1 s2.0 S1366554511001426 Main

Transportation Research Part E 48 (2012) 616–636
Contents lists available at SciVerse ScienceDirect
Transportation Research Part E

journal homepage: www.elsevier.com/locate/tre
The time dependent vehicle routing problem with time windows:

Benchmark problems, an efficient solution algorithm, and solution
characteristics
Miguel Andres Figliozzi
Maseeh College of Engineering and Computer Science, PO Box 0751, Portland State University, Portland, OR 97207-0751, United States
a r t i c l e i n f o a b s t r a c t
Article history: An algorithm that can tackle time dependent vehicle routing problems with hard or soft
Received 15 August 2011 time windows without any alteration in its structure is presented. Analytical and experi-
Received in revised form 23 September 2011 mental results indicate that average computational time increases proportionally to the
Accepted 6 November 2011
number of customers squared. New replicable test problems that capture the typical speed
variations of congested urban settings are proposed. Solution quality, time window pertur-
bations, and computational time results are discussed as well as a method to study the
Keywords:
impact of perturbations by problem type. The algorithm efficiency and simplicity is well
Vehicle routing
Benchmark problems
suited for urban areas where fast running times may be required.
Time dependent travel time Ó 2011 Elsevier Ltd. All rights reserved.
Variable speed
Congestion
1. Introduction
Congestion is a common phenomenon in most urban areas of the world. Congestion creates a substantial variation in tra-
vel speeds during peak morning and evening hours. This is problematic for all vehicle routing models that rely on a constant
value to represent vehicle speeds. Urban route designs that ignore these significant speed variations result in inefficient and
suboptimal solutions. Poorly designed routes that lead freight vehicles into congested arteries and streets not only increase
supply chain and logistics costs but also worsen externalities associated with freight traffic in urban areas such as green-
house gases, noise, and air pollution. Travel time between customers and depot is found to be a crucial factor that amplifies
the negative impacts of congestion; congestion also affects carriers’ cost structure and the relative weight of wages and over-
time expenses (Figliozzi, 2010a).
Routing models with time-varying travel times are gaining greater attention in vehicle routing literature and industry.
However, research on the time dependent vehicle routing problem (TDVRP) is still comparatively meager in relation to
the body of literature accumulated for the classical vehicle routing problem (VRP) and vehicle routing problem with time
windows (VRPTW). In addition, published algorithms and related results can neither be readily benchmarked nor do they
cover all practical and relevant objective functions or time window constraint types. Without readily available benchmark
problems the solution quality of TDVRP algorithms cannot be properly analyzed. Furthermore, both solution quality and
computation times are key attributes of a desirable TDVRP algorithm. Fast algorithms are particularly critical in urban TDVRP
because travel time updates and new solutions may be necessary due to non-recurrent events such as accidents.
The goals of this research are to: (a) formulate a time dependent vehicle routing problem with a general cost function and
time window constraints, (b) present an intuitive and efficient solution methodology for problems with time dependent
speeds, (c) introduce readily replicable time dependent instances and analyze the computational results and robustness
E-mail address: figliozzi@pdx.edu
1366-5545/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.tre.2011.11.006
M. Andres Figliozzi / Transportation Research Part E 48 (2012) 616–636 617
of the solutions, and (d) analyze the computational complexity of the solution approach. This paper is organized as follows:
Section 2 presents a literature review for the TDVRP; Section 3 introduces notation and formulates the problems. Section 4
presents the Iterative Route Construction and Improvement algorithm (IRCI) to solve time dependent routing problems and
minimize fleet size; Section 5 presents algorithms to reduce soft time window penalties and route durations; Section 6 pre-
sents benchmark problems; Section 7 discusses computational results; Section 8 analyses the worst case and average com-
putational complexity of the algorithms presented in Sections 4 and 5; and Section 9 concludes the paper.
2. Literature review
Unlike widely studied versions of the VRP, e.g. capacitated VRP or time windows VRP with constant travel speed or no
travel time variability, time dependent problems have received considerably little attention. The time dependent VRP was
first formulated by Malandraki (1989) and Malandraki and Daskin (1992) using a mixed integer linear programming formu-
lation. A greedy nearest-neighbor heuristic based on travel time between customers was proposed, as well as a branch and
cut algorithm to solve TDVRP without time windows. Hill and Benton (1992) considered a node based time dependent vehi-
cle routing problem (without time windows). Computational results for one vehicle and five customers were reported. Ahn
and Shin (1991) discussed modifications to the savings, insertion, and local improvement algorithms to better deal with
TDVRP. In randomly generated instances, they reported computation time reductions as a percentage of ‘‘unmodified’’ sav-
ings, insertion, and local improvement algorithms. Malandraki and Dial (1996) proposed a dynamic programming algorithm
for the time dependent traveling salesman problem, i.e. for a fleet of just one vehicle. A nearest-neighbor type heuristic was
used to solve randomly generated problems.
