MARIT. POL. MGMT.

, OCTOBER–DECEMBER 2004
VOL. 31, NO. 4, 259–268

Designing optimal routes in a liner shipping problem

KJETIL FAGERHOLT*
Norwegian University of Science and Technology, Department of
Marine Technology, Trondheim, Norway

A real liner shipping problem of deciding optimal weekly routes for a given fleet
of ships is considered and a solution method for solving the problem is proposed.
First, all feasible routes for each ship are generated together with the cost and the
duration for each route. The routes are given as input to an integer programming
(IP) problem. By solving the IP problem, routes for each ship are selected such
that total transportation costs are minimized and the demand at each port is
satisfied. The total duration for the routes that are selected for a given ship
must not exceed one week.
The real liner shipping problem is solved together with four randomly gener-
ated test problems. The computational results show that proposed solution
method is suitable for designing optimal routes in several liner shipping problems.

1. Introduction
International trade has experienced a continuous growth during this century due to
an increase in world population and an increase in standard of living. Shipping is
the major transportation mode of international trade. A ship may cost thousands of
dollars daily; thus a modest improvement of the routing and scheduling can result in
large monetary savings. Despite this, relatively little has been done in quantitative
aspects of managing seaborne shipping systems. Explanations for the low attention
on ship routing and scheduling are discussed in [1–3].
Lawrence [4] distinguishes between three main modes of operation in the shipping
industry: liner, tramp and industrial operation. In industrial shipping, the cargo
owner or shipper also controls the ships. Industrial operators try to ship all their
cargoes at minimal cost. Tramp ships follow the available cargoes, like a taxi. Liners
operate according to a published itinerary and schedule similar to a bus line. Most of
the contributions on ship routing and scheduling in the literature pertain to indus-
trial operation, see for instance [5–11]. References on tramp shipping are fewer
because many shipping companies traditionally often have seen the tramp market
as a secondary one, see for instance [12–13].
In liner shipping there may be a variety of problems involving decisions at
different planning levels [14]:
(1) Route and schedule design, i.e. design an optimal set of liner routes to be
serviced by a fleet of vessels. This is most often referred to as a strategic
planning problem [15].
(2) Fleet size and mix, i.e. design an optimal fleet for a given liner service. This is
also a strategic planning problem.

*e-mail: [email protected]

Maritime Policy & Management ISSN 0308–8839 print/ISSN 1464–5254 online # 2004 Taylor & Francis Ltd
http://www.tandf.co.uk/journals
DOI: 10.1080/0308883042000259819
260 K. Fagerholt

(3) Fleet deployment, i.e. assignment of vessels to given routes. This is most
often defined as a tactical planning problem.
(4) Cargo booking, i.e. choosing which cargoes to accept or reject for a given
voyage, which is an operational planning problem.
The background of this paper is a real problem that arose when several major
shippers with production units along the Norwegian coast started to discuss entering
into a joint containerized transportation system for the markets in Europe. There-
fore, several shipping companies started developing an ocean shipping system in
order to meet the requirements of the shippers. The study in this paper was used
as a decision support in the process of designing parts of this transportation system.
To achieve an efficient shipping system a hub or a central port was suggested. This
partitions the system into two parts. The first part of the shipping system, which is
called the feeder system, concerns the transportation from the ports at the produc-
tion units to the hub. The second part is the transportation from the hub to the
markets. Both these two parts will have its own, more or less independent shipping
system, separated by the hub. The focus of this paper is to develop optimal weekly
routes for the feeder system with a given a fleet of ships and transportation demand
at each production unit. Since the transportation demand at the production units do
not vary much over time, a set of ship routes is decided for a fixed period (the length
of this period depends on factors such as seasonal fluctuations and market require-
ments for the production units involved). The problem of deciding optimal weekly
sailing routes for the ships in the fleet can thus be considered as a route design
problem in liner shipping, which according to [16] can be regarded as a strategic
planning problem.
There a few references in the literature dealing with similar problems, see for
instance [17–20]. In contrast to the problem considered here, where the fleet is
fixed and given, these studies’ main issue is deciding an optimal fleet size.
For the interested reader, we can refer to a recent review article on ship routing
and scheduling [21], which systemizes and presents almost 60 references on ship
routing and scheduling mostly published in the period 1993–2003.
In the next section a more detailed description of the problem considered in this
paper is given. It is shown that the problem is similar to a multi-trip ship routing
problem, which also has several other important practical applications. Section 3
describes the proposed method to find the optimal routes for the ships in the fleet.
The method is in addition to the real liner shipping problem also tested on some
randomly generated test problems. The generation of the test problems and the
computational results are presented in section 5, while concluding remarks and a
discussion of future research follow in the last section.

