You are on page 1of 80

AMP

High level problems


Transportation
problem
Transshipment
problem

Management science 3 Multiple commodities


Summary
Low level problems
Shortest path problem

Transportation Problems Traveling salesman


problem
Vehicle routing
problem
Heterogeneous fleet
VRP
Feeder network design
Remy Spliet problem
Comparison
Uncertainty
Erasmus University Rotterdam Operational decisions
Tactical decisions
AMP

High level problems


Transportation
problem
Transshipment
problem
Multiple commodities
Summary
Low level problems
Shortest path problem

High level
Traveling salesman
problem
Vehicle routing
problem
Heterogeneous fleet
VRP

transportation problems Feeder network design


problem
Comparison
Uncertainty
Operational decisions
Tactical decisions
AMP
Transportation problem
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
Summary
Low level problems
Shortest path problem
Traveling salesman
problem
Vehicle routing
problem
Heterogeneous fleet
VRP
Feeder network design
problem
Comparison
Uncertainty
Operational decisions
Tactical decisions

I Moving goods from supply locations to demand locations is


called transportation.
AMP
Transportation problem
High level problems
Transportation
problem
I S: the set of supply locations Transshipment
problem
I si : total supply at location i ∈ S Multiple commodities
Summary
I D: the set of demand locations Low level problems
Shortest path problem
I dj : total demand at location j ∈ D Traveling salesman
problem
Vehicle routing
I cij : the costs of transporting one unit from supply location problem
Heterogeneous fleet
i ∈ S to demand location j ∈ D. VRP
Feeder network design
problem
Comparison

Decisions: Uncertainty
Operational decisions
I xij : the amount of goods transported from supply location Tactical decisions

i ∈ S to demand location j ∈ D.

Goal:
I Minimize the total transportation costs of satisfying all
demand.
AMP
Transportation problem
High level problems
Transportation
problem
Problem network: Transshipment
problem
Multiple commodities
Summary
3 Low level problems
Shortest path problem
Traveling salesman
c 13 problem

1 c14 Vehicle routing


problem
c1 Heterogeneous fleet
5 VRP
Feeder network design
problem
4 Comparison

c 23 Uncertainty
Operational decisions
c 24 Tactical decisions
2 c25

5
AMP
Transportation problem
High level problems
Transportation
problem
Transshipment
problem
Linear programming formulation: Multiple commodities
Summary
Low level problems
Shortest path problem
Traveling salesman
problem
Vehicle routing
X
min cij xij problem
Heterogeneous fleet
i∈S,j∈D VRP
X Feeder network design
problem
xij ≤ si ∀i ∈ S (Supply capacity) Comparison

j∈D Uncertainty
X Operational decisions
xij = dj ∀j ∈ D (Demand constraint) Tactical decisions

i∈S
xij ≥ 0 ∀i ∈ S, j ∈ D
AMP
Transportation problem
! The costs are incurred per unit transported. High level problems
Transportation
problem
! Transport of fractional units is allowed Transshipment
problem
Multiple commodities
Summary
X The model works best when in reality supply and demand are Low level problems
much greater than vehicle capacity. Shortest path problem
Traveling salesman
problem
Vehicle routing
Example: problem
Heterogeneous fleet
I Optimal solution: transport 1000.1 units from A to B. VRP
Feeder network design
I Vehicle capacity: 1 unit problem
Comparison
I Model costs of 1000.1 vehicles does not differ much from
Uncertainty
using 1001 vehicles in reality. Operational decisions
Tactical decisions

x The model does not work well if supply and demand is small
compared to vehicle capacity.

Example:
I Optimal solution: transport 0.1 units from A to B.
I Vehicle capacity: 1 unit
I Model costs of 0.1 vehicles differs substantially from 1 vehicle
in reality.
AMP
Transshipment problem
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
Summary
I Consider transporting goods from supply to demand locations Low level problems
using transshipment points. Shortest path problem
Traveling salesman
problem
Vehicle routing
problem
Example: Heterogeneous fleet
VRP
I Supply at U.S. ports Feeder network design
problem
Comparison
I Transport by ship from U.S. to European ports Uncertainty
Operational decisions
I Transshipment at European ports Tactical decisions

I Transportation by truck from European ports


I Demand at European inland destinations
AMP
Transshipment problem
High level problems
Transportation
problem
I Let S, si , D and dj be defined as before. Transshipment
problem
I T : the set of transshipment locations. Multiple commodities
Summary
I A: the arcs connecting i ∈ S to t ∈ T , and t ∈ T to j ∈ D. Low level problems
Shortest path problem
I cij : the costs of transporting one unit along arc (i, j) ∈ A. Traveling salesman
problem
Vehicle routing
* We introduce A to write down the LP more concisely: problem
Heterogeneous fleet
I (i, j) ∈ A may now refer to a connection i ∈ S and t ∈ T or VRP
Feeder network design
t ∈ T and j ∈ D problem
Comparison
Uncertainty
Operational decisions
Decisions: Tactical decisions

I xij : the amount of goods transported along arc (i, j) ∈ A.

Goal:
I Minimize the total transportation costs of satisfying all
demand.
AMP
Transshipment problem
High level problems
Transportation
problem
Problem network: Transshipment
problem
Multiple commodities
Summary
1 c14 6 Low level problems
c1 Shortest path problem
5 Traveling salesman
c 46 problem