An important property for time dependent problems is the First In–First Out (FIFO) property (Ahn and Shin, 1991; Ichoua
et al., 2003). A model with a FIFO property guarantees that if a vehicle leaves customer i to go to customer j at any time t, any
identical vehicle with the same destination leaving customer i at a time t + e, where e > 0, will always arrive later. This is an
intuitive and desirable property though it is not present in all models. Earlier formulations and solutions methods,
Malandraki (1989), Malandraki and Daskin (1992), Hill and Benton (1992), and Malandraki and Dial (1996), do not guarantee
the FIFO property as reported by Ichoua et al. (2003). Later research efforts have modeled travel time variability using
‘‘constant speed’’ time periods which guarantees the FIFO property, as shown by Ichoua et al. (2003).
Ichoua et al. (2003) proposed a tabu search solution method, based on the work of Taillard et al. (1997), in order to solve
time dependent vehicle routing problems with soft time windows. Ichoua et al. showed that ignoring time dependency, i.e.
using VRP models with constant speed, can lead to poor solutions. Ichoua et al. tested their method using the Solomon prob-
lem set, soft time windows, three time periods, and three types of time dependent arcs. The objective was to minimize the
sum of total travel time plus penalties associated with time window violations. Summarizing, Ichoua et al. solved a different
and simpler problem than the one analyzed in this paper. This research deals with a problem with hard time windows
whereas Ichoua et al. deals with a problem with soft time windows only. Finally, Ichoua et al. assume that the number of
routes is known a priori and that violations to time window constraints are acceptable. Furthermore, the objective function
in Ichoua et al. is to minimize a weighted sum of total distance travelled and total lateness over all customers whereas this
research follows the more traditional hierarchical approach used in the vast majority of VRP problems: first minimize num-
ber of routes and second minimize time and/or distance.
Other approaches include the work of Fleischmann et al. (2004) who utilized a route construction methods already pro-
posed in the literature, savings and insertion, to solve uncapacitated time dependent VRP with and without time windows.
Fleischmann et al. tested their algorithms in instances created from Berlin travel time data. Jung and Haghani proposed a
genetic algorithm to solve time dependent problems (Jung and Haghani, 2001; Haghani and Jung, 2005). Using randomly
generated test problems, the performance of the genetic algorithm was evaluated by comparing its results with exact solu-
tions (up to 9 customers and 15 time periods) and a lower bound (up to 25 customers and 10 time periods).
More recently Van Woensel et al. (2008) used a tabu search to solve the capacitated vehicle routing problem with time
dependent travel times. To determine travel speed, approximations based on queuing theory and the volumes of vehicles in a
link were used. Van Woensel et al. solved capacitated VRP (with no time windows) for problem sizes between 32 and 80
customers. Donati et al. (2008) proposed a solution adapting the ant colony heuristic approach and a local search improve-
ment approach that stores and updates the slack times or feasible delays. The heuristic was tested using a real life network in
Padua, Italy, and some variations of the Solomon problem set. The latest work includes a paper by Soler et al. (2009) who
proposed a method to solve, optimally, TDVRP instances that are too small for practical purposes and likely exponential
growth of computational time as a function of problem size. Dabia et al. (2010) deals with a one-vehicle vehicle routing prob-
lem using a dynamic programming approach. Kok (2010) deals with the TDVRP with a focus on departure time optimization
and driver break scheduling. This work actually uses a modification of the set of benchmark instances for the VRP with time-
dependent travel speeds proposed by an early working paper by Figliozzi (2009). In addition to the work of Kok, only two
papers use well known benchmark problems with time windows. Ichoua et al. (2003) used the widely known Solomon prob-
lems for the VRP with time windows. However, capacity constraints were not considered, optimal fleet size was given, and
no details were provided regarding how links were associated with ‘‘categories’’ that represent differences in the urban
network (i.e. main arteries, local streets, etc.). Donati et al. (2008) also used Solomon instances, however, the results cannot
be compared with previous results by Ichoua et al. (2003) because a different time speed function was used and capacity
constraints were considered. In addition, the exact instances used by Donati et al. (2008) cannot be reconstructed because
618 M. Andres Figliozzi / Transportation Research Part E 48 (2012) 616–636
the different travel speeds were randomly assigned to arcs. Therefore, no study published to date can be swiftly replicated
and solution qualities and computation times cannot be compared.
