Ann Oper Res

DOI 10.1007/s10479-014-1527-4

Flow-based integer linear programs to solve the weekly

log-truck scheduling problem

Nizar El Hachemi · Issmail El Hallaoui ·

Michel Gendreau · Louis-Martin Rousseau

© Springer Science+Business Media New York 2014

Abstract In this paper we present the solution to a weekly log-truck scheduling problem
(LTSP) integrating the routing and scheduling of trucks where all goods are transported
in full truckloads. We must take into account pick-up and delivery requirements, multiple
products, inventory levels, and lunch breaks. The objective is to minimize the overall trans-
portation cost including wait times and the empty and loaded distance traveled. Our solution
is based on a two-phase approach. The first phase involves an integer linear program that
determines the destinations of full truckloads. The second phase uses an implicit integer lin-
ear program based on an arc formulation to ensure that the trucks are routed and scheduled
at a minimum cost. Experiments have been conducted using Cplex 12.4.0, and almost all
instances were solved within six hours with a reasonable gap.

Keywords Forestry · Transportation · Routing · Scheduling · Mixed integer programming

Notations
Parameters
B Set of all bases
cl Cost of waiting time of a log-loader per unit of time

B
N. El Hachemi ( )
Ecole Mohammadia d’Ingénieurs (Rabat), Université Mohammed V Agdal, Avenue Ibnsina, B.P. 765,
Agdal Rabat, Maroc
e-mail: [email protected]

I. El Hallaoui · M. Gendreau · L.-M. Rousseau

Département de Mathématiques et Génie Industriel, Ecole Polytechnique de Montréal, C.P. 6079,
Succursale Centre-Ville, Montreal, Québec H3C 3A7, Canada
I. El Hallaoui
e-mail: [email protected]
M. Gendreau
e-mail: [email protected]
L.-M. Rousseau
e-mail: [email protected]
 Ann Oper Res

ca Cost associated with arc a (empty driven arcs and truck waiting times)
tl Loading time of one shipment
tu Unloading time of one shipment
H Optimization horizon
Tas Start time associated with arc a
Tae End time associated with arc a
Kf m Number of full truckloads to be delivered from forest area f to wood mill m
Lbf m Set of loaded-trip arcs linking forest area f and wood mill m and associated with
base
 b
Lb f ∈F m∈M Lbf m , set of loaded-trip arcs associated with base b
A+ (sb ) Set of exiting arcs from the source node sb associated with base b
A− (tb ) Set of entering arcs into the sink node tb associated with base b
A+ (i) Set of exiting arcs from node i
A− (i) Set of entering arcs into node i
ATbf Set of loading arcs (ab ) loading at forest area f and associated with base b, such
that their start times Tasb ∈ [T , T + t l [, T ≤ H
ATbm Set of unloading arcs (ab ) unloading at wood mill m and associated with base b,
such that their start times Tasb ∈ [T , T + t u [, T ≤ H
Cbf Set
 of loading arcs associated with forest area f and base b
Cb f ∈F Cbf , set of loading arcs associated with base b
Db Set of unloading arcs associated with base b
Eb Set of empty driven arcs associated with base b
Wb Set of truck waiting arcs associated with base b
Nb Set of nodes associated with base b
Vb Number of trucks for base b
A Set of all arcs.
Variables
xab Number of trucks that follow arc ab ∈ A
tfmin Start time of log-loader associated with forest area f
tfmax End time of log-loader associated with forest area f

1 Introduction

The forestry industry manufactures goods from timber grown in forests. It provides a large
range of products including paper, wrappings, building materials, and furniture. As wood
harvesting has advanced over the years, public concerns and ancillary uses of the forest
(recreation, and the preservation of wildlife species and vegetation) have developed in par-
allel with the development of the forestry industry. The forestry industry occupies an im-
portant place in the economy of countries such as Chile, Canada, Sweden, Finland, New
Zealand, and Austria. In Canada, the industry employed about 600,000 people in 2011, and
233,900 were directly employed. In the same year, it contributed $ 23.7 billion to Canada’s
economy (about 1.9 % of Canada’s GDP).1 Forests cover a large part of Sweden and Finland,
and these countries have developed a flourishing forestry industry that contributes about 8 %
to Finland’s GDP, putting it in second place after the electronic sector.2

1 http://cfs.nrcan.gc.ca/pages/269?lang=en_CA.
2 http://www.fao.org/forestry/country/57478/en/fin/.
Ann Oper Res