4 c47 Vehicle routing


problem
c4 Heterogeneous fleet
VRP
8
c 24 Feeder network design
problem
2 c25 7 Comparison

c 56 Uncertainty
Operational decisions
c 57 Tactical decisions
5 c58
c 34
c 35
3 8
AMP
Transshipment problem
High level problems
Transportation
problem
Linear programming formulation: Transshipment
problem
Multiple commodities
Summary
Low level problems
Shortest path problem
X Traveling salesman
min cij xij problem
Vehicle routing
(i,j)∈A problem
X Heterogeneous fleet
VRP
xij ≤ si ∀i ∈ S (Supply capacity) Feeder network design
problem
(i,j)∈A Comparison
X X Uncertainty
xij = xji ∀j ∈ T (Transshipment flow) Operational decisions
Tactical decisions
(i,j)∈A (j,i)∈A
X
xij = dj ∀j ∈ D (Demand constraint)
(i,j)∈A
xij ≥ 0 ∀(i, j) ∈ A
AMP
Transshipment problem
High level problems
Transportation
I In case a transportation connection (i, j) ∈ A has a capacity problem
Transshipment
Qij , we add the constraint problem
Multiple commodities
Summary

xij ≤ Qij Low level problems


Shortest path problem
Traveling salesman
problem
Vehicle routing
problem
I In case a transshipment point j ∈ T has a processing capacity Heterogeneous fleet
VRP
of Qj units, we add the constraint Feeder network design
problem
Comparison
X
xij ≤ Qj Uncertainty
Operational decisions
(i,j)∈A Tactical decisions

I If we wish to connect a supply location i ∈ S and demand


location j ∈ S, we simply add (i, j) to A.
I In case a transportation connection (i, j) ∈ A can not be
used, we simply remove (i, j) from A.
AMP
Transshipment problem
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
Summary
Low level problems
! Suppose Shortest path problem
Traveling salesman
I T is empty problem
Vehicle routing
I All connections (i, j) for i ∈ S and j ∈ D are added to A problem
Heterogeneous fleet
then the transshipment problem reverts to the transportation VRP
Feeder network design
problem. problem
Comparison
Uncertainty
Operational decisions
X Hence, the transshipment problem is a generalization of the Tactical decisions
transportation problem.
AMP
Multiple commodity transshipment problem
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
Summary
Low level problems
Shortest path problem
Traveling salesman
problem
Vehicle routing
problem
Heterogeneous fleet
VRP
Feeder network design
problem
Comparison
Uncertainty
Operational decisions
Tactical decisions

I Consider the case where multiple commodities are


transported.
AMP
Multiple commodity transshipment problem
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
Summary
I K : the set of commodities. Low level problems
Shortest path problem
I Duplicate all variables and parameters of the transshipment Traveling salesman
problem
formulation for each commodity. Vehicle routing
problem
Example: Heterogeneous fleet
VRP
I xijk : the amount of commodity k ∈ K transported along arc Feeder network design
problem
(i, j) ∈ A. Comparison
Uncertainty
Operational decisions
Tactical decisions
I The total amount of goods transported along each arc
(i, j) ∈ A may not exceed the capacity Qij
AMP
Multiple commodity transshipment problem
High level problems
Linear programming formulation: Transportation
problem
Transshipment
problem
Multiple commodities
Summary
X X Low level problems
min cij xijk Shortest path problem
Traveling salesman
(i,j)∈A k∈K problem
X Vehicle routing
problem
xijk ≤ sik ∀i ∈ S, k ∈ K (Supply capacity) Heterogeneous fleet
VRP
(i,j)∈A Feeder network design
X X problem

xijk = xjik ∀j ∈ T , k ∈ K (Transshipment flow) Comparison


Uncertainty
(i,j)∈A (j,i)∈A Operational decisions
X Tactical decisions
xijk = djk ∀j ∈ D, k ∈ K (Demand constraint)
(i,j)∈A
X
xijk ≤ Qij ∀(i, j) ∈ A (Connection capacity)
k∈K
xijk ≥ 0 ∀(i, j) ∈ A, k ∈ K
AMP
Multiple commodity transshipment problem
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
Summary
Low level problems
Shortest path problem
X The multiple commodity transshipment problem is a Traveling salesman
problem
generalization of the transshipment problem (and therefore Vehicle routing
problem
also of the transportation problem). Heterogeneous fleet
VRP
Feeder network design
problem
! If the connection capacity constraints would not be imposed Comparison
Uncertainty
the problem decomposes into (single commodity) Operational decisions
transshipment problems, one for each commodity. Tactical decisions
AMP
Summary
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
X High level transportation problems can be modeled as linear Summary

programming problems. Low level problems


Shortest path problem
Traveling salesman
problem
Vehicle routing
! Remember: linear programming problems can typically be problem
Heterogeneous fleet
solved efficiently. VRP
Feeder network design
problem
Comparison
X Hence, large high level transportation problems can be solved Uncertainty

fast. Operational decisions


Tactical decisions

x High level transportation models do not take detailed routing


decisions and vehicle utilization into account.
AMP

High level problems


Transportation
problem
Transshipment
problem
Multiple commodities
Summary
Low level problems
Shortest path problem

Low level transportation


Traveling salesman
problem
Vehicle routing
problem
Heterogeneous fleet
VRP

problems Feeder network design


problem
Comparison
Uncertainty
Operational decisions
Tactical decisions
AMP
Shortest path problem
High level problems
Transportation
B 5 G problem
Transshipment
2 2 problem
Multiple commodities
2 E 5 Summary

4 Low level problems

1 4 5 Shortest path problem


Traveling salesman
A C 6 H J problem
3 Vehicle routing
problem
Heterogeneous fleet
F 5 1 VRP
Feeder network design
3 problem
Comparison

D I Uncertainty
Operational decisions
Tactical decisions

I Find the shortest path (SP) from the start to the end
location.
I Example: Find the shortest path from A to J in the above
network.
I This problem is occasionally referred to as the
shortest route problem.
AMP
Shortest path problem
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
Summary
I Assume all costs are positive. Low level problems
Shortest path problem
Traveling salesman
problem
X Dijkstra’s algorithm can be used to solve the shortest path Vehicle routing
problem
problem with positive costs. Heterogeneous fleet
VRP
Feeder network design
problem
Comparison
Notation: Uncertainty
Operational decisions
I Let cij be the costs of traveling from node i to node j. Tactical decisions