2. Problem description
The basic vehicle routing problem (VRP) is the problem of deciding optimal routes
for the vehicles in a fleet, where the vehicles make exactly one route or trip. Each
route consists of departing from a depot or a hub, visiting one or more customers
and returning to the depot. Each customer with a given demand is visited once, and
only once, by one of the vehicles. Each vehicle has a given capacity. Time constraints
on the maximum duration of each route are also often imposed in the VRP. Bodin et
al. [22] give a mathematical formulation of such a time-constrained VRP.
 Designing optimal routes 261

For the liner shipping problem considered here, the task is to develop optimal
weekly routes for a given heterogeneous fleet of ships. Each ship has a given cost
structure, service speed and capacity regarding the number of containers that can be
loaded onboard the ship. The customers are the production ports with given
estimates on the number of containers to be delivered weekly. The ships also bring
empty containers that are discharged at the ports of the production units, while full
containers are loaded into the ships. Since the number of containers that are loaded
onboard the ship usually is the same as the number of empty containers discharged
at each port, it is assumed that only the transportation of full containers influences
on the capacity constraints. Each production port must be serviced at least once a
week.
A liner ship route always starts and ends at the depot or a hub, visiting at least one
production port in between. The challenge is to find optimal weekly routes for the
ships in the fleet, i.e. the routes that minimize the total transportation costs given
the capacity and service constraints. The transportation costs considered consist of
the operational costs of the fleet, such as fuel and port costs. The fixed fleet costs are
left out, as they do not depend on the route selection.
The problem is illustrated for the real liner shipping case in figure 1. The ports
at the production units along the Norwegian coast are indicated with their weekly
transportation demands. The depot or hub is indicated by a square. Due to reasons
related to confidentiality, the transportation demands indicated in the figure are not
the same as for the real problem.
The problem of deciding optimal weekly liner routes looks similar to the time-
constrained VRP. Here, the maximum allowed route time is one week, including
sailing time, service time in ports and necessary slack that must be imposed due
to bad weather or any other unforeseen events. However, for the liner shipping

+90
+50
+10
+40

+10 +260
+180
+10
+50
+90 +210

+90
+150
+120

Figure 1. Transportation demand for the feeder system.

262 K. Fagerholt

problem, several routes or multiple trips are allowed for each ship, as long as the
total duration of the routes that are selected for any given ship does not exceed
one week.
Thus, the problem of deciding weekly liner routes can be considered as a multi-
trip VRP. Despite the fact that multiple trips are important in many practical routing
applications where there are time constraints, the multi-trip VRP has received little
attention in the operational research literature. In [23], the multi-trip VRP is solved
by tabu search heuristics.

3. Solution algorithm
In this paper, an optimization algorithm consisting of two phases is proposed for
solving the multi-trip VRP. In phase 1, all routes that are feasible regarding capacity
and time constraints are generated for each ship. These routes are given as input to
an integer programming model in phase 2. By solving the integer programming
model, the routes for each ship that minimize the total transportation costs and
ensure that all ports are serviced are selected. A similar solution approach is also
applied to other ship routing and scheduling problems given in [24–30]. However, in
these problems, the model in phase 2 becomes different than in our problem, because
it does not consider the multiple trips aspect. Therefore, in most of these studies a set
partitioning model is obtained that is easier to solve, compared with our integer
programming model.
Section 3.1 describes the route generation of phase 1, while formulation of the
integer programming model of phase 2 is given in section 3.2.

3.1. Phase 1: route generation

In the route generation, or phase 1, of the solution algorithm all routes that are
feasible with respect to time and capacity constraints are generated. This may be
illustrated by a simplified example:
Example 1. Suppose we have a problem with a ship with a capacity of 240 units and
the maximum allowed duration is seven days. A network for the problem is shown in
figure 2. In figure 2, the figures by the nodes indicate the number of units to be

+80
1 1 +140
2
3
3
2
3

2 3 +160
0
Figure 2. Example problem network.
 Designing optimal routes 263

Table 1. Feasible routes for example 1.