Forestry operations consist of many different activities such as planting, building access
to forest locations, and transporting the harvest to wood mills. In Canada, the distances be-
tween forest locations and wood mills are generally large. Furthermore, backhauling, which
is usually performed empty, represents a waste of resources (time, fuel, etc.). It is thus im-
portant to reduce unproductive activities during transportation, for both economic and envi-
ronmental considerations.
In this paper, we consider a variant of the weekly log-truck scheduling problem (LTSP),
namely assigning trucks to move wood loads from forest locations to paper mills. In this
variant, trucks and log-loaders need to be synchronized for loading and unloading opera-
tions. The purpose of this paper is to address a number of new challenges that were not
considered in El Hachemi et al. (2013), such as lunch breaks, supply constraints, and home
bases of trucks.
The paper is organized as follows. Sections 2, 3, and 4 present the LTSP, the literature
review, and the approach adopted to solve the problem. The experimental setting is described
in Sect. 5, where computational results are also reported. Section 6 concludes the paper.

2 Log-truck scheduling problem

We consider a set F of forest areas, a set M of wood mills, a set P of wood assortments, and
a planning horizon corresponding to a set J of days. On day j of the planning horizon, wood
j
mill m ∈ M requires a quantity Dmp (the demand) of assortment p ∈ P . Similarly, a weekly
quantity Sfp (the supply) of product p ∈ P is available in forest area f ∈ F . The demand
and supply quantities are expressed in numbers of full truckloads. Loads are transported
from the forest areas to the wood mills by trucks, one load at a time. In each forest location
and at each wood mill, a single log-loader is available for loading/unloading operations.
Therefore, a single truck can be loaded/unloaded at any time in any location. If several
trucks are present, the additional trucks must wait for the log-loader to become available.
These waiting times significantly increase the transportation costs, since the drivers usually
leave the engine running while they are waiting to load or unload, so this situation must be
avoided. It should be noted that although partial truckloads are common in other countries,
full truckloads are typical in Canada because the harvested volumes are large. We therefore
choose to express all quantities in truckloads rather than in cubic meters.
The LTSP consists in constructing a transportation plan over the planning horizon in
which the required number of loads is delivered to wood mills at a minimum total cost while
respecting constraints on the supplies in the forest areas and the stock constraints at the wood
mills. The objective function includes both the cost of unproductive activities (waiting times
and deadheads) and the cost of transportation of full truckloads from forest areas to wood
mills. In our application, we also have a set of regional bases. Each base is associated with
a set of trucks, so that each truck must begin and end each day at its base. We consider
instances with one and three bases.
The LTSP is related to routing problems encountered in other industries, in particular,
the vehicle routing problem (VRP) and pick-up and delivery problems with time windows
(PDPTW). The book edited by Toth and Vigo (2001) discusses variants of the problem and
different algorithms. However, the LTSP differs significantly from these problems.

3 Literature review

Since the mid-1990s, several projects have been conducted in the forestry sector with the
aim of improving the efficiency of the transportation activities such as the control and qual-
 Ann Oper Res