I Let Fi be the costs of the shortest path from the start node
to node i.
AMP
Shortest path problem
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
Summary
General outline of Dijkstra’s algorithm:
Low level problems
I Initialize Fi to be very high. Shortest path problem
Traveling salesman
problem
I Make a pass trough the network to update Fi . Vehicle routing
problem
I For the end node n, Fn will provide the costs of the shortest Heterogeneous fleet
VRP
path. Feeder network design
problem
Comparison
Uncertainty
Backtracking: Operational decisions
Tactical decisions
I Keep track of which node is the cheapest predecessor.
I This way the shortest path can be found by backtracking.
AMP
Shortest path problem
Dijkstra’s algorithm: High level problems
Step 1 Initialize Fi = ∞ for all nodes i in the network Transportation
problem
except the starting node 0, let F0 = 0. Transshipment
problem
Multiple commodities
Summary
Step 2 Select a node i with the lowest costs Fi among the Low level problems
nodes for which all predecessors have been Shortest path problem
Traveling salesman
”processed”. problem
Vehicle routing
problem
Heterogeneous fleet
Step 3 Select the cheapest arc (i, j) going out of the VRP
Feeder network design
current node. problem
Comparison
Uncertainty
Step 4 Update the costs Fj Operational decisions
Tactical decisions

Fj = min{Fj , Fi + cij }

I In case Fj was updated to Fi + cij , i is the


current cheapest predecessor of j.

Step 5 If all nodes are processed, stop. Otherwise return to


step 2.
AMP
Shortest path problem - example 1
High level problems
Transportation
problem
Transshipment
B 5 G problem
Multiple commodities
2 2 Summary
Low level problems
2 E 5 Shortest path problem
4 Traveling salesman

1 4 5
problem
Vehicle routing
A C 6 H J problem
Heterogeneous fleet
3 VRP
Feeder network design
problem
F 5 1 Comparison
3 Uncertainty
Operational decisions
D I Tactical decisions

Initialization:
Node A B C D E F G H I J
Fi 0 ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞
Pred. - - - - - - - - - -
AMP
Shortest path problem - example 1
High level problems
Transportation
problem
Transshipment
B 5 G problem
Multiple commodities
2 2 Summary
Low level problems
2 E 5 Shortest path problem
4 Traveling salesman

1 4 5
problem
Vehicle routing
A C 6 H J problem
Heterogeneous fleet
3 VRP
Feeder network design
problem
F 5 1 Comparison
3 Uncertainty
Operational decisions
D I Tactical decisions

Initialization:
Node A B C D E F G H I J
Fi 0 ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞
Pred. - - - - - - - - - -
AMP
Shortest path problem - example 1
High level problems
Transportation
problem
Transshipment
B 5 G problem
Multiple commodities
2 2 Summary
Low level problems
2 E 5 Shortest path problem
4 Traveling salesman

x 4 5
problem
Vehicle routing
A C 6 H J problem
Heterogeneous fleet
3 VRP
Feeder network design
problem
F 5 1 Comparison
3 Uncertainty
Operational decisions
D I Tactical decisions

Iteration 1:
Node A B C D E F G H I J
Fi 0 ∞ 1 ∞ ∞ ∞ ∞ ∞ ∞ ∞
Pred. - - A - - - - - - -
AMP
Shortest path problem - example 1
High level problems
Transportation
problem
Transshipment
B 5 G problem
Multiple commodities
2 2 Summary
Low level problems
x E 5 Shortest path problem
4 Traveling salesman

x 4 5
problem
Vehicle routing
A C 6 H J problem
Heterogeneous fleet
3 VRP
Feeder network design
problem
F 5 1 Comparison
3 Uncertainty
Operational decisions
D I Tactical decisions

Iteration 2:
Node A B C D E F G H I J
Fi 0 2 1 ∞ ∞ ∞ ∞ ∞ ∞ ∞
Pred. - A A - - - - - - -
AMP
Shortest path problem - example 1
High level problems
Transportation
problem
Transshipment
B 5 G problem
Multiple commodities
2 2 Summary
Low level problems
x E 5 Shortest path problem
4 Traveling salesman

x 4 5
problem
Vehicle routing
A C 6 H J problem
Heterogeneous fleet
x VRP
Feeder network design
problem
F 5 1 Comparison
3 Uncertainty
Operational decisions
D I Tactical decisions

Iteration 3:
Node A B C D E F G H I J
Fi 0 2 1 3 ∞ ∞ ∞ ∞ ∞ ∞
Pred. - A A A - - - - - -
AMP
Shortest path problem - example 1
High level problems
Transportation
problem
Transshipment
B 5 G problem
Multiple commodities
2 2 Summary
Low level problems
x E 5 Shortest path problem
x Traveling salesman

x 4 5
problem
Vehicle routing
A C 6 H J problem
Heterogeneous fleet
x VRP
Feeder network design
problem
F 5 1 Comparison
3 Uncertainty
Operational decisions
D I Tactical decisions

Iteration 4:
Node A B C D E F G H I J
Fi 0 2 1 3 5 ∞ ∞ ∞ ∞ ∞
Pred. - A A A C - - - - -
AMP
Shortest path problem - example 1
High level problems
Transportation
problem
Transshipment
B 5 G problem
Multiple commodities
2 2 Summary
Low level problems
x E 5 Shortest path problem
x Traveling salesman

x 4 5
problem
Vehicle routing
A C x H J problem
Heterogeneous fleet
x VRP
Feeder network design
problem
F 5 1 Comparison
3 Uncertainty
Operational decisions
D I Tactical decisions

Iteration 5:
Node A B C D E F G H I J
Fi 0 2 1 3 5 7 ∞ ∞ ∞ ∞
Pred. - A A A C C - - - -
AMP
Shortest path problem - example 1
High level problems
Transportation
problem
Transshipment
B x G problem
Multiple commodities
2 2 Summary
Low level problems
x E 5 Shortest path problem
x Traveling salesman

x 4 5
problem
Vehicle routing
A C x H J problem
Heterogeneous fleet
x VRP
Feeder network design
problem
F 5 1 Comparison
3 Uncertainty
Operational decisions
D I Tactical decisions

