Upload
others
View
3
Download
0
Embed Size (px)
Citation preview
General rights Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.
• Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain • You may freely distribute the URL identifying the publication in the public portal
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.
Downloaded from orbit.dtu.dk on: Dec 17, 2017
A Greedy Construction Heuristic for the Liner Shipping Network Design Problem
Brouer, Berit Dangaard
Publication date:2010
Document VersionEarly version, also known as pre-print
Link back to DTU Orbit
Citation (APA):Løfstedt, B. (2010). A Greedy Construction Heuristic for the Liner Shipping Network Design Problem[Sound/Visual production (digital)]. Optimization Days 2010, Montréal, Canada, 10/05/2010
A greedy construction heuristic for the LinerService Network Design Problem
Berit Løfstedt
Operations ResearchDepartment of Management Engineering
Technical University of Denmark
May 5, 2010
Outline(jg
1 The Liner Service Network Design Problem (LS-NDP)2 Methods based on integer and linear programming relaxations3 LS-NDP as a multilayered Multiple Quadratic Knapsack Problem4 The greedy construction heuristic5 Critique of model and method6 Future work
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 2/22
Outline(jg
1 The Liner Service Network Design Problem (LS-NDP)2 Methods based on integer and linear programming relaxations3 LS-NDP as a multilayered Multiple Quadratic Knapsack Problem4 The greedy construction heuristic5 Critique of model and method6 Future work
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 2/22
Outline(jg
1 The Liner Service Network Design Problem (LS-NDP)2 Methods based on integer and linear programming relaxations3 LS-NDP as a multilayered Multiple Quadratic Knapsack Problem4 The greedy construction heuristic5 Critique of model and method6 Future work
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 2/22
Outline(jg
1 The Liner Service Network Design Problem (LS-NDP)2 Methods based on integer and linear programming relaxations3 LS-NDP as a multilayered Multiple Quadratic Knapsack Problem4 The greedy construction heuristic5 Critique of model and method6 Future work
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 2/22
Outline(jg
1 The Liner Service Network Design Problem (LS-NDP)2 Methods based on integer and linear programming relaxations3 LS-NDP as a multilayered Multiple Quadratic Knapsack Problem4 The greedy construction heuristic5 Critique of model and method6 Future work
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 2/22
Outline(jg
1 The Liner Service Network Design Problem (LS-NDP)2 Methods based on integer and linear programming relaxations3 LS-NDP as a multilayered Multiple Quadratic Knapsack Problem4 The greedy construction heuristic5 Critique of model and method6 Future work
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 2/22
Outline(jg
1 The Liner Service Network Design Problem (LS-NDP)2 Methods based on integer and linear programming relaxations3 LS-NDP as a multilayered Multiple Quadratic Knapsack Problem4 The greedy construction heuristic5 Critique of model and method6 Future work
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 2/22
Problem definition(jg
The Liner shipping network design problemGiven a complete graph G′ between a set of ports P, a fleet dividedinto vessel classes A and a set of commodities K determine aminimum cost network G = (V ,E) consisting of disjoint non-simplecyclic vessel routes to transport the most profitable subset of thecommodities.
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 3/22
Characteristics of a service(jg
Figure: Example of a single service
CyclicNon-simpleInbound vs. outbound direction
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 4/22
Characteristics of a network(jg
Figure: Network design
Transhipment of cargo at transhipment hubs and main portsCapacity classes: feeder, panamax, super panamaxFixed schedule -mainly based on weekly port visits
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 5/22
Selection of previous work(jg
Figure: Transhipment of cargo
Focus:Multiple routings (i.e.network design)Multiple hubs
Relevant literature:#models = #articlesMain difference:transhipment
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 6/22
Previous work(jg
Article Method Optimal Transhipment vessels/ports[1] Lagrange, Benders No No 3v, 20p[2] Branch-&-Cut Yes Yes, handling cost per container 6v, 20p[3] greedy, column generation, Benders No Yes, no cost 50v, 10p[4] tabu search, LP solver No Yes, individual cost per container 100v, 120p
Table: Overview of main articles with multiple route construction
[1]: Rana & Vickson 1991[2]: Reinhardt & Kallehauge 2007[3]: Agarwal & Ergun 2008[4]: Alvarez 2009
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 7/22
Going global....(jg
Challenges
Scaling to a global liner shipping network200+ ports, 200+ vessels
Scalability Issues:
Symmetry:Cyclic RoutingVessel Specs
Large scalemulticommodity flow
problem
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 8/22
Motivation(jg
Good solutions to the liner shipping network design problem