ity of the truck scheduling. Among these, one should note the seminal work of Weintraub
et al. (1996) in Chile in the ASICAM project, as well as the EPO system developed by Lin-
nainmaa et al. (1995) in Finland. More recently, Palmgren et al. (2003, 2004) have proposed
a column generation scheme for tackling the LTSP. More details on this work can be found
in El Hachemi et al. (2011).
Tabu search (TS) heuristics were also applied by Gronalt and Hirsch (2007) to solve a
restricted variant of the LTSP, where the destination of each load is given a priori. El Ha-
chemi et al. (2011) considered the same problem and proposed a two-step hybrid solution
procedure: in the first step an assignment model produces a set of deadhead movements, and
in the second step a constraint programming (CP) model is used to schedule activities taking
into account the synchronization constraints between trucks and log-loaders. In El Hachemi
et al. (2013), the same authors revisit this problem and propose a constraint-based local
search procedure. This procedure is coupled with a mixed integer programming model to
address the weekly LTSP, in which the assignments of loads from forest areas to mills must
be optimized.
Flisberg et al. (2008) and Andersson et al. (2008) proposed a two-phase approach to deal
with the daily problem. First, an LP model is used to determine the flow of wood from
supply points to demand points. The result is passed to a second model that sequences these
transportation nodes into complete routes using a standard TS. The approach is run in real
time: a dispatching procedure continuously monitors the truck routes during the day and
performs the necessary updates. This procedure, which builds a route one trip at a time,
was inspired by Rönnqvist and Ryan (1995) and Rönnqvist et al. (1998). Rey et al. (2010)
addressed the problem of the daily assignment of available trucks to the delivery of the
forest products required at different destinations. They proposed an integer programming
model based on a column generation formulation, with each column representing a given
truck’s schedule for a working day. The solution approach first finds the optimal solution of
the linear relaxation of the model and then solves the integer model constructed with all the
columns generated previously. However, these papers do not consider the synchronization of
trucks and loaders. For a comprehensive discussion of synchronization problems in several
areas, see Bredström and Rönnqvist (2008), Eveborn et al. (2006, 2009).
With respect to problem size, most papers Murphy (2003), Palmgren et al. (2003),
Gronalt and Hirsch (2007), El Hachemi et al. (2011)) present case studies with 9 to 28 trucks
and 30 to 85 transport tasks per day. Flisberg et al. (2008) and Andersson et al. (2008), on
the other hand, have solved substantially larger problems ranging from 188 transport tasks
to about 2, 500 full truckloads with 15 to 110 trucks. We solve the weekly problems of
El Hachemi et al. (2013), with up to approximately 700 transport tasks and around 14 to
32 trucks, and we introduce two new instances of about the same size. For a more detailed
description of optimization problems in the forestry sector, we refer the reader to Rönnqvist
(2003) and D’Amours et al. (2008).

4 Solution approach

We focus on developing a solution method to solve a modified version of the weekly LTSP,
allowing the wood mills to operate in just-in-time mode. El Hachemi et al. (2013) propose
a hybrid local search (LS)/CP method to deal with this problem. However, new challenges,
such as accounting for lunch breaks, supply constraints, and the trucks’ home bases, led us
to seek a new model that better represents the problem at hand.
As in El Hachemi et al. (2013), we use a two-phase approach to solve the weekly LTSP.
The first phase, which we call the “tactical problem,” is identical to that presented in El
Ann Oper Res

Hachemi et al. (2013); it involves solving an IP formulation taking into account demand,
supply, and stock constraints. In this step, there are restrictions on the time availability of
trucks. This phase yields seven daily LTSPs that are solved in the second phase. For each of
these LTSPs, there is a fixed set of transportation requests. The novelty of this paper is a new
flow-based IP model to address all the other issues of the daily LTSP. Each component of a
truck trip is represented in the network as an arc, and the constraints are enforced through
capacity constraints on the truck flows. The tactical model has been presented in El Hachemi
et al. (2013); we briefly recall the tactical problem before presenting the flow-based model
of the second phase.

4.1 Tactical problem

In the tactical model, we take into consideration the fact that different wood products must
be shipped to the wood mills. These multi-product demand constraints arise because many
wood mills order logs in specific lengths and diameters to produce given final products and
because the properties of a wood depend on the particular tree species from which it comes.
Thus, the logs are sorted into different groups that depend on species, usage, quality, and
dimension. Each supply point consists of a given group (up to 5 products in our case), and
each demand point represents a requirement for a given group. In general, the inventory is
known at the beginning of the week since it is the stock associated with the last day of the
previous week. In some cases, when the demands at the wood mills and the supplies in the
forest areas remain constant for a long period, it is sensible to generate a weekly solution
that can be repeated for the whole period. To achieve this goal, we require the inventory at
the beginning of the first day to be equal to the inventory at the end of the last day of the
week.
In this phase, the objective is to minimize the cost of full truckloads subject to the fol-
lowing constraints. First, we ensure that the daily stock of any product at any wood mill
does not exceed a given capacity. Second, we satisfy the wood mills’ daily demands for
each product over the whole week. We consider an upper limit on the number of hours that a
loader may operate on any day. Similarly, if a loader is active on a given day it must operate
for a minimum number of hours. These limits can also be expressed in loads.

4.2 Phase 2: Daily synchronized log-truck scheduling problem

We present an IP model to deal with the daily synchronized log-truck scheduling problem
(SLTSP). This model is very close to a network flow formulation. Each component (activ-
ity) of a truck trip (deadhead, loading, loaded travel, unloading) is modeled as an arc in a
network; truck waiting times are also modeled (see Fig. 1). Our model relies on a space-
time representation of the truck movements. Time is discretized into time slices of a fixed
duration. The time step is a key parameter of the model, since it determines the precision of
the solutions.