Iteration 6:
Node A B C D E F G H I J
Fi 0 2 1 3 5 7 ∞ ∞ ∞ ∞
Pred. - A A A C C - - - -
AMP
Shortest path problem - example 1
High level problems
Transportation
problem
Transshipment
B x G problem
Multiple commodities
2 2 Summary
Low level problems
x E 5 Shortest path problem
x Traveling salesman

x 4 5
problem
Vehicle routing
A C x H J problem
Heterogeneous fleet
x VRP
Feeder network design
problem
F 5 1 Comparison
x Uncertainty
Operational decisions
D I Tactical decisions

Iteration 7:
Node A B C D E F G H I J
Fi 0 2 1 3 5 6 ∞ ∞ ∞ ∞
Pred. - A A A C D - - - -
AMP
Shortest path problem - example 1
High level problems
Transportation
problem
Transshipment
B x G problem
Multiple commodities
x 2 Summary
Low level problems
x E 5 Shortest path problem
x Traveling salesman

x 4 5
problem
Vehicle routing
A C x H J problem
Heterogeneous fleet
x VRP
Feeder network design
problem
F 5 1 Comparison
x Uncertainty
Operational decisions
D I Tactical decisions

Iteration 8:
Node A B C D E F G H I J
Fi 0 2 1 3 5 6 7 ∞ ∞ ∞
Pred. - A A A C D E - - -
AMP
Shortest path problem - example 1
High level problems
Transportation
problem
Transshipment
B x G problem
Multiple commodities
x 2 Summary
Low level problems
x E 5 Shortest path problem
x Traveling salesman

x x 5
problem
Vehicle routing
A C x H J problem
Heterogeneous fleet
x VRP
Feeder network design
problem
F 5 1 Comparison
x Uncertainty
Operational decisions
D I Tactical decisions

Iteration 9:
Node A B C D E F G H I J
Fi 0 2 1 3 5 6 7 ∞ 9 ∞
Pred. - A A A C D E - E -
AMP
Shortest path problem - example 1
High level problems
Transportation
problem
Transshipment
B x G problem
Multiple commodities
x 2 Summary
Low level problems
x E x Shortest path problem
x Traveling salesman

x x 5
problem
Vehicle routing
A C x H J problem
Heterogeneous fleet
x VRP
Feeder network design
problem
F 5 1 Comparison
x Uncertainty
Operational decisions
D I Tactical decisions

Iteration 10:
Node A B C D E F G H I J
Fi 0 2 1 3 5 6 7 10 9 ∞
Pred. - A A A C D E E E -
AMP
Shortest path problem - example 1
High level problems
Transportation
problem
Transshipment
B x G problem
Multiple commodities
x 2 Summary
Low level problems
x E x Shortest path problem
x Traveling salesman

x x 5
problem
Vehicle routing
A C x H J problem
Heterogeneous fleet
x VRP
Feeder network design
problem
F x 1 Comparison
x Uncertainty
Operational decisions
D I Tactical decisions

Iteration 11:
Node A B C D E F G H I J
Fi 0 2 1 3 5 6 7 10 9 ∞
Pred. - A A A C D E E E -
AMP
Shortest path problem - example 1
High level problems
Transportation
problem
Transshipment
B x G problem
Multiple commodities
x x Summary
Low level problems
x E x Shortest path problem
x Traveling salesman

x x 5
problem
Vehicle routing
A C x H J problem
Heterogeneous fleet
x VRP
Feeder network design
problem
F x 1 Comparison
x Uncertainty
Operational decisions
D I Tactical decisions

Iteration 12:
Node A B C D E F G H I J
Fi 0 2 1 3 5 6 7 10 9 9
Pred. - A A A C D E E E G
AMP
Shortest path problem - example 1
High level problems
Transportation
problem
Transshipment
B x G problem
Multiple commodities
x x Summary
Low level problems
x E x Shortest path problem
x Traveling salesman

x x 5
problem
Vehicle routing
A C x H J problem
Heterogeneous fleet
x VRP
Feeder network design
problem
F x x Comparison
x Uncertainty
Operational decisions
D I Tactical decisions

Iteration 13:
Node A B C D E F G H I J
Fi 0 2 1 3 5 6 7 10 9 9
Pred. - A A A C D E E E G
AMP
Shortest path problem - example 1
High level problems
Transportation
problem
Transshipment
B x G problem
Multiple commodities
x x Summary
Low level problems
x E x Shortest path problem
x Traveling salesman

x x x
problem
Vehicle routing
A C x H J problem
Heterogeneous fleet
x VRP
Feeder network design
problem
F x x Comparison
x Uncertainty
Operational decisions
D I Tactical decisions

Iteration 14:
Node A B C D E F G H I J
Fi 0 2 1 3 5 6 7 10 9 9
Pred. - A A A C D E E E G
AMP
Shortest path problem - example 1
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
Summary
Low level problems
Shortest path problem
Node A B C D E F G H I J Traveling salesman
problem
Fi 0 2 1 3 5 6 7 10 9 9 Vehicle routing
Pred. - A A A C D E E E G problem
Heterogeneous fleet
VRP
Feeder network design
problem
I By backtracking we find the shortest path: A-C-E-G-J. Comparison
Uncertainty
Operational decisions
I The costs of the shortest path are FJ = 9. Tactical decisions
AMP
Shortest path problem
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
Summary

