 avg(r/t) stdv(r/t) avg(i) stdv(i) avg(p) stdv(p) Master schedule evaluated on 4-segment line, low and high demand, 100 realizations The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.28/29

64 Conclusions (2) Demand Adaptive transit Systems DAS scheduling different from DAR and traditional transit The Meta-Schedule problem May be efficiently addressed by sampling (single segment) Results for general framework are satisfactory under different demand and line scenarios The design of Demand-Adaptive public transportation Systems: Meta-Schedules p.29/29