4.3 Flow-based model

    
Min cab xab + cl tfmax − tfmin
b∈B ab ∈Eb ∪Wb f ∈F

xab ≤ Vb , ∀b ∈ B (1)
ab ∈A+ (sb )
 Ann Oper Res

Fig. 1 Enriched Network Model

 
xab = xab , ∀b ∈ B (2)
ab ∈A+ (sb ) ab ∈A− (tb )
 
xab = xab , ∀b ∈ B, ∀i ∈ Nb − {sb , tb } (3)
ab ∈A+ (i) ab ∈A− (i)
 
xab ≤ 1, ∀f ∈ F, ∀T ≤ H (4)
b∈B ab ∈AT
bf
 
xab ≤ 1, ∀m ∈ M, ∀T ≤ H (5)
b∈B ab ∈AT
bm
 
H − Tasb xab ≤ H − tfmin , ∀b ∈ B, ∀f ∈ F, ∀ab ∈ Cbf (6)
≤
Taeb xab tfmax , ∀b ∈ B, ∀f ∈ F, ∀ab ∈ Cbf (7)
 
xab = K f m , ∀f ∈ F, ∀m ∈ M (8)
b∈B ab ∈Lbf m

xab ∈ {0, 1}, ∀b ∈ B, ∀ab ∈ Cb ∪ Db ∪ Eb ∪ Lb (9)

xab ∈ {0, 1, 2, . . . , Vb }, ∀b ∈ B, ∀ab ∈ Wb (10)
Ann Oper Res

The objective function is to minimize the daily deadheads and waiting-time cost of the
trucks and forest log-loaders. Constraint (1) restricts the number of vehicles available at each
regional base. Constraints (2) and (3) are flow conservation constraints for each base. Con-
straints (4) and (5) ensure that each loader serves only one truck at a given time. Constraints
(6) and (7) compute the start time and end time of each forest log-loader; these values appear
in the objective function. Since the goal is to minimize the overall cost of unproductive ac-
tivities, the optimizer will set each variable tfmin to its largest possible value while satisfying
constraint (6) and set each tfmax to its smallest possible value while satisfying constraint (7).
Constraint (8) indicates the daily number of requests to be met. Constraint (9) ensures that
all arcs, except waiting-time arcs, have a unit capacity. Finally, constraint (10) ensures that
for each base b, the number of trucks waiting before or after unloading does not exceed Vb .
Forestry companies need to ensure a one-hour break between 11 a.m. and 3 p.m. for each
truck at any wood mill, when the drivers can eat lunch and the trucks can be refueled. To
satisfy this new constraint, we divide the network into two parts at the wood-mill nodes, one
before the break and one after. The arcs linking the two parts represent a one-hour break,
as specified by the forestry companies (see Fig. 1). It is important to note that a truck must
wait before leaving the wood mill or before being loaded in a forest area. Since we expect,
in general, the number of mills to be significantly smaller than the number of forest areas,
we have decided to make the trucks wait at the mills. Therefore, waiting arcs do not appear
in the representation of the forest areas.
The network contains four node types: start, end, forest area, and wood mill. There is a
single start node (s) and end node (t) for each base and each period of the planning horizon.
There is a pair of nodes grouped by forest area and wood mill for each period of the hori-
zon and each base. This clustering of nodes implicitly represents the spatial dimension of
the problem. All the nodes are sorted in chronological order; this is represented vertically in
Fig. 1. The network involves eight arc types: start of schedule (from start nodes to forest-area
nodes), end of schedule (from wood-mill nodes to end nodes), loaded trip (from forest-area
nodes to wood-mill nodes), deadhead (from wood-mill nodes to forest-area nodes), break
(linking the two parts of the network), waiting (linking vertically two successive wood-mill
nodes), loading (linking horizontally two successive forest-area nodes) and unloading (link-
ing horizontally two successive wood-mill nodes). For start of schedule, end of schedule,
loaded trip, and deadhead arcs, the cost is proportional to the associated distance and the
operational cost provided by our partner. Each waiting-time arc has a fixed duration and
therefore a fixed cost proportional to its operational cost. It should be noted that loads are
not represented in this network, but constraint (8) ensures that enough loads are transported
between each forest area and each wood mill.