Route number Route Elapsed time

1 0 1 0 6
2 0 2 0 6
3 0 3 0 4
4 0 1 2 0 7

loaded, while the figures by the arcs show the sailing time in days. The route
0 1 3 0 is infeasible since the elapsed time for the route is 8 (days). The route
0 2 3 0 is infeasible due to the capacity constraints. Table 1 shows all feasible
routes for the example problem in figure 2.
Figure 3 illustrates the conceptual flowchart of the solution algorithm for the
route generation. For a given ship, we initialize the route generation by calculating
the routes consisting of only one port (in addition to the depot) that are feasible
regarding capacity and duration. The first three routes in table 1 are such routes.
These routes are kept in a route list. The total elapsed time and the cost are
calculated and stored for each route.
The first route in the route list is then selected to generate new routes by
adding one more port visit. The port to extend the route is selected in a way
that ensures that the same route will not be generated twice and the capacity of
the ship is not exceeded. To find the optimal visiting sequence for the new extended
route, we have to solve a Travelling Salesman Problem (TSP). Suppose we have
the route 0 1 2 4 0 that we want to extend by port i. We solve a TSP with
the nodes 0, 1, 2, 4 and i, where the depot node 0 is the start and ending node.
We also have to check that the new, extended route is feasible regarding the time
constraints.
For the real liner shipping problem, the TSP is rather simple, due to the geo-
graphical location of the ports along the Norwegian coast, see figure 1.
The TSP is solved by a dynamic programming algorithm with minimization of
route time as objective. This also will correspond to the least cost route, since the
sailing costs are directly related to the distance travelled and hence the sailing time
(for a given set of port calls, the port costs will be constant regardless of visiting
sequence). For the real liner shipping problem considered in this paper, the capacity
and the time constraints are relatively tight, which result in a small number of ports
to be sequenced in each TSP. Therefore, all feasible routes can easily be generated for
the given ship. This procedure is repeated for all ships in the fleet.

3.2. Phase 2: integer programming formulation

Let K be the set of ships in the fleet, indexed by k. Let Rk be the set of candidate
routes for ship k, indexed by r. Akri is a constant which is equal to one if route r for
ship k services node i, and zero otherwise. Crk is the operational cost (including port
charges) for ship k to sail route r, while Trk is the elapsed time for ship k to sail
route r. Akri , Crk and Trk are calculated in the route generation procedure in phase 1.
TMAX is the maximum time allowed to use for a ship, which is one week minus
necessary slack for the real liner shipping problem. Let xkr be a binary variable that is
equal to one if ship k uses route r and zero otherwise.
264 K. Fagerholt

START STOP

k=1 NO

YES Are there any ships

Select ship k
left in the fleet?

Initialize route list to routes

with only one port visit (in
addition to the depot)

Are there any routes NO

k=k+1
left in the route list?

YES

Delete route u from the route Select route u (the first route in
list the route list) to generate new
routes

Let n be the maximum port

number serviced by route u and
YES let i = n + 1

Is i > number of ports NO

Select port i to extend route u
in the problem?

NO Is the new route

i=i+1 feasible regarding capacity
constraints?

YES

Solve the TSP to find the

optimal visiting port sequence
for the new route

NO Is the new route

feasible regarding time
constraints?

YES

Add the new route to the route

list and keep it as a candidate
route for the master problem

Figure 3. Conceptual flowchart of the route generation algorithm.

The multi-trip ship routing problem can now be formulated as:

XX k k
min Cr xr ð1Þ
k2K r2Rk

XX
Akri xkr
1 8i 2 N ð2Þ
k2K r2Rk
 Designing optimal routes 265
X
Trk xkr  TMAX 8k 2 K ð3Þ
r2Rk
X
xkr
1 8k 2 K ð4Þ
r2Rk

xkr 2 f0, 1g 8k 2 K, 8r 2 Rk ð5Þ

The objective function (1) minimizes the total transportation costs, i.e. the operating
costs (including port charges) for the whole fleet. Constraints (2) ensure that all
production ports are serviced at least once per week and (3) ensure that the total
duration for the selected routes for a given ship does not exceed maximum allowed.
Constraints (4) ensure that each ship sails at least one route. These constraints could
be omitted, allowing some ships being unused. (5) imposes binary restrictions on the
variables.

4. Computational results
The solution procedure is tested on data from the real liner shipping problem. To test
the algorithm on other similar problems of different size some test problems are also
generated. The generation of these test problems is described in the next section,
while results are presented in section 4.2.