Comparisons are also problematical because objective functions and routing constraints for time dependent problems are
often dissimilar, unlike VRPTW research where the objective function is hierarchical and usually considers fleet size (primary
objective), distance (secondary objective), and total route duration. Ichoua et al. (2003) study the TDVRP with soft time win-
dows and consider as the objective function total duration plus lateness and assume that the optimal fleet size is given a
priori. Haghani and Jung (2005) minimize the sum of costs associated with number of vehicles, distance, duration, and late-
ness. Fleischmann et al. (2004) minimize number of vehicles and total duration. Donati et al. (2008) optimizes fleet size (pri-
mary objective) and total route duration (secondary objective).
Summarizing, different solution approaches such as tabu search, ant colony, and genetic algorithms have been proposed
for the VRPTD. The computational complexity of the existing TDVRP approaches have never been discussed or analyzed
though some solution approaches are not fast as later discussed in Section 7 where solution quality and running times
are compared. In addition, benchmark instances have not been yet proposed. Section 8 discusses the Iterative Route Con-
struction and Improvement (IRCI) TDVRP algorithmic complexity; benchmark instances that can be clearly and unmistakably
replicated by future researchers are detailed in Section 6. The next section introduces mathematical notation and defines the
problem under study. To the best of the author’s knowledge there is no published work that presents replicable time depen-
dent vehicle routing problems with hard time windows constraints.
3. Problem definition
Using a traditional flow-arc formulation (Desrochers et al., 1988), the time dependent vehicle routing problem with hard
time windows studied in this research can be described as follows. Let G = (V, A) be a graph where A = {(vi, vj): i – j ^ i, j e V} is
an arc set and the vertex set is V = (v0, . . . , vn+1). Vertices v0 and vn+1 denote the depot at which vehicles of capacity qmax are
based. Each vertex in V has an associated demand qi P 0, a service time g i P 0, and a service time window [ei, li]; in partic-
ular the depot has g0 = 0 and q0 = 0. The set of vertices C = {v1, . . . , vn} specifies a set of n customers. The arrival time of a vehi-
cle at customer i, i e C is denoted ai and its departure time bi. Each arc (vi, vj) has an associated constant distance dij P 0 and a
travel time tij ðbi Þ P 0 which is a function of the departure time from customer i. The set of available vehicles is denoted K.
The cost per unit of route duration is denoted ct; the cost per unit distance traveled is denoted cd. It is assumed that the prob-
lem is feasible, i.e. it is always possible to feasibly serve any individual customer starting from the depot.
The primary objective function for the TDVRP is the minimization of the number of routes; the optimal number of routes
is unknown. A secondary objective is the minimization of total time or distance. There are two decision variables in this for-
mulation; xkij is a binary decision variable that indicates whether vehicle k travels between customers i and j. The real number
decision variable yki indicates service start time for customer i served by vehicle k. The TDVRP is formulated as follows:
Primary objective:
XX
minimize xk0j ; ð1Þ
k2K j2C
Secondary objective:
X X k
XX
minimize cd dij xkij þ ct ðyknþ1 yk0 Þxk0j ; ð2Þ
k2K ði;jÞ2A k2K j2C
subject to:
X X
qi xkij 6 qmax ; 8k 2 K ð3Þ
i2C j2V
XX
xkij ¼ 1; 8i 2 C ð4Þ
k2K j2V
X X
xkil xklj ¼ 0; 8l 2 C; 8k 2 K ð5Þ
i2V i2V
xki0 ¼ 0; xknþ1;i ¼ 0; 8i 2 V; 8k 2 K ð6Þ

X
xk0j ¼ 1; 8k 2 K ð7Þ
j2V
X
xkj;nþ1 ¼ 1; 8k 2 K ð8Þ
j2V
X
ei xkij 6 yki ; 8i 2 V; 8k 2 K ð9Þ
j2V
X
li xkij P yki ; 8i 2 V; 8k 2 K ð10Þ
j2V
xki;j ðyki þ g i þ t i;j ðyki þ g i ÞÞ 6 ykj ; 8ði; jÞ 2 A; 8k 2 K ð11Þ
xkij 2 f0; 1g; 8ði; jÞ 2 A; 8k 2 K ð12Þ
yki 2 R; 8i 2 V; 8k 2 K ð13Þ
The primary and secondary objectives are defined by (1) and (2) respectively. The constraints are defined as follows: vehicle
capacity cannot be exceeded (3); all customers must be served (4); if a vehicle arrives at a customer it must also depart from
that customer (5); routes must start and end at the depot (6); each vehicle leaves from and returns to the depot exactly once,
(7) and (8) respectively; service times must satisfy time window start (9) and ending (10) times; and service start time must
allow for travel time between customers (11). Decision variables type and domain are indicated in (12) and (13).