4.4 Discretization

One of the most interesting aspects of our study is that we can exploit the fact that loading
and unloading times are approximately equal (around 20 minutes, although in practice load-
ing takes slightly longer than unloading). This property allows us to set the discretization
step equal to the loading time; it thus leads to a simplification of the constraints and the
removal of constraints (4) and (5). This is particularly advantageous in the case where we
have only one base, since the restriction that each log-loader cannot serve more than one
truck at the same time is automatically enforced by the flow conservation constraints and
the unit capacity on the loading arc.
 Ann Oper Res

4.5 Algorithmic approach to synchronized log-truck scheduling problem

We propose a two-stage approach to deal with each daily problem resulting from the tactical
problem. We use a 20-minute step-size in the first stage and a 10-minute step-size in the
second stage. We use a fixed computational time that depends only on the number of bases
involved. We carry out the second stage only if the relative gap in the first stage is less than
1 %. We use a warm-start strategy: the initial solution for the second stage is the best solution
found in the first stage. The goal is to use a smaller step-size to refine the best solution found
in the first stage.

4.6 Branching strategy

We developed a branching strategy that we call “start late end early” (SLEE) based on re-
ducing the forest-loader wait time to minimize the wait-time component of the objective
function. We focused on forest log-loader wait times since the hourly log-loader wait-time
cost is approximately double the hourly cost of a truck. To implement this, we use two load-
ing arcs (variables) for the same forest area, one in each direction, and we branch first on
the earlier and then the later partial loading arc (variable). This strategy leads to a reduction
in the work time for each forest log-loader, and since the daily requests in the second phase
are known in advance (a result of phase 1), this approach will induce the minimization of
the forest log-loader wait times. Thus, the role of the SLEE branching strategy is to help the
optimizer set each variable tfmin to its largest possible value by branching on the earlier frac-
max
tional xab and to set each variable
 tf to its smallest possible value by branching on the later
fractional xab such that ab ∈ b∈B Cbf . Our branching procedure consists to assign a static
priority to all loading variables as follows: ∀f ∈ F, ∀b ∈ B, ∀T ≤ H such that T ≤ H − T ,
H −T
∀a ∈ ATbf , ∀a ∗ ∈ Abf , we set that the priority of xa is equal to the priority of xa ∗ = H − T .
Variables with higher priorities will be branched on before variables with lower priorities
(when the variables have fractional values). We make sure to first branch down.

5 Experimental results

We were provided with four different case studies by FPInnovations.3 Cases 1 and 2 in-
volve six forest areas and five wood mills. Case 1 involves approximately 400 shipments
(logs) per week, and it has an average cycle time of about 4 hours to transport a ship-
ment. Case 2 involves approximately 700 shipments per week, with an average cycle time
of about 5.5 hours. Both cases have three products to transport. Cases 3 and 4 have nine
and eleven forest areas, respectively, and seven wood mills. The number of shipments
per week is 560 and 583 respectively and the average cycle times are about 6.3 and 4.16
hours. There are five products to transport. The data for these case studies can be found at
http://w1.cirrelt.umontreal.ca/~louism/Data_LTSP.xls.
FPInnovations also provided approximate operational costs: $60 per hour for each truck
if it is waiting and $70 if it is traveling. A loader costs more: $100 per hour when it is
waiting. We performed tests with different numbers of trucks and with either one or three
regional bases. Each test consists in solving seven daily instances (weekdays) resulting from
the tactical phase. We used a computational time of two hours for one base and six hours for

3 FPInnovation is a private, not-for-profit R&D organization whose mission is to improve Canadian forestry
operations.
Ann Oper Res

Table 1 Tactical phase results

Case Full truckload cost ($)

Case 1 24,990
Case 2 80,103
Case 3 66,407
Case 4 73,733

Table 2 Comparison of the approaches (10 minutes of computational time)

Case V ILS/CP approach ($) Flow-based approach ($) Improvement (%)