4.1. Generation of test problems

Four different test problems are generated, with 10, 20, 30 and 40 ports to be serviced
respectively. The basis on the generation of the test problems is the real shipping
problem described in the first two sections. For simplicity, and in contrast to the
real problem, it is assumed that a homogeneous fleet of ships is used in the random
test problems, i.e. all ships have the same speed and capacity. It should, however, be
emphasized that the algorithm works just as well for heterogeneous fleets. The ship
that is used in the test problems is selected from the real problem. This ship is
also used in calculating the operational costs of the candidate routes generated in
phase 1.
The locations of the ports or customers are randomly (uniformly) generated from
a 2-dimensional Euclidean space, such that the sailing distances between the ports
and the depot are Euclidean and in the same order of magnitude as for the real
problem. Figure 4 illustrates the generation of the port locations, where the sailing
times are indicated along the two axes. The depot is located in the origin. The port
demands are randomly generated from [0.1–0.9] of the ship capacity, which give
relatively tight capacity constraints.

4.2. Computational performance

The route generation algorithm is written and compiled in Borland Pascal 7.0, while
the integer programming model was implemented and solved by GAMS/CPLEX
version 5.0. The real problem consists of 15 ports that must be serviced in addition
to the depot. For this problem a heterogeneous fleet of five ships is available. It is
assumed that a homogeneous fleet of ships is used for the random test problems and
the number of ships in the fleet is, for simplicity, assumed to be half of the number of
ports. All problems are solved with a minimum weekly slack of six hours, i.e. the
total duration for the routes selected by a ship can not exceed 162 hours (one week
266 K. Fagerholt

40

Sailing time [hours]

x
40
Sailing time [hours]
Figure 4. Generation of port locations.

Table 2. Computational results.

Test problem
Real 1 2 3 4

No. nodes/ports 15 10 20 30 40
No. ships 5 5 10 15 20
No. candidate routes 1689 240 1760 18 720 41 420
Computation time, phase 1 [s] 5.5 0.1 0.2 3.7 9.0
Computation time, phase 2 [s] 17.0 0.4 1.6 20.9 177.5
No. sailed routes in optimal solution 8 6 13 16 22

minus six hours). In addition, some slack is included in the selection of the service
speed of the ships in order to maintain route regularity and allow for bad weather.
The service time for each port visit consists of two parts: a fixed time to get the ship
prepared for loading/discharging and a time depending on the number of units to be
handled. Some slack is also included in the parameters defining the service time in
ports.
Table 2 gives the computational results for the different problems. The route
generation of phase 1 was performed within a few seconds for all instances. All
problems were also optimally solved within seconds, except for test problem 4,
where we experienced memory problems by trying to solve the problem to optimal-
ity. Here, the solution was only proved to be within 1% of the optimal solution.
For the real problem, the number of sailed routes in the optimal solution was
eight (the number of ships was five). This indicates the occurrence of multiple trips.
In fact, one ship sailed three routes, while another sailed two. Fewer multiple trips
occurred in the optimal solutions for the randomly generated test problems. For
problems 1, 2, 3 and 4, there were one, three, one and two ships that sailed two
routes, respectively. This indicates that the timing constraints were a little tighter for
these problem instances than for the real problem.

5. Concluding remarks and future research

A real liner shipping problem of deciding optimal weekly routes for a given fleet
of ships is considered. It is shown that the problem can be considered as a
 Designing optimal routes 267

multi-trip VRP. The proposed solution method consists of two phases. In phase 1, all
feasible routes are generated for each ship in the fleet together with the cost and the
duration of the routes. These routes are given as input to an integer programming
model in phase 2. By solving the integer programming model, the routes for each
ship that minimize total operational costs for the whole fleet are determined. Each
ship may sail multiple routes as long as the total time consumption does not exceed
one week.
The computational results show that the proposed solution method works.
The real liner shipping problem is solved to optimality within seconds. The four
randomly generated test problems are also solved to optimality within reasonable
computer time, except the solution of test problem 4, which due to computer
memory problems was only proved to be within 1% of the optimal solution.
The proposed solution method may also be extended for solving similar problems
for other shipping companies operating a liner fleet. Based on estimates for the
expected demand for transport services between the different geographical regions
or ports, the solution method can be used to determine liner routes for their fleet
such that the expected profits are maximized.
It should, however, be emphasized that the problems considered here are
relatively small and well constrained because of high demands at each port compared
with the ship capacities and also because of high sailing times compared with the
maximum time available. Therefore, the number of feasible sailing routes is limited
compared to what can be expected of many other routing problems of the same size.
The proposed solution method is not expected to work that well on larger and not
so well constrained problems. However, many real-life ship routing and scheduling
problems do not involve more customers or ports and ships than examined here.
These problems are also often well constrained.
If the problem size and structure should be so that it is not possible to generate all
feasible routes and solve the problem to optimality within reasonable time, heuristic
methods can be used in the route generation of phase 1 to reduce the number of
routes. Examples of such approaches can be found in [31–33], where heuristic rules
regarding the utilisation of the ships are used to eliminate the number of sailing
routes.
Another possible future extension of this work could be to allow for split delivery,
where the demand at any port can be split between any two routes. Dror and
Trudeau [34,35] present a route improvement heuristic for solving split delivery
routing problems. They report savings in the travelled distance of about 7% when
the demands are in the same size as for the problems considered here. However, we
cannot expect these improvements in costs by introducing split delivery for this ship
routing problem, since split delivery involves more port arrivals, which again result
in additional time in the ports and increased port costs.