X Dijkstra’s algorithm can find the shortest or cheapest route to Low level problems
Shortest path problem
transport goods from one location to another. Traveling salesman
problem
Vehicle routing
problem
x Dijkstra’s algorithm does not work with negative costs Heterogeneous fleet
VRP
(like a standard road network). Feeder network design
problem
Comparison
Uncertainty
X The algorithm can be modified to work for specific cases with Operational decisions
Tactical decisions
negative costs
(e.g. Bellman-Ford’s algorithm).
AMP
Traveling salesman problem
High level problems
Transportation
problem
I The traveling salesman problem (TSP) is the problem of Transshipment
problem
finding the shortest tour along all locations while ending back Multiple commodities
at the start location. Summary
Low level problems
Shortest path problem
Traveling salesman
problem
Vehicle routing
problem
Heterogeneous fleet
VRP
Feeder network design
problem
Comparison
Uncertainty
Operational decisions
Tactical decisions
AMP
Traveling salesman problem
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
I N: the set of nodes. Summary
Low level problems
I cij the costs of traveling from i ∈ N to j ∈ N. Shortest path problem
Traveling salesman
problem
Vehicle routing
problem
Decisions: Heterogeneous fleet
VRP
I xij : the binary variable xij = 1 when traveling directly from Feeder network design
problem
i ∈ N to j ∈ N, and xij = 0 otherwise. Comparison
Uncertainty
Operational decisions
Tactical decisions
Goal:
I Minimize the total traveling costs of visiting every location
while ending back at the start location.
AMP
Traveling salesman problem
High level problems
Transportation
problem
Transshipment
Integer linear programming formulation: problem
Multiple commodities
Summary
Low level problems
Shortest path problem
Traveling salesman
X
min cij xij problem
Vehicle routing
i,j∈N problem
Heterogeneous fleet
VRP
X
xij = 1 ∀j ∈ N (In-degree constraints) Feeder network design
problem
i∈N Comparison
X Uncertainty
xij = 1 ∀i ∈ N (Out-degree constraints) Operational decisions
Tactical decisions
j∈N
X
xij ≥ 1 ∀S ⊂ N (Subtour elimination)
i∈S,j6∈S
xij ∈ {0, 1} ∀i, j ∈ N
AMP
Traveling salesman problem
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
I The subtour elimination constraints make sure that no Summary

subtours can occur. Low level problems


Shortest path problem
Traveling salesman
B D problem
Vehicle routing
1 1 problem
Heterogeneous fleet
VRP
Feeder network design
A 1 1 F problem
Comparison
Uncertainty

1 1 Operational decisions
Tactical decisions

C E

! There are exponentially many subtour elimination constraints.


AMP
Traveling salesman problem
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
Summary

x The TSP is difficult to solve. Low level problems


Shortest path problem
Traveling salesman
problem
x General purpose integer linear programming software will not Vehicle routing
problem
be able to solve large TSP’s with the presented formulation. Heterogeneous fleet
VRP
Feeder network design
I Binary (integer) decision variables. problem
I Exponentially many constraints. Comparison
Uncertainty
Operational decisions
Tactical decisions
X Specialized software exists to solve TSP’s with hundreds or
even thousands of nodes.
AMP
Vehicle routing problem
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
I Consider a single depot where goods are stored. Summary
Low level problems
Shortest path problem
Traveling salesman
I Consider customers with demand to be satisfied from the problem
Vehicle routing
depot. problem
Heterogeneous fleet
VRP
Feeder network design
I Consider an unlimited fleet of homogeneous vehicles, i.e., all problem
Comparison
vehicles have equal capacity, used to transport demand. Uncertainty
Operational decisions
Tactical decisions
I The vehicle routing problem (VRP) is to find routes for the
vehicles starting and ending at the depot satisfy all demand
at minimum costs.
AMP
Vehicle routing problem
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
Summary
Low level problems
Shortest path problem
Traveling salesman
problem
Vehicle routing
problem
Heterogeneous fleet
VRP
Feeder network design
problem
Comparison
Uncertainty
Operational decisions
Tactical decisions
AMP
Vehicle routing problem
High level problems

I N 0 : the customers Transportation


problem
Transshipment
I N : the customers and the depot. problem
Multiple commodities
Summary
I cij : the costs of traveling from i ∈ N to j ∈ N.
Low level problems
I di : the demand of customer i ∈ N 0 . Shortest path problem
Traveling salesman
problem
I Q: the vehicle capacity. Vehicle routing
problem
I b(S): the minimum number of vehicles required to satisfy Heterogeneous fleet
VRP
demand of all customers in S ⊆ N 0 . Feeder network design
problem
Comparison
Uncertainty
Decisions: Operational decisions
Tactical decisions
I xij : the binary variable xij = 1 when traveling directly from
i ∈ N to j ∈ N, and xij = 0 otherwise.

Goal:
I Minimize the total routing costs of visiting every customer
while satisfying the capacity constraints.
AMP
Vehicle routing problem
High level problems
Transportation
problem
Transshipment
Integer linear programming formulation: problem
Multiple commodities
Summary
Low level problems
Shortest path problem
Traveling salesman
X
min cij xij problem
Vehicle routing
i,j∈N problem
Heterogeneous fleet
VRP
X
xij = 1 ∀j ∈ N (In-degree constraints) Feeder network design
problem
i∈N Comparison
X Uncertainty
xij = 1 ∀i ∈ N (Out-degree constraints) Operational decisions
Tactical decisions
j∈N
X
xij ≥ b(S) ∀S ⊆ N 0 (Subtours + capacity)
i∈S,j6∈S
xij ∈ {0, 1} ∀i, j ∈ N
AMP
Vehicle routing problem
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
Summary
x Computing b(S) is very difficult Low level problems
(it is a bin-packing problem). Shortest path problem
Traveling salesman
problem
Vehicle routing
! Suppose b(S) = 1 for all S ⊆ N 0 problem
Heterogeneous fleet
VRP
I (this is the case when the vehicles have enough capacity to Feeder network design
problem
visit all customers) Comparison

then the VRP reverts to a TSP. Uncertainty


Operational decisions
Tactical decisions

X Hence, the VRP is a generalization of the TSP,


i.e., VRP is at least as difficult as TSP.
AMP
Heterogeneous fleet VRP
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
I The heterogeneous fleet VRP (HVRP) is the VRP where the Summary
Low level problems
vehicles have different capacities and operating costs. Shortest path problem
Traveling salesman
problem
Vehicle routing
Notation: problem
Heterogeneous fleet
VRP
I V : the set of vehicles Feeder network design
problem
I cijv : the costs of traveling from i ∈ N to j ∈ N using vehicle Comparison
Uncertainty
v ∈ V. Operational decisions

I Q v : the capacity of vehicle v ∈ V Tactical decisions

I xijv : the binary variable xijv = 1 when traveling directly from


i ∈ N to j ∈ N using vehicle v ∈ V , and xijv = 0 otherwise.
AMP
Heterogeneous fleet VRP
High level problems

Integer linear programming formulation: Transportation


problem
Transshipment
problem
Multiple commodities
Summary
X X Low level problems
min cijv xijv Shortest path problem
Traveling salesman
v ∈V i,j∈N problem
XX Vehicle routing

∀j ∈ N 0 (In-degree constraints)
problem
xijv = 1 Heterogeneous fleet
VRP
v ∈V i∈N Feeder network design
X X problem
xijv = xjiv ∀j ∈ N, v ∈ V (Flow constraints) Comparison
Uncertainty
i∈N i∈N Operational decisions
X 1 X v Tactical decisions
xijv ≥ xij ∀S ⊂ N 0 , v ∈ V (Subtour elimination)
|S|
i∈S,j6∈S i,j∈S
X X
di xijv ≤ Q v
∀v ∈ V (Vehicle capacity)
i∈N 0 j∈N
xijv ∈ {0, 1} ∀i, j ∈ N, v ∈ V
AMP
Heterogeneous fleet VRP
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
Summary
Low level problems
About the subtour elimination constraints: Shortest path problem
Traveling salesman
problem
X 1 X v Vehicle routing
xijv ≥ xij problem
|S| Heterogeneous fleet
VRP
i∈S,j6∈S i,j∈S Feeder network design
problem
Comparison

I The right hand side is greater than 0 (and at most 1) if Uncertainty


Operational decisions
vehicle v is used to visit customers in S, and 0 otherwise. Tactical decisions
AMP
Heterogeneous fleet VRP
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
Summary
Low level problems
Shortest path problem
Traveling salesman
! Suppose the capacity Q v of each vehicle is the same then the problem
Vehicle routing
HVRP reverts to a VRP. problem
Heterogeneous fleet
VRP
Feeder network design
problem
X Hence, the HVRP is a generalization of the VRP (and TSP), Comparison

i.e., the HVRP is at least as difficult as VRP (and TSP). Uncertainty


Operational decisions
Tactical decisions
AMP
Feeder network design problem
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
Summary
Low level problems
Shortest path problem
Traveling salesman
problem
Vehicle routing
problem
Heterogeneous fleet
VRP
Feeder network design
problem
Comparison
Uncertainty
Operational decisions
Tactical decisions

Source: O. Polat, H. Günther, O. Kulak (2014). The feeder network design problem: Application to container services in the
Black Sea region. Maritime Economics & Logistics.

I Hub ports: depot where ships start and end their route.
I Feeder ports: other ports with supply and/or demand.
AMP
Feeder network design problem
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
Summary
Low level problems
Shortest path problem
Traveling salesman
problem
Vehicle routing
problem
Heterogeneous fleet
VRP
Feeder network design
problem
Comparison
Uncertainty
Operational decisions
Tactical decisions

Source: O. Polat, H. Günther, O. Kulak (2014). The feeder network design problem: Application to container services in the
Black Sea region. Maritime Economics & Logistics.

I A heterogeneous fleet of ships is available to perform a


shuttle or cyclic liner service in which demand is picked up
and deliverd.
AMP
Feeder network design problem
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
Summary
Low level problems
Pick-up and delivery: Shortest path problem
Traveling salesman
problem
I Consider one type of commodity, e.g., oil. Vehicle routing
problem
I Every port either has demand or supply. Heterogeneous fleet
VRP
Feeder network design
problem
Comparison
The feeder network design problem is the problem of designing
Uncertainty
routes for the heterogeneous fleet of vehicles to pick-up and Operational decisions
Tactical decisions
deliver goods at minimum costs.
AMP
Feeder network design problem
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
Summary
Low level problems
I N 0 : the feeder ports. Shortest path problem
Traveling salesman
I N : the feeder ports and the hub. problem
Vehicle routing
problem
I cijv : the costs of traveling from i ∈ N to j ∈ N using vehicle Heterogeneous fleet
VRP
v ∈ V. Feeder network design
problem
I Q v : the capacity of vehicle v ∈ V Comparison
Uncertainty
I di : the demand of port i ∈ N. Operational decisions
Tactical decisions
I si : the supply of port i ∈ N.
AMP
Feeder network design problem
High level problems
Transportation
problem
Transshipment
problem
Decisions: Multiple commodities
Summary
I x v : the binary variable x v = 1 when traveling directly from Low level problems
ij ij
Shortest path problem
i ∈ N to j ∈ N using vehicle v ∈ V , and xijv = 0 otherwise. Traveling salesman
problem
I Pij : amount of goods picked up from ports up to i and Vehicle routing
problem
transported between i and j. Heterogeneous fleet
VRP
Feeder network design
I Pij : amount of goods to be delivered after visiting i and problem
Comparison
transported between i and j. Uncertainty
Operational decisions
Tactical decisions
Goal:
I Minimize the total traveling costs of picking up and delivering
all goods.
AMP
Feeder network design problem
Integer linear programming formulation: High level problems
X X
cijv xijv
Transportation
min problem
Transshipment
v ∈V i,j∈N problem
XX Multiple commodities