Case 1 14 54,170 52,743 2.63

15 53,430 51,503 3.60
16 53,043 51,590 2.73

Case 2 30 137,883 130,394 5.43

31 139,550 130,567 6.05
32 139,966 129,687 6.35

three bases. In the tactical phase, we fixed the computation time to 5 minutes; all the tactical
problems were solved optimally in less than 1 minute. The associated results are reported in
Table 1. All the tests were run on a cluster of Intel Itanium II 1.5-GHz processors.
We compared our flow-based approach with the ILS/CP method recently published by El
Hachemi et al. (2013). Table 2 presents the results for cases 1 and 2; note that V = b Vb .
Clearly, our flow-based approach outperforms ILS/CP. The difference is considerable for
case 2, where there are fewer trips and vehicles.
We report in Tables 3 and 4 the average gap, the maximum and minimum gaps, the
average distance from the best known solution (BKS) where the BKS is the best solution
found by all the approaches studied in this paper, the average waiting-time cost, the average
total cost, and the time to reach the best solution. Comparing the default branching strategy
(DBS) of Cplex 12.4.0 with the SLEE strategy for our flow-based model shows that the
latter generally outperforms the former in terms of solution quality except in case 3 with
one base. Note that the difference between the two strategies is not large (less than 1 % for
all scenarios). This is because the waiting-time cost represents less than 6 % of the overall
cost, and therefore the SLEE strategy has an impact on only a small component of the overall
objective function. Although the reduction in terms of the average total cost is less than 1 %,
it is helpful to minimize the average waiting-time cost because this part of the solution is
directly perceived by the operators of the log-loaders.
The gap size increases when the number of bases increases. This is expected since the
problem size (constraints and variables) also increases. The results show that the average gap
size is smaller for large problems than for smaller problems (case 1) (see Tables 3 and 4).
Bramel and Simchi-Levi (1997) have discussed the impact of the problem size on the gap
for set covering formulations of the vehicle routing problem with time windows. They show
that the relative gap between fractional and integer solutions becomes progressively smaller
as the number of customers increases.
The analysis of the relative gap is complex, since it is associated with the lower bound
(linear relaxation in our case) and the integer solution. The difference observed between the
 Ann Oper Res

Table 3 Results for one regional base

Criteria Case 1 Case 2 Case 3 Case 4

DBS SLEE DBS SLEE DBS SLEE DBS SLEE

Gap(%) 3.4 3.1 1 1 1.5 1.6 1.6 1.5

Gapmax (%) 5 4.4 1.2 1.2 1.7 1.8 2 2
Gapmin (%) 1.8 1.4 0.8 0.9 1.2 1.4 1.2 1.2
GapBKS (%) 0.5 0 0.5 0 0 0.08 0.01 0
W aiting ($) 2,430 2,035 4,565 4,480 3,892 4,101 3,792 3,786
Cost ($) 65,692 65,341 157,908 157,065 140,909 141,016 132,049 132,034
T ime (s) 10,854 6,738 4,042 4,111 13,419 12,064 23,231 12,095

Table 4 Results for three regional bases

Criteria Case 1 Case 2 Case 3 Case 4

DBS SLEE DBS SLEE DBS SLEE DBS SLEE

Gap (%) 5.3 5.1 1.9 1.7 4.5 4.2 5.5 4.6
Gapmax (%) 7.4 5.9 2.3 2 5.1 4.7 6.4 5.5
Gapmin (%) 3.4 4.2 1.6 1.5 3.7 3.7 4.7 3.9
GapBKS (%) 0.26 0 0.1 0 0.17 0 0.72 0
W aiting ($) 3,461 3,188 5,185 4,978 6,748 6,395 7,731 6,785
Cost ($) 66,548 66,372 158,275 158,098 143,552 143,296 135,394 134,418
T ime (s) 22,539 16,078 14,205 7,866 19,072 17,616 17,041 17,633

relative gap of the two strategies is in most cases in favor of the SLEE strategy. In terms of
computational time, we notice that the SLEE strategy usually finds a good-quality solution
quickly. This reduction in the average time to reach the best solution can reach 48 %.
Finally, as mentioned earlier, we use a 10-minute step-size only when the relative gap
for the 20-minute step-size is less than 1 %. Our goal is to refine the best solution found by
the larger step-size. In practice, cases 2, 3, and 4 were solved with both step sizes, but the
smaller step-size improved the solutions of just two daily SLTSPs associated with one base;
the improvement was about 2 %.