#
In the TDVRP with soft time windows, customer service time windows are defined by two intervals [ei, li] and ½e# i ; li
# # #
where ei 6 e#i ; l i 6 l i . The interval ½e i ; l i indicates the interval of time where service can start without incurring a penalty.
The interval [ei, li] indicates the interval of time where service can start but there are additional costs, associated to ce or cl, if
#
service starts early or late, respectively – i.e. during the early interval ½ei ; e# e
i or during the late interval ½li ; li . Defining xi and
l
xi as auxiliary binary variables that indicate whether a penalty is incurred, the objective functions can be expressed as
follows:
X
minimize xli ; 8i 2 C ð14Þ
i2C
X
minimize xei ; 8i 2 C ð15Þ
i2C
X X k
XX XX # XX
minimize cd dij xkij þ ct ðyknþ1 yk0 Þxk0j þ ce ðei yki Þþ þ cl ðyki e#
l Þ
þ
ð16Þ
k2K ði;jÞ2A k2K j2C k2K i2C k2K i2C
subject to
þ
xli ðyki e# l
l Þ 6 xi ð17Þ
k þ
xei ðe# e
i y i Þ 6 xi ð18Þ
xei 2 f0; 1g; xli 2 f0; 1g ð19Þ

The primary objective function for the TDVRP with soft time windows is still the minimization of the number of routes. Using
a customer service perspective ranking, a secondary objective is the minimization of the number of late penalties1 (14); a
tertiary objective is the minimization of early penalties (15); a final objective is the minimization of the combined distance,
route duration, and soft time window costs (16). Logical constraints (17) and (18) are used to determine if service times must
be penalized due to early or late time window utilization, respectively.
It is important to notice that the depot time windows as well as the maximum route duration are not changed as a result
of the customers’ time window relaxation. The TDVRP with hard time windows is a special case of the soft time window
#
formulation. If ei ¼ e#i and li ¼ li ; then (14) and (15) are redundant and (16) is reduced to (2). The travel time speed in
any arc is a positive and continuous function of time, sij(t) > 0, which guarantees the FIFO property (Ahn and Shin, 1991).
In addition, in the presented TDVRP travel times may be asymmetrical, i.e. t i;j ðyki Þ–t j;i ðykj Þ even if yki ¼ ykj .
Unlike previous formulations of the TDVRP (Malandraki, 1989; Jung and Haghani, 2001) time is not partitioned into dis-
crete intervals. Furthermore, the decision variable yki allows for waiting at customer i; service start time may not necessarily
be the same as arrival time. For example, if the vehicle arrives too early, it can wait at the customer location to avoid early
service penalties. However, waiting may have an impact on future travel times. The following two sections describe a solu-
tion approach to tackle the TDVRP.
4. Solution approach
Time dependent travel times require significant modifications to local search approaches and metaheuristics that have
been successfully applied to the traditional constant time VRPTW (Braysy and Gendreau, 2005a,b). A customer insertion
or a local improvement not only influences the arrival and departure times of a ‘‘local’’ subset of customers but it may also
1
Although the cost of early and late service times are application dependent, in numerous real life problems early services are preferred over late services,
e.g. blood transport, just-in-time production systems, express mail delivery, etc.
significantly change travel times among ‘‘local’’ customers. Furthermore, the impact of altering a routing sequence is not just
‘‘local’’ but potentially affects all subsequent travel times. Changes in travel times have a subsequent impact on feasibility. To
a certain degree, introducing soft time windows ameliorates the computational burden and loss of efficiency introduced by
time dependent travel times. However, hard time constraints are more difficult to accommodate and this is reflected in the
literature review. There are no published results in a set of standard benchmark problems with hard time windows and time
dependent travel times. The absence of standard problems can also be attributed to the multiple different versions of the
time dependent VRP.
The presented Iterative Route Construction and Improvement (IRCI) solution approach for the TDVRP employs algorithms
that do not require modifications to accommodate constant or time dependent travel speeds. This research builds upon pre-
vious work to solve the constant speed VRP with soft and hard time windows (Figliozzi, 2008, 2010b).