Competitive networkLow cost networkInclusion of dynamic non-linear bunker cost calculationNo optimality guarantee
Sheet1
Page 1
14 16 18 20 22 24 26 28
0
100
200
300
400
500
600
Bunker curve
non linear function
Mton/day
Knots
MTons
/day
Figure: Fictious example of non linear bunker curve
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 9/22
Work in progress...(jg
Create a good model including bunker costBuild a local search framework (ALNS)
Combining sets of:
1 Construction Heuristics2 Destruction Heuristics
Topic of this talk:First building block:
1 Greedy construction heuristic2 Based on a simplified LS-NDP model with simplified cost structures
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 10/22
Work in progress...(jg
Create a good model including bunker costBuild a local search framework (ALNS)Combining sets of:
1 Construction Heuristics2 Destruction Heuristics
Topic of this talk:First building block:
1 Greedy construction heuristic2 Based on a simplified LS-NDP model with simplified cost structures
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 10/22
Work in progress...(jg
Create a good model including bunker costBuild a local search framework (ALNS)Combining sets of:
1 Construction Heuristics2 Destruction Heuristics
Topic of this talk:
First building block:
1 Greedy construction heuristic2 Based on a simplified LS-NDP model with simplified cost structures
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 10/22
Work in progress...(jg
Create a good model including bunker costBuild a local search framework (ALNS)Combining sets of:
1 Construction Heuristics2 Destruction Heuristics
Topic of this talk:First building block:
1 Greedy construction heuristic
2 Based on a simplified LS-NDP model with simplified cost structures
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 10/22
Work in progress...(jg
Create a good model including bunker costBuild a local search framework (ALNS)Combining sets of:
1 Construction Heuristics2 Destruction Heuristics
Topic of this talk:First building block:
1 Greedy construction heuristic
2 Based on a simplified LS-NDP model with simplified cost structures
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 10/22
Work in progress...(jg
Create a good model including bunker costBuild a local search framework (ALNS)Combining sets of:
1 Construction Heuristics2 Destruction Heuristics
Topic of this talk:First building block:
1 Greedy construction heuristic2 Based on a simplified LS-NDP model with simplified cost structures
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 10/22
Model simplifications(jg
Rephrase the problem:
1 A set of routes2 Place port calls on
routes
Avoid evaluatinga large scale
multicommodity flowproblem
Multiple QuadraticKnapsack Problem
(MQKP)Routes=Knapsacks
Port calls=items
Profit function, f :f (distance,
demand ,
transhipment)
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 11/22
Model simplifications(jg
Rephrase the problem:
1 A set of routes
2 Place port calls onroutes
Avoid evaluatinga large scale
multicommodity flowproblem
Multiple QuadraticKnapsack Problem
(MQKP)Routes=Knapsacks
Port calls=items
Profit function, f :f (distance,
demand ,
transhipment)
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 11/22
Model simplifications(jg
Rephrase the problem:
1 A set of routes2 Place port calls on
routes
Avoid evaluatinga large scale
multicommodity flowproblem
Multiple QuadraticKnapsack Problem
(MQKP)Routes=Knapsacks
Port calls=items
Profit function, f :f (distance,
demand ,
transhipment)
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 11/22
Model simplifications(jg
Rephrase the problem:
1 A set of routes2 Place port calls on
routes
Avoid evaluatinga large scale
multicommodity flowproblem
Multiple QuadraticKnapsack Problem
(MQKP)Routes=Knapsacks
Port calls=items
Profit function, f :f (distance,
demand ,
transhipment)
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 11/22
Model simplifications(jg
Rephrase the problem:
1 A set of routes2 Place port calls on
routes
Avoid evaluatinga large scale
multicommodity flowproblem
Multiple QuadraticKnapsack Problem
(MQKP)Routes=Knapsacks
Port calls=items
Profit function, f :f (distance,
demand ,
transhipment)
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 11/22
Model simplifications(jg
Rephrase the problem:
1 A set of routes2 Place port calls on
routes
Avoid evaluatinga large scale
multicommodity flowproblem
Multiple QuadraticKnapsack Problem
(MQKP)Routes=Knapsacks
Port calls=items
Profit function, f :f (distance,
demand ,
transhipment)
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 11/22
Layer characteristics(jg
Layer Port types Distances Direct Transportto Hub
Weeks
FeederSpokes Short secondary primary 1-3
Main portsHubs
Panamax Main ports Medium primary secondary 3-8Hubs
Super Main ports Long secondary primary 6-12panamax Hubs
Table: Layer classification
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 12/22
Layer characteristics(jg
Layer Port types Distances Direct Transportto Hub
Weeks
FeederSpokes Short secondary primary 1-3
Main portsHubs
Panamax Main ports Medium primary secondary 3-8Hubs
Super Main ports Long secondary primary 6-12panamax Hubs
Table: Layer classification
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 12/22