6 Conclusion

We have described a two-phase method to solve the weekly log-truck scheduling problem,
which includes a just-in-time delivery policy, a subject of great interest to the forestry in-
dustry.
The first phase determines the optimal number of full truckloads to deliver from forest
areas to wood mills by minimizing the loaded-transportation and forest-area costs. This
phase yields seven daily SLTSPs. To solve these, we developed a flow-based model and a
specialized branching strategy. Our computational experiments show that this specialized
strategy provides solutions more rapidly and of higher quality than those produced by the
default branching strategy of Cplex 12.4.0.
Future research directions will involve solving the problem in one step, using a La-
grangian relaxation approach by relaxing all the constraints that link two successive days.
Ann Oper Res

We also plan to produce robust solutions by including stochasticity in the trip times and
the loading/unloading times. We would also like to include additional constraints related
to driver breaks and shift changes and to investigate the case where the number of home
bases equals the number of trucks. We believe that column generation approaches may be
beneficial in this context.

References

Andersson, G., Flisberg, P., Lidén, B., & Rönnqvist, M. (2008). RuttOpt—a decision support system for
routing of logging trucks. Canadian Journal of Forest Research, 38, 1784–1796.
Bramel, J., & Simchi-Levi, D. (1997). On the effectiveness of set covering formulations for the vehicle routing
problem with time windows. Operational Research, 45(2), 295–301.
Bredström, D., & Rönnqvist, M. (2008). Combined vehicle routing and scheduling with temporal precedence
and synchronization constraints. European Journal of Operational Research, 191(1), 19–31.
D’Amours, S., Rönnqvist, M., & Weintraub, A. (2008). Using operational research for supply chain planning
in the forest products industry. INFOR. Information Systems and Operational Research, 46(4), 265–
281.
El Hachemi, N., Gendreau, M., & Rousseau, L.-M. (2011). A hybrid constraint programming approach to the
log-truck scheduling problem. Annals of Operations Research, 184(1), 163–178.
El Hachemi, N., Gendreau, M., & Rousseau, L.-M. (2013). A heuristic to solve the synchronized log-truck
scheduling problem. Computers & Operations Research, 40(3), 666–673.
Eveborn, P., Flisberg, P., & Rönnqvist, M. (2006). Laps care—an operational system for staff planning of
home care. European Journal of Operational Research, 171(3), 962–976.
Eveborn, P., Rönnqvist, M., Einarsdóttir, H., Eklund, M., Lidén, K., & Almroth, M. (2009). Operations re-
search improves quality and efficiency in home care. Interfaces, 39(1), 18–34.
Flisberg, P., Lidén, B., & Rönnqvist, M. (2008). A hybrid method based on linear programming and tabu
search for routing of logging trucks. Computers & Operations Research, 36(4), 1122–1144.
Gronalt, M., & Hirsch, P. (2007). Log-truck scheduling with tabu search strategy. Metaheuristics, 39, 65–88.
Linnainmaa, S., Savalo, J., & Jokinen, O. E. P. (1995). A knowledge based system for wood procurement
management. Paper presented at the 7th Annual Conference on Artificial Intelligence, Montreal
Murphy, G. (2003). Reducing trucks on the road through optimal route scheduling and shared log transport
services. Southern Journal of Applied Forestry, 27(3), 198–205.
Palmgren, M., Rönnqvist, M., & Värbrand, P. (2003). A solution approach for log truck scheduling based
on composite pricing and branch and bound. International Transactions in Operational Research, 10,
433–447.
Palmgren, M., Rönnqvist, M., & Värbrand, P. (2004). A near-exact method for solving the log-truck schedul-
ing problem. International Transactions in Operational Research, 11, 447–464.
Rey, P. A., Muñoz, J. A., & Weintraub, A. (2010). A column generation model for truck routing in the Chilean
forest industry. INFOR. Information Systems and Operational Research, 47(3), 215–221.
Rönnqvist, M. (2003). Optimization in forestry. Mathematical Programming, 97, 267–284.
Rönnqvist, M., Sahlin, H., & Carlsson, D. (1998). Operative planning and dispatching of forestry transporta-
tion. Report LiTH-MAT-R-1998-18, Linköping University, Linköping, Sweden
Rönnqvist, M., & Ryan, D. (1995). Solving truck dispatch problems in real time. In Proceedings of the 31st
annual conference of the operational research society of New Zealand, Wellington, New Zealand, 31
August–1 September 1995 (pp. 165–172). Auckland: Operational Research Society of New Zealand.
Toth, P., & Vigo, D. (2001). The vehicle routing problem. Philadelphia: Society for Industrial and Applied
Mathematics.
Weintraub, A., Epstein, R., Morales, R., Seron, J., & Traverso, P. (1996). A truck scheduling system improves
efficiency in the forest industries. Interfaces, 26(4), 1–12.