The solution method to minimize fleet size is divided into two phases and algorithms: route construction and route
improvement. A third algorithm, an auxiliary route building heuristic, is reiterated during the execution of the construction
heuristic. The construction and auxiliary algorithms are sequential, i.e. the routes are built sequentially one customer at the
time but always looking ahead at the cost of building a feasible solution when a given customer is routed or sequenced. Com-
plete and feasible solutions are always returned by the route construction algorithm. The following pseudo-code illustrates
the interaction between the route construction algorithm and the auxiliary routing algorithm in the construction phase (the
calls are underlined). Sets are denoted by capital letters in italics (for example set of customers C); functions are denoted by
small letters in bold (for example the cost function c(C)); sub-algorithms are denoted by bold capital letters (for example Hr).
4.1. Construction phase
Start
1 Read problem data
2 Input initial search space D
3 for each D e D
4
Call route construction algorithm for the set of unrouted customers C
5 Set depot as the initial customer i
6 do
7 Select a set of customer candidates J to follow i, JeC
8
Call the auxiliary routing algorithm for each pair (i, j) where jeJ
9 Find pair (i, j) with least cost solution and route i
10 Update best solution Z if there was an improvement
11 Update initial customer, i j, remove i from C
12 while J – {}
13 Return best solution found Z⁄ and corresponding D
14 end for
Output:
Best solution found Z⁄and corresponding D
The second pseudo-code illustrates the interaction between the route improvement algorithm and the route construction
algorithm. The route construction algorithm is executed (one or more times) every time the improvement algorithm is exe-
cuted. The improvement procedure is also at the route level, i.e. it is not a local improvement since improvements are ob-
tained by grouping routes and improving a set of routes at the time.
4.2. Improvement phase
Start
1 Input best solution found Z and corresponding D
2 Input criteria to select subsets of routes z from Z, z Z
3 Form set X of subsets z to explore
4 For each subset z e X
5 Call route construction algorithm for customers in z
6 Return best solution found z0
7 If z0 improves the partial solution z, then replace z and update Z
8 end for
Output:
Best solution found Z
The same basic algorithm applies to problems with soft and hard time windows. Hence, the presented algorithm pro-
duces solutions for time dependent vehicle routing problems with hard and soft time windows with similar computation
times.
Using a bottom up approach the detailed descriptions of the algorithms are introduced in the following order: (a) an
algorithm to sequence any given set of customers (auxiliary algorithm), (b) the route construction algorithm, and (c)
the route improvement algorithm. In addition, due to the nature of the TDVRP, advancing or delaying service time may
have a favorable impact on future travel times and costs. Hence another algorithm (d) is described in Section 5 to optimize
service times in problems with soft time windows given a set of routes obtained from (c). Travel time calculations are nec-
essary to execute (a)–(d). However, unlike algorithms (a)–(d), travel time calculations are heavily dependent on the specific
type of speed function. Hence, the algorithm used to calculate travel times is presented in Appendix A for a specific type of
speed function.
4.3. The auxiliary routing algorithm Hr
The auxiliary routing algorithm Hr can be any heuristic that given a starting vertex, a set of customers, and a depot loca-
tion returns a set of routes that satisfy the constraints of the TDVRP with soft or hard time windows. The auxiliary route heu-
ristic is defined as Hr(D, vi, C, v0) where D = {d0, d1, . . . , d6} are the parameters of the generalized cost function, vi is the vertex
where the first route starts, C is the set of customers to route, and v0 is the depot where all routes end and all additional
routes start, with the exception of the first route which starts at vi.
In this research, Hr is a generalized nearest-neighbor heuristic (GNNH). The GNNH starts every route k by finding, from a
subset of C, the unrouted customer with the least appending ‘‘generalized cost’’. At every subsequent iteration, the heuristic
searches for the remaining unrouted customer with the least appending cost. Let i denote the initial vertex and let j denote a
potential customer to append next. Let qki denote the remaining capacity of the vehicle k after serving customer i. The service
at customer i in route k begins at the earliest feasible time, which is yki ¼ maxðai ; ei Þ, and the departure time is given by
yki þ g i . The generalized cost of going from customer i to customer j is estimated as:
k þ # þ
gðD; i; j; kÞ ¼ d1 dij þ d2 ðykj ðyki þ g i ÞÞ þ d3 ðlj ðyki þ g i þ t ij ðyki þ g i ÞÞÞ þ d4 ðqki qj Þ þ d5 ½e# k
j yj þ d6 ½yj ej ð20Þ
If customer j is infeasible, i.e. it cannot be visited after serving customer i, the cost of ending customer i’s route and starting a
new one to serve customer j is estimated as:
k þ # þ
gðD; i; jÞ ¼ d0 þ d1 d0j þ d2 ðykj e0 Þ þ d3 ðlj t 0j ðe0 ÞÞ þ d4 ðqmax dj Þ þ d5 ½e# k
j yj þ d6 ½yj ej ð21Þ
where d0 is the cost of adding a new vehicle.