References
1. Ronen, D., 1983, Cargo ships routing and scheduling: survey of models and problems.
European Journal of Operational Research, 12, 119–126.
2. Ronen, D., 1993, Ship scheduling: the last decade. European Journal of Operational
Research, 71, 325–333.
3. Christiansen, M., Fagerholt, K. and Ronen, D., 2004, Ship routing and scheduling:
status and perspectives. Transportation Science, 38(1), 1–18.
4. Lawrence, S. A., 1972, International Sea Transport: The Years Ahead (Lexington, MA:
Lexington Books).
268 Designing optimal routes

5. Bausch, D. O., Brown, G. G. and Ronen, D., 1998, Scheduling short-term marine
transport of bulk products. Maritime Policy and Management, 25(4), 335–348.
6. Brown, G. C., Graves, G. W. and Ronen, D., 1987, Scheduling ocean transportation
of crude oil. Management Science, 33(3), 335–346.
7. Christiansen, M. and Fagerholt, K., 2002, Robust ship scheduling with multiple
time windows. Naval Research Logistics, 49(6), 611–625.
8. Fagerholt, K., 2001, Ship scheduling with soft time windows. European Journal of
Operational Research, 131, 559–571.
9. Fagerholt, K. and Christiansen, M., 2000, A combined ship scheduling and alloca-
tion problem. Journal of the Operational Research Society, 51(7), 834–842.
10. Fisher, M. L. and Rosenwein, M. B., 1989, An interactive optimization system for
bulk-cargo ship scheduling. Naval Research Logistics, 36, 27–42.
11. Perakis, A. N. and Jaramillo, D. I., 1991, An operational tanker scheduling optimiza-
tion system: background, current practice and model formulation. Maritime Policy and
Management, 19(3), 177–187.
12. Appelgren, L., 1969, A column generation for a ship scheduling problem.
Transportation Science, 3, 53–68.
13. Kim, S.-H. and Lee, K.-K., 1997, An optimization-based decision support system for
ship scheduling. Computers and Industrial Engineering, 33(3–4), 689–692.
14. Ibid. 3.
15. Ibid. 3.
16. Ibid. 3.
17. Bendall, H. B. and Stent, A. F., 2001, A scheduling model for a high speed con-
tainer service: a hub and spoke short-sea application. International Journal of Maritime
Economics, 3(3), 262–277.
18. Fagerholt, K., 1999, Optimal fleet design in a ship routing problem. International
Transactions in Operational Research, 6(5), 453–464.
19. Fagerholt, K. and Lindstad, H., 2000, Optimal policies for maintaining a supply
service in the Norwegian Sea. OMEGA—The International Journal of Management
Science, 28(3), 269–275.
20. Powell, B. J. and Perakis, A. N., 1997, Fleet deployment optimization for liner
shipping: an integer programming model. Maritime Policy and Management, 24(2),
183–192.
21. Ibid. 3.
22. Bodin, L., Golden, B., Assad, A. and Ball, M., 1983, Routing and scheduling
of vehicles and crews: the state of the art. Computers and Operations Research, 10(2),
63–211.
23. Brando, J. and Mercer, A., 1997, A tabu search algorithm for the multi-trip vehicle
routing and scheduling problem. European Journal of Operational Research, 100,
180–191.
24. Ibid. 5.
25. Ibid. 6.
26. Ibid. 7.
27. Ibid. 8.
28. Ibid. 9.
29. Ibid. 10.
30. Ibid. 11.
31. Ibid. 9.
32. Ibid. 10.
33. Ibid. 11.
34. Dror, M. and Trudeau, P., 1989, Savings by split delivery routing. Transportation
Science, 23(2), 141–145.
35. Dror, M. and Trudeau, P., 1990, Split delivery routing. Naval Research Logistics, 37,
383–402.