xijv = 1 ∀j ∈ N 0 (In-degree constraints) Summary


Low level problems
v ∈V i∈N Shortest path problem
X X Traveling salesman
xijv = xjiv ∀j ∈ N, v ∈ V (Flow constraints) problem
Vehicle routing
i∈N i∈N problem
Heterogeneous fleet
X 1 X v VRP
xijv ≥ xij ∀S ⊂ N 0 , v ∈ V (Subtour elimination) Feeder network design
|S| problem
i∈S,j6∈S i,j∈S Comparison
X X Uncertainty
Pij − Pji = si ∀i ∈ N (Supply constraints) Operational decisions
Tactical decisions
j∈N j∈N
X X
Dij − Dji = di ∀i ∈ N (Demand constraints)
i∈N i∈N
X
Dij + Pij ≤ Q v xijv ∀i, j ∈ N (Vehicle capacity)
v ∈V
xijv ∈ {0, 1} ∀i, j ∈ N, v ∈ V
Pij , Dij ≥ 0 ∀i, j ∈ N
AMP
Feeder network design problem
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
Summary
Low level problems
About the capacity constraints: Shortest path problem
Traveling salesman
problem
X Vehicle routing
Dij + Pij ≤ Q v xijv problem
Heterogeneous fleet
VRP
v ∈V
Feeder network design
problem
Comparison

I The right hand side is the capacity of the vehicle selected to Uncertainty
Operational decisions
travel between i ∈ V and j ∈ V . Tactical decisions
AMP
Feeder network design problem
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
Summary
Low level problems

! Suppose the supply si of all feeder ports i ∈ N 0 is 0 then the Shortest path problem
Traveling salesman
problem
problem reverts to a heterogeneous fleet VRP. Vehicle routing
problem
Heterogeneous fleet
VRP
X Hence, the FNDP is a generalization of the HVRP (and Feeder network design
problem
VRP and TSP), Comparison
Uncertainty
i.e., the FNDP is at least as difficult as HVRP (and VRP and Operational decisions
TSP). Tactical decisions
AMP
Comparison
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
Summary
SP TSP VRP HVRP FNDP
Low level problems
# Routes 1 path 1 tour Multiple Multiple Multiple Shortest path problem
Fleet No No Homogen. Heterogen. Heterogen. Traveling salesman
problem
Pick-up/ - - Or Or And Vehicle routing
Delivery problem
Heterogeneous fleet
Solvable Virtually ±1000 ±100 ±25 ±10 VRP
# nodes unbounded Feeder network design
problem
Comparison
Uncertainty
Operational decisions
I To solve transportation problems with more nodes heuristic Tactical decisions

algorithms are developed


(algorithms that do not guarantee optimality).
AMP

High level problems


Transportation
problem
Transshipment
problem
Multiple commodities
Summary
Low level problems
Shortest path problem
Traveling salesman
problem

Uncertainty
Vehicle routing
problem
Heterogeneous fleet
VRP
Feeder network design
problem
Comparison
Uncertainty
Operational decisions
Tactical decisions
AMP
Uncertainty
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
Overview: Summary
Low level problems
Shortest path problem
Traveling salesman
I We will discuss a model with uncertain port handling times problem
Vehicle routing
and uncertain travel times. problem
Heterogeneous fleet
VRP
Feeder network design
problem
I When faced with delays there is the opportunity to increase Comparison

sailing time during operations. Uncertainty


Operational decisions
Tactical decisions

I The operational decisions will be integrated with tactical


scheduling decisions.
AMP
Uncertainty
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
Summary
Low level problems
Shortest path problem
Traveling salesman
problem
Vehicle routing
problem
Heterogeneous fleet
VRP
Feeder network design
problem
Comparison
Uncertainty
Operational decisions
Tactical decisions

Source: www2.nykline.com

I Consider a fixed cyclic route schedule.


I Given uncertainty in sailing time, what departure times do we
schedule?
AMP
Operational decisions
Given: High level problems
I A cyclic route schedule. Transportation
problem
I A fixed departure time from each port. Transshipment
problem
Multiple commodities
I A current position on the route. Summary
I The current delay (possibly 0). Low level problems
Shortest path problem
I High costs if the arrival time at the next port is not early Traveling salesman
problem
enough to perform all handling activities before departure Vehicle routing
problem
(cut-and-go costs). Heterogeneous fleet
VRP
I A set of (corrective) actions (e.g., decrease, maintain or Feeder network design
problem
increase sailing speed) at specified costs. Comparison
Uncertainty
Operational decisions
Which (corrective) action should we take to minimize Tactical decisions

expected costs of a complete tour?

I Note that this decision should be contingent on the


continuation of the tour.
I Perhaps there is enough slack on the next legs to compensate
the current delay.
I Perhaps there is no delay now, but we expect some delay on
the next legs.
AMP
Operational decisions
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
State: Summary
Any possible situation is related to an individual state. Low level problems
Shortest path problem
I Location on the route. Traveling salesman
problem
I Current delay. Vehicle routing
problem
Heterogeneous fleet
VRP
Feeder network design
Example: problem
Comparison
I The ship has currently departed from Rotterdam with 4 hours Uncertainty
Operational decisions
of delay. Tactical decisions

State space:
The set of all states is called the state space.
AMP
Operational decisions
High level problems
Transportation
problem
Transshipment
Policy: problem
Multiple commodities
A prescription of which action to take in every state is called a Summary

policy. Low level problems


Shortest path problem
Traveling salesman
problem
Vehicle routing
I It might be optimal to take an action only with a certain problem
Heterogeneous fleet
probability. VRP
Feeder network design
problem
Comparison
I The policy affects the probability that we are in a certain Uncertainty

state at a random moment. Operational decisions


Tactical decisions

I Because the route is cyclic, we are interested in steady state


probabilities, the probabilities (fraction of time) we are in a
certain state in the long term.
AMP
Operational decisions
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
I S: the set of states. Summary

I A: the set of actions. Low level problems


Shortest path problem
I cia : the costs of taking action a ∈ A when in state i ∈ S. Traveling salesman
problem
Vehicle routing
I pija : the probability of going from state i ∈ S to state j ∈ S problem
Heterogeneous fleet
when taking action a ∈ A. VRP
Feeder network design
Example: problem
Comparison
I i =(Departed from Rotterdam, 4 hours of delay) Uncertainty
I j =(Arrival in Antwerp, 0 hours of delay) Operational decisions
Tactical decisions
I a =(increase sailing speed to 30 knots)
I pija = 0.6
I Note that pija is dependent on the probabilities of incurring
additional delay between Rotterdam and Antwerp.
AMP
Operational decisions
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
Summary

Cut-and-go Low level problems


Shortest path problem
I It is common to include the cut-and-go action for each port Traveling salesman
problem
activity. Vehicle routing
problem
Heterogeneous fleet
I This action represents departing directly without having VRP
Feeder network design
unloaded or loaded all goods. problem
Comparison
I The costs of this action are very high. Uncertainty
Operational decisions
I This action is used to put a bound on the total delay. Tactical decisions

I We aim at never using this action.


AMP
Operational decisions
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
Summary
Low level problems
Decisions: Shortest path problem
Traveling salesman
I πia : the steady state probability of being in state i ∈ S and problem
Vehicle routing
problem
taking action a ∈ A. Heterogeneous fleet
VRP
Feeder network design
problem
Goal: Comparison
Uncertainty
I Minimize the expected long term costs of sailing the cyclic Operational decisions
Tactical decisions
route.
AMP
Operational decisions
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
Linear programming formulation: Summary
Low level problems
Shortest path problem
Traveling salesman
problem
XX Vehicle routing
min cia πia problem
Heterogeneous fleet
VRP
i∈S a∈A
XX Feeder network design
problem
πia = 1 (Probabilities) Comparison

i∈S a∈A Uncertainty


XX X Operational decisions
pija πia = πja ∀j ∈ S (Transition) Tactical decisions

i∈S a∈A a∈A


πia ≥ 0 ∀i ∈ S, a ∈ A
AMP
Operational decisions
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
Summary
X The operational decision problem can be solved efficiently, for Low level problems
Shortest path problem
instance using linear programming software. Traveling salesman
problem
Vehicle routing
problem
x The operational decision problem suffers from the Heterogeneous fleet
VRP
curse of dimensionality, i.e., the size of the state space Feeder network design
problem
quickly explodes when more detailed information is used to Comparison
identify states: Uncertainty
Operational decisions
I Delays in days as opposed to minutes. Tactical decisions
I Considering only port locations to make decisions as opposed
to also making decision at way-points on route.
AMP
Tactical decisions
High level problems
Transportation
problem
Transshipment
problem
I Suppose the minimal sailing time of a cyclic route is 21 days. Multiple commodities
Summary
Low level problems
Shortest path problem
I Suppose the cycle repeats itself every 30 days. Traveling salesman
problem
Vehicle routing
problem
I There is a buffer of 9 days. Heterogeneous fleet
VRP
Feeder network design
problem
Comparison
I Suppose we use two out of the 9 days and assign them to one Uncertainty
port, the chance of having a delay at this port decreases and Operational decisions
Tactical decisions
so do the expected cut-and-go costs.

How should the total buffer be divided over the ports to


minimize expected delay costs?
AMP
Tactical decisions
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
State space: Summary
Low level problems
I Recall that each state was a combination (port,delay). Shortest path problem
Traveling salesman
I We now modify the state space to include the allocated problem
Vehicle routing
buffer. problem
Heterogeneous fleet
VRP
I Let P be the set of ports. Feeder network design
problem
I Let D be the set of delays. Comparison
Uncertainty
I Let B be the set of buffers. Operational decisions
Tactical decisions
I For example B = {12 hours, 24 hours, 36 hours, 48 hours}.
I Hence, we can write state i ∈ S as i = (p, d, b) for p ∈ P,
d ∈ D and b ∈ B.
AMP
Tactical decisions
High level problems
Transportation
problem
Transshipment
problem
I Let S, P, D, B, A, cia , pija and πia be defined as before. Multiple commodities
Summary

I M: the total buffer time to be divided Low level problems


Shortest path problem
I tb : buffer time of buffer b ∈ B. Traveling salesman
problem
Vehicle routing
problem
Heterogeneous fleet
Decisions: VRP
Feeder network design
problem
I ypb : the binary variable ypb = 1 when buffer b ∈ B is Comparison

allocated to port p ∈ P, and yip = 0 otherwise. Uncertainty


Operational decisions
Tactical decisions

Goal:
I Allocate the total buffer such that the expected long term
costs of sailing the cyclic route.
AMP
Tactical decisions
Mixed integer programming formulation: High level problems
Transportation
problem
Transshipment
problem
XX Multiple commodities
Summary
min cia πia
Low level problems
i∈S a∈A Shortest path problem
XX Traveling salesman
πia = 1 (Probabilities) problem
Vehicle routing
problem
i∈S a∈A
XX X Heterogeneous fleet
VRP
pija πia = πja ∀j ∈ S (Transition) Feeder network design
problem
i∈S a∈A a∈A Comparison
XX Uncertainty
π(p,d,b),a ≤ ypb ∀p ∈ P, b ∈ B (States in use) Operational decisions
Tactical decisions
d∈D a∈A
XX
tb ypb ≤ M (Maximum buffer time)
p∈P b∈B
X
ypb = 1 (Buffer allocation)
b∈B
πia ≥ 0 ∀i ∈ S, a ∈ A
ypb ∈ {0, 1} ∀p ∈ P, b ∈ B
AMP
Tactical decisions
High level problems
Transportation
problem
Transshipment
problem
Multiple commodities
I The costs resulting from tactical decisions are determined by Summary

the operational decision. Low level problems


Shortest path problem
Traveling salesman
problem
I When dealing with uncertainty, tactical and operational Vehicle routing
problem
decisions are often integrated. Heterogeneous fleet
VRP
I Tactical decisions are made before uncertain parameters are Feeder network design
problem
observed. Comparison

I Operational decisions are made to respond after uncertain Uncertainty


Operational decisions
parameters are observed. Tactical decisions

I Integrated models quickly become intractable when using


more detailed information.

You might also like