The parameter d1 takes into account the relative distance between customers. The parameter d1 is included, even in prob-
lems with time windows, to reduce route length; this is particularly important in instances where time windows are wide or
not ‘‘binding’’ in all routes. The parameter d2 accounts for the ‘‘slack’’ between the completion of service at i and beginning of
service at j. Following Solomon’s approach (Solomon, 1987), the parameter d3 takes into account the ‘‘urgency’’ of serving
customer j, expressed as the time remaining until the vehicle’s last possible service time start. The parameter d4 takes into
account the capacity slack of the vehicle after serving customer j. The parameters d5 and d6 are added to account for possible
early or late service penalties, respectively.
4.4. The route construction algorithm Hc
In this algorithm, denoted Hc, routes are constructed sequentially. Given a partial solution and a set of unrouted custom-
ers, the algorithm uses the auxiliary heuristic Hr to search for the feasible least cost set of routes. The algorithm also uses an
auxiliary function w(vi, C, g, W) that, given a set of unrouted customers C, a vertex vi, vi R C, and a generalized cost function
g(D, i, j, k), returns a set of vertices of cardinality W with the lowest generalized costs g(D, i, j, k) for all vj e C. The function
c(sequenceofcustomers) simply evaluates the true cost of a sequence of costumers utilizing the objective function (16).
Parameters:
Hr: Route building heuristic
W: Width of the search, the number of solutions to be built and compared before adding a customer
to a route, W 6 jCj
D: Search space of the route heuristic generalized cost parameters, each D e D is a vector of
cost parameters
Data:
C: Set of customers to route
v0: the depot
vi: initial vertex (customer that is last in the partially formed route, or the depot in the special case
when the route is empty)
(con
Start Hc
1 start v0
2 lowestCost 1
3 bestSequence v0
4 for each D e D
5 while C – £ do
6 C w(start, C, g, W)
7 for each vi e C
8 if c(bestSequence [ Hr(D, vi, C, v0)) < lowestCost then
9 lowestCost c(bestSequence [ Hr(D, vi, C, v0))
10 lowestNext vi
11 end if
12 end for
13 start lowestNext
14 C C n lowestNext
15 bestSequence bestSequence [ lowestNext
16 R bestSequence [ Hr(D, lowestNext, C, v0)
17 end while
18 end for
Output:
Best set of routes found R to serve all C customers (R and its associated D)
END Hc
This algorithm will sequentially construct routes. In line 6, up to W ‘‘potentially’’ good candidates are selected. Line 8 at-
tempts to ‘‘look ahead’’ and estimate the future costs of adding vi to the existing sequence of routed customers. Between lines
7 to 12 complete routes (built using the W candidates and the heuristic Hr) are compared. The candidate with the least gen-
eralized cost is selected. The generalized cost function g that is used in Hr must not be confused with the objective cost func-
tion c that is used in Hc or the improvement heuristic Hi; the latter cost function is the sum of the accrued vehicle, distance,
time, or penalty costs as indicated in the objective function.
Each D e D is likely to produce a different solution in terms of fleet size, distance traveled, and route duration. The search
space is limited by these constraints:
d1 þ d2 þ d3 ¼ 1
The parameter d4 is bounded by zero and the ratio between the median intercustomer distance and the median customer
demand. The parameters d5 and d6 are bounded between zero and the possible early or late service penalties, respectively.
All parameters are assumed to be non-negative.
4.5. The route improvement algorithm Hi
The route construction algorithm generates an initial set of routes. Any route obtained from Hc is a special cluster of cus-
tomers with the desirable property that there is at least one feasible sequence that satisfies all the constraints of the TDVRP.
Fleet size and routing costs can be further reduced using a route improvement algorithm. The improvement algorithm works
on a subset of routes. The gist of the algorithm resembles the ruin and recreate approach presented by Schrimpf et al. (2000)
for the TSP and VRP.
The motivation of the route improvement algorithm is to combine these routes, ‘‘feasible clusters’’, to consolidate or im-
prove the efficiency of routes that are not fully utilized in terms of vehicle capacity, route duration, number of customers
serviced, etc. In the Hi algorithm two functions are introduced. The function ks(R, s) orders the set of routes R from smallest
to largest based on the number of customers per route and then returns a set of s P 1 routes with the least number of cus-
tomers; e.g. ks(R, 1) will return the route with the least number of customers. The number of customers per route is used as a
proxy measure of potential route utilization in terms of duration or vehicle capacity. If two or more routes have the same
number of customers, ties are solved drawing random numbers.