Layer characteristics(jg
Layer Port types Distances Direct Transportto Hub
Weeks
FeederSpokes Short secondary primary 1-3
Main portsHubs
Panamax Main ports Medium primary secondary 3-8Hubs
Super Main ports Long secondary primary 6-12panamax Hubs
Table: Layer classification
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 12/22
Layer characteristics(jg
Layer Port types Distances Direct Transportto Hub
Weeks
FeederSpokes Short secondary primary 1-3
Main portsHubs
Panamax Main ports Medium primary secondary 3-8Hubs
Super Main ports Long secondary primary 6-12panamax Hubs
Table: Layer classification
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 12/22
Multilayered algorithm(jg
Figure: Multi layered knapsack interpretation of the LS-NDP
Three layers: feeder, panamax and super panamaxPort items: Scheduled port visitsEach layer may have multiple visits to a port
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 13/22
Solve an MQKP for each layer(jg
i 0 1 20 0 287 3061 -25 42 7422 14 513 0
Table: Profit matrix
Vlayer :items (scheduled port calls with the capacity class of thislayer)Rlayer : knapsacks (Services)Services are assigned a standard number of vesselsNumber of vessels = Duration in weeks
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 14/22
Specialised MQKP - Mathematical model(jg
maximize(MQKP) = ∑r∈R
∑i∈V
∑j∈V
pij xri x r
j + ∑r∈R
∑j∈V
pj xrj
subject to: ∑r∈R
x ri = 1 ∀i ∈ V (Mutually exclusive)
x ri x r
j ≥ y rij ∀i ∈ V , j ∈ V , r ∈R (Activate edge variable)
∑j∈V
y rij − ∑
j∈Vy r
ji = 0 ∀i ∈ V , r ∈R (Cyclic)
∑j∈V
y rij ≤ 1 ∀i ∈ V , r ∈R (Simple)
uri −ur
j +y rij ∑
i∈Vx r
i ≤ ∑i∈V
x ri −1 ∀i ∈ V , j ∈ V , r ∈R (Connected)
∑i∈V
∑j∈V
y rij (tij + ti )≤ σ(Ca) ∀r ∈Ra ,a ∈A (Duration)
x ri ∈ {0,1} ∀i ∈ V , r ∈R
y rij ∈ {0,1} ∀i ∈ V , j ∈ V , r ∈R
uri ∈Z + ∀i ∈ V , r ∈R
Quadratic objective function - heuristic solution method
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 15/22
Greedy parallel insertion(jg
The football teaming principleThe knapsacks take turn at choosing the most profitable item amongthe remaining items
Principle: parallel insertionMotivation: Distribution of difficult items
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 16/22
Algorithm(jg
GREEDYCONSTRUCTION (instance)
1 layers← FLEETTOLAYERS(instance)2 SCHEDULETOITEMS(instance, layers)3 profitIncrease← TRUE4 for each layer ∈ layers5 do MAKEKNAPSACKS()6 while (Vlayer 6= /0∪profitIncrease )7 do profitIncrease← FALSE8 for each r ∈ Rlayer9 best ← NULL
10 bestValue← 011 for each i ∈ Vlayer12 deltaValue← ∑j∈r pij13 if (deltaValue > bestValue)
then14 bestValue← deltaValue15 best ← i16 if (bestValue > 0)
then17 profitIncrease← TRUE18 UPDATEDEMANDMATRICES(knapsack , best )19 r ← best20 Vlayer ← Vlayer \best
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 17/22
Results(jg
Solve an instance of 234 ports and roughly 14000 demands in 33secondsEvaluated by Network specialists at Maersk Line
1 The routings are overall realistic2 Emphasis on direct transportation3 Transhipment fascilities are weak4 Good basis for a local search
Conclusion:Good construction heuristic as initial solution for further local search
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 18/22
Critique of the approach(jg
Not based on the true objective i.e. the MCF problemLittle interaction between layersOnly tested on a single instance of the Maerskline networkNo transhipment cost, bunker cost or vessel deployment costNote: Integration in ALNS will provide evaluation of true cost
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 19/22
Future work for MQKP heuristic(jg
Interaction between layersMore realistic goal function
1 Solve uncapacitated MCF2 Evaluate the transit times and the potential throughput
Test on real life data (Benchmark suite in progress)Compare results to the network cost of the initial schedule
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 20/22
Future work for ALNS framework(jg
Fast delta evaluation of multi commodity flow problemDestruction/ construction heuristicsBenchmark suite for Liner shipping
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 21/22
Bibliography(jg
[K. Rana and R.G. Vickson]“Routing container Ships Using Lagrangian Relaxation and Decomposition”Transportation Science 25, 201-214 (1991)
[L. B. Reinhardt and B. Kallehauge]“Network Design for Container Shipping Using Cutting Planes”Conference on Maritime & Intermodal Logistics, Singapore, December 2007
[R. Agarwal and O. Ergun]”Ship Scheduling and Network Design for Cargo routing in Liner Shipping”Transportation Science 42, 175-196 (2008).
[J.F. Alvarez]“ Joint Routing and Deployment of a Fleet of Container Vessels”Maritime Economics and Logistics 11, 186-208 (2009)
[S. Røpke and D. Pisinger],“A general heuristic for vehicle routing problems”,Computers and Operations Research 34, 2403-2435 (2007)
[I. Akio and K. Shintani and S.Papadimitriou]”Multi-port vs. Hub-and-Spoke port calls by containerships”Transportation Research Part E 42 , 740-757 (2009)
[D. Pisinger, S. Ropke]”Large Neighborhood Search”
Book Chapter, http://www.diku.dk/hjemmesider/ansatte/sropke/Papers/lns.pdf
Berit Løfstedt (DTU Management) LS-NDP May 5, 2010 22/22