A Multicommodity Flow Approach for a Routing Problem with Time Synchronization Constraint Alain QUILLIOT
Loïc YON
{alain.quilliot, loic.yon}@isima.fr LIMOS, UMR 6158 - CNRS Université Blaise Pascal Clermont-Ferrand - France
CO 2004, 28-30 March
Introduction Related Problems MODT Problem Conclusion
Introduction
Modeling of Transportation Problems Configuration of a transportation offer for a bus company in an urban network I
cost for the bus company
I
customer satisfaction (Quality of Service)
Alain QUILLIOT, Loïc YON
CO 2004 - The MODT Problem
1/21
Introduction Related Problems MODT Problem Conclusion
Contents Introduction Related Problems Vehicle Routing Problem Network Design
MODT Problem Problem definition Resolution Schemes Hierarchical decomposition Auxiliary problem
Conclusion Numerical Experiments Alain QUILLIOT, Loïc YON
CO 2004 - The MODT Problem
2/21
Introduction Related Problems MODT Problem Conclusion
Public transportation
Quality of Service :
Routes : I
shape, duration, labor constraints
I
heterogeneous fleet : capacity, autonomy
I
areas : political, commercial
I
frequencies
Alain QUILLIOT, Loïc YON
I
demand
I
riding/travel time
I
frequency
I
price
CO 2004 - The MODT Problem
3/21
Introduction Related Problems MODT Problem Conclusion
Vehicle Routing Problem Network Design
School Bus Routing Problem "Providing public transportation service for students from their home to their school" I
equity
I
efficiency
I
lowest costs
A Multiobjective Optimization Approach to Urban School Routing: Formulation and Solution Method, BOWERMAN, HALL, CALAMAI, Transportation Research, 1995 A Computerized Approach to the New-York City School Bus Routing Problem, BRACA, BRAMEL, POSNER, SIMCHI-LEVI, IIE Transactions, 1997 Decision-aid Methodology for the School Bus Routing and Scheduling Problem, SPADA, BIERLAIRE, LIEBLING, 3rd STRC, 2003
Alain QUILLIOT, Loïc YON
CO 2004 - The MODT Problem
4/21
Introduction Related Problems MODT Problem Conclusion
Vehicle Routing Problem Network Design
Dial-a-Ride Problem "Designing routes and schedules for vehicles to pick up or drop off people and/or goods" (Ex: disabled or elderly people transportation) I
best possible service
I
lowest costs
(PDVRP/VRPTW extension)
The Dial-a-Ride Problem: Variants, Modeling Issues and Algorithms, CORDEAU, LAPORTE, 4-OR, 2003
Alain QUILLIOT, Loïc YON
CO 2004 - The MODT Problem
5/21
Introduction Related Problems MODT Problem Conclusion
Vehicle Routing Problem Network Design
Capacity & Flow Assignment Problem "Designing topology for packet-switch networks with QoS constraints" I
capacity requirements
I
lowest costs
trade-off between investment costs and congestion Minimum Cost Capacity Installation for Multicommodity Network Flows, BIENSTOCK, CHOPRA, GÜNLÜCK, TSAI, Mathematical Programming, 1998 Capacity and Flow assignment of data networks by generalized Benders decomposition, MAHEY, BENCHAKROUN, BOYER, Journal of Global Optimization, 2001
Alain QUILLIOT, Loïc YON
CO 2004 - The MODT Problem
6/21
Introduction Related Problems MODT Problem Conclusion
Problem definition Resolution Schemes Hierarchical decomposition Auxiliary problem
Passengers behavior pattern
SP-mode
Alain QUILLIOT, Loïc YON
Realistic mode
CO 2004 - The MODT Problem
7/21
Introduction Related Problems MODT Problem Conclusion
Problem definition Resolution Schemes Hierarchical decomposition Auxiliary problem
Notations G = (V , E ) directed graph E = A ∪ A where A is the set of "fast" arcs F integer flow f fractional multiflow P Sum(f ) = e∈E fe (ce ), (pe ) costs `(e) length of arc e k k (MAXe ), (Cmin ),(Cmax ) capacities e e o k , d k , D k for each commodity bvk for flow conservation constraints
Alain QUILLIOT, Loïc YON
CO 2004 - The MODT Problem
8/21
Introduction Related Problems MODT Problem Conclusion
Problem definition Resolution Schemes Hierarchical decomposition Auxiliary problem
Multi-Origin/Destination Transportation Problem minimize c.F + p.Sum(f ) subject to
X
F ≤ MAX
(1)
Cmin ≤ f ≤ Cmax X Fe − Fe = 0
(2)
e∈ω − (v )
X e∈ω − (v )
∀v ∈ V
(3)
∀v ∈ V , ∀k ∈ K
(4)
∀e ∈ A
(5)
e∈ω + (v )
fek −
X
fek = bvk
e∈ω + (v )
dSum(f )e e ≤ Fe F ∈N
(6)
+
(7)
f ∈R Alain QUILLIOT, Loïc YON
CO 2004 - The MODT Problem
9/21
Introduction Related Problems MODT Problem Conclusion
Problem definition Resolution Schemes Hierarchical decomposition Auxiliary problem
Application (1) K commodities (o k , d k , D k , tfk , Lkf ) I
D k : demand
I
tfk : maximal time for arrival at destination
I
Lkf : maximal length of the travel
1 depot : each tour begins and ends at this depot homogenous fleet of vehicles design and schedule a route system for a shuttle fleet I
minimal cost : c.F + p.f
I
demands
Alain QUILLIOT, Loïc YON
CO 2004 - The MODT Problem
10/21
Introduction Related Problems MODT Problem Conclusion
Problem definition Resolution Schemes Hierarchical decomposition Auxiliary problem
Application (2) : Dynamic Graph
temporal discrete dynamic graph: 0 −→ N.δ each node is duplicate to correspond to an instant i ∈ {0, . . . , N} special fictitious nodes D and U arc [x, y ] ∈ E becomes [xr , yr +`∗ (e) ], ∀0 ≤ r ≤ N − `∗ (e) `(e) e where `∗ (e) = d δ Drawback : this graph can be really big.
Alain QUILLIOT, Loïc YON
CO 2004 - The MODT Problem
11/21
Introduction Related Problems MODT Problem Conclusion
Problem definition Resolution Schemes Hierarchical decomposition Auxiliary problem
Several approaches
NP-hard I
Lagrangian relaxation
I
Hierarchical decomposition
I
Metaheuristics (GRASP, Tabu)
⇒ all but metaheuristics lead to an auxiliary problem
Alain QUILLIOT, Loïc YON
CO 2004 - The MODT Problem
12/21
Introduction Related Problems MODT Problem Conclusion
Problem definition Resolution Schemes Hierarchical decomposition Auxiliary problem
Lagrangian Relaxations
I
integrity constraints
I
flow constraints (*)
I
coupling constraints (*)
if MAX = ∞, all optimal values are equal
Alain QUILLIOT, Loïc YON
CO 2004 - The MODT Problem
13/21
Introduction Related Problems MODT Problem Conclusion
Problem definition Resolution Schemes Hierarchical decomposition Auxiliary problem
Master Slave Decomposition Heuristic Scheme Initializing F et f , feasible for MODT Stop ← false while not (Stop) do Solving MODTf Extracting dual vector λ (coupling constraints) Evaluating ∆ = c.F + p.Sum(f ) Solving|Improving P-aux(λ) if f is unchanged OR ∆ not improved during a given number of iterations then Stop ← true end if end while
Alain QUILLIOT, Loïc YON
CO 2004 - The MODT Problem
14/21
Introduction Related Problems MODT Problem Conclusion
Problem definition Resolution Schemes Hierarchical decomposition Auxiliary problem
P-aux(λ,f) (1)
minimize λ. dSum(f )e + p.Sum(f ) subject to
X e∈ω − (v )
Cmin ≤ f ≤ Cmax X fek − fek = bvk
(8) ∀v ∈ V , ∀k ∈ K
(9)
e∈ω + (v )
Sum(f )e ≤ Fe f ∈R
Alain QUILLIOT, Loïc YON
∀e ∈ A
+
CO 2004 - The MODT Problem
(10) (11)
15/21
Introduction Related Problems MODT Problem Conclusion
Problem definition Resolution Schemes Hierarchical decomposition Auxiliary problem
P-aux(λ,f) (2)
NP-hard (location problem) Special case of P-aux : f is a single flow (or all components of f can be treated separately) solved by finding iteratively a cycle improving the objective function in the residual graph : negative cycle canceling (CYGEN)
Alain QUILLIOT, Loïc YON
CO 2004 - The MODT Problem
16/21
Introduction Related Problems MODT Problem Conclusion
Problem definition Resolution Schemes Hierarchical decomposition Auxiliary problem
Special Case : Aggregated flow
transportation application : commodity set really big D k small compared to total demand Assumption : each component f is routed on a single path only possible only if G is strong connected f → Sum(f ) ???
f ← Sum(f )
Alain QUILLIOT, Loïc YON
CO 2004 - The MODT Problem
17/21
Introduction Related Problems MODT Problem Conclusion
Problem definition Resolution Schemes Hierarchical decomposition Auxiliary problem
Metric inequalities for a flow g Given Z ⊂ V , OD(Z ) = {k ∈ K /ok ∈ Z et dk ∈ / Z} ∀Z ⊂ V ,
X e∈ω + (Z )
X
ge ≥
Dk
k∈OD(Z )
I
necessary conditions for Sum(f ) → f
I
cuts (Benders) injected in CYGEN and MSD scheme
On feasibility conditions of multicommodity flows in networks, ONAGA, KALUSHO, IEEE Transactions on Circuit Theory, 1971
Alain QUILLIOT, Loïc YON
CO 2004 - The MODT Problem
18/21
Introduction Related Problems MODT Problem Conclusion
Numerical Experiments
Numerical Experiments 20 instances G = (V , E ) 10 ≤ |V | ≤ 20 2|V | ≤ |E | ≤ 4|V | same results for MSD and relaxations on 12 instances I
Decomposition: 5%
I
Coupling constraints relaxation: 3%
I
Flow constraint relaxation: 3%
CYGEN : 4%
Alain QUILLIOT, Loïc YON
CO 2004 - The MODT Problem
19/21
Introduction Related Problems MODT Problem Conclusion
Numerical Experiments
Conclusion & Further work
I
instances are too small
I
→ bigger instances with metaheuristics
I
more tests on CYGEN with the aggregation process
I
integration of travel patterns, elastic demand
Alain QUILLIOT, Loïc YON
CO 2004 - The MODT Problem
20/21
Introduction Related Problems MODT Problem Conclusion
Numerical Experiments
Clermont-Ferrand city 80
81
47 83
82
6 46
77 48 76
42 45
41
16
49 75 5
17
43 78
18
73
50
74
40
19
39
15
44
51
92 91 20
37
38 14
4
89
52
87
88
21 79
13 36 53
57
54 55
58 90
56
22
72
12
35
3 71
59
61
60 23
62
34
70
11
65
66
63
69 67
64
2
68
24
1
33 10 25
84
9 85 86 32 8
7
26
31
27
30 28 29
Alain QUILLIOT, Loïc YON
Nodes Arcs Commodities
CO 2004 - The MODT Problem
92 304 992
21/21