The function kg(ri, S, p) returns a set of p routes that belong to S (a complete solution already) and are good matches for
route ri. The term ‘‘good matches’’ refers to routes that as a group have the potential to be consolidated or improved. In this
research a linear combination of two distinct measures are used to evaluate the quality of a potential ‘‘match’’: (1) geograph-
ical proximity – the distance between any two routes’ center of gravity is used as a proxy measure of geographic proximity
assuming that close routes have the potential to be improved and (2) utilization – the number of customers per route is used
as a proxy measure of potential route capacity utilization assuming that routes are poorly utilized have the potential to be
combined. Other measures of geo-temporal proximity or utilization can be used to find good potential matches. By defini-
tion, the route ri is always included in the output of the set function kg(ri, S, p). To simplify notation the term C(G) is the set of
customers served by the set of routes G.
Data:
R: Set of routes
Parameters:
Hc: Route construction heuristic
W: Number of solutions to be built and compared in the construction heuristic
D: Generalized cost parameters of the auxiliary routing algorithm
s: Number of routes potentially considered for improvement
p: Number of neighboring routes that are reconstructed
ks and Route selection functions
kg:
Start Hi
1 s min(s, |R|)
2 p min(s, p)
3 S ks(R, s) # R
4 S0 RnS
5 r ks(S, 1)
6 while |S| > 1 do
7 G kg(r, S, p)
8 G0 Hc(Hr, W, D, C(G))
9 if c(G0 ) < c(G) then
10 R RnG
11 R R [ G0
12 S SnG
13 S S [ G0
14 r ks(S, 1)
15 end if
16 r S n ks ðS; jSj 1Þ
17 S Snr
18 if |S0 | > 0 then
19 r0 = ks(S0 , 1)
20 S0 S0 /r0
21 S S [ r0
22 end while
Output:
R set of improved routes
END Hi
The smallest route is selected in line 5 using function ks while ‘‘matching’’ routes are selected in line 7 using function kg.
The routes are reconstructed in line 8. If the generalized cost is improved (line 9), the routes with the least generalized cost
are selected. The algorithm continues adding new routes until all routes have had at least one opportunity to be recon-
structed and ‘‘re-optimized’’.
5. Service time improvement
The previous algorithms deal with the minimization of costs via sequencing of customers and their assignment to routes.
The Hyf and Hyb algorithms aim to reduce costs, if total route duration is the key secondary objective, by improving customer
service start times for a given set of routes produced by Hi. This goal has already been analyzed by Desrosiers et al. (1995) in
the context of scheduling traditional VRP fixed routes.
For any given route k, an approach similar to dynamic programming can be used to determine the optimal service start
times yki for customer i belonging to route k given the arrival time ai; each customer is associated with a stage, the decision
variable is the service time yki , and the state is defined by the arrival time ai. For the sake of notational simplicity, let’s denote
a sequence of vertices in a route by their location in the route; hence, any given route k is defined by the sequence of cus-
tomers (0, 1, 2, . . . , q, q + 1) where 0 and q + 1 denote the depot. If the cost to minimize is the sum of distance traveled, route
durations, and soft time window utilization given by Expression (16), the cost function, p(yq, aq), for the last customer is2:
pðyq ; aq Þ ¼ cd dq;qþ1 þ ct ðaqþ1 yq Þ þ ce ðe#q ykq Þþ þ cl ðykq e#l Þþ ð22Þ
where aq+1 = yq + gq + tq,q+1(yq + gq) and subject to lq P yq P aq .
2
The distance term can be eliminated from (22) because it is not affected by service time.
Using a backward solution approach, for each customer it is possible to define a stage cost and an optimal cost to go
function. Further, for each customer, it is possible to limit the feasible space of customer service time to a closed time
interval. For a customer i belonging to route k, let yki and y
ki denote, respectively, the earliest and latest feasible service
times.
Lemma 1. Given any route k, the optimal service times at any customer i belong to the time interval ½yki ; y
ki and can be
calculated using a forward and backward algorithm.
Proof 1. Starting from the depot, earliest possible arrival at customer 1 is a1 = e0 + t01(e0) due to FIFO property; earliest ser-
vice time at customer 1 is yk1 ¼ maxða1 ; e1 Þ; earliest departure at customer 1 is yk1 þ g 1 . Earliest possible arrival at customer 2
is a2 yk1 þ g 1 þ t 1;2 ðyk1 þ g 1 Þ due to FIFO property; earliest service time at customer 2 is yk2 ¼ maxða2 ; e2 Þ; earliest departure at
customer 2 is yk2 þ g 2 and so on until reaching the last customer.On the other hand, starting from the depot, latest possible
departure time last from customer q is: arg maxy2R y; s:t:ðy þ tq;qþ1 ðyÞ 6 lqþ1 Þ; due to the continuous speed function and the
FIFO property this value is unique. The latest possible service time at customer q is y kq ¼ minðy g q ; lq Þ. Latest possible depar-
k
q Þ; latest possible service time at customer q 1 is
ture time from customer q 1 is arg maxy2R y; s:t:ðy þ tq1;q ðyÞ 6 y
kq1 ¼ minðy g q1 ; lq1 Þ and so on until reaching the first customer.
y h
It is possible to state properties that simplify the determination of service start times.
Property 1. Given a route k outputted by Hi, the customer service times are the earliest feasible times.
Proof. Due to the workings of the Hr algorithm, the service at any customer i in route k begins at the earliest feasible time,
which is yki ¼ yki ¼ maxðai ; ei Þ, and the departure time is given by yki þ g i . Due to the FIFO property, for the given routes, cus-
tomers cannot be serviced earlier than the provided service start times. h
Property 2. Given a route k outputted by Hi, total route duration cannot be reduced further.
Proof. Due to Property 1, service times cannot be advanced. Then, the FIFO property guarantees that route duration cannot
be reduced further unless the set of routes is altered. The arrival times at each customer are the earliest possible for the
sequence given by route k. h
Property 3. Given a route k outputted by Hi, a TDVRP with hard time windows requires no service time optimization for route k.
Proof. Due to Property 2, route durations cannot be reduced. Start times do not affect distance traveled and there are no
soft time windows penalties or costs to be reduced. Hence, altering service start time will not reduce any objective
function. h
Property 4. Given a route k outputted by Hi, if a customer uses the ‘‘late’’ soft time window, no improvement can be made by
changing the service time.
Proof. Due to Property 1, the service time cannot be advanced without losing feasibility. If the service time is delayed, there
is a greater late penalty. Hence, if a customer in the route outputted by Hi uses a late time window, the provided service time
for that customer cannot be improved. h
Corollary. In a route outputted by Hi, the service time optimization problem can be decomposed into smaller problems delimited
by customers using ‘‘late’’ soft time windows.
5.1. Service time improvement algorithms
For each customer that uses an early soft time window, the Hyb algorithm attempts to reduce early soft time window
usage without allowing the introduction of service delays that increase late time window usage. This algorithm operates
backwards. For the sake of presentation simplicity, periods of constant travel time are assumed. The depot working time
[e0, l0] is partitioned into p time periods T = T1, T2, . . . , Tp; each period Tk has an associated constant travel speed sk in the time
interval T k ¼ ½tk ; tk .
Data:
T and S: Time intervals and speeds
v i ; v j ; ykj : Two customers served in this order in route k, ykj is the current service time at customer j
START Hyb
1 #
If ykj < lj & ykj < y
kj then
2 ykj
#
kj Þ
minðlj ; y
3 end if
4 find k, tk 6 ykj 6 t k
5 bi ykj dji =sk
6 d dji ; t ykj
7 while bi < tk do
8 d d ðt tk ÞSk
9 t tk
10 bi t d/sk+1
11 k k+1
12 end while
13 ki
y minðbi g i ; li Þ
Output:
ykj ; y
ki
END Hyb
After early time windows have been reduced, a final task is to reduce route duration without increasing the number of
late soft time windows. The following forward algorithm, Hyf, reduces route duration without increasing soft time windows.
Data:
T and S: Time intervals and speeds
vi, vj, yj: Two customers served in this order in route k, yki is the current service time at customer j
START Hyf
1 if yki > e# k k
j & yi > yi then
2 yki maxðe# k
i ; yi Þ
3 end if
4 find k; tk 6 yki 6 t k
5 aj yki þ dij =sk
6 d dij ; t yki
7 while aj > t k do
8 d d ðt k tÞsk
9 t tk
10 aj t + d/sk+1
11 k k+1
12 end while
13 ykj maxðaj ; ej Þ
Output:
yki ; ykj
END Hyf
Both algorithms try to reduce the interval ½yki ; y

ki where the optimal service start time is found for a given route k.
6. Proposed benchmark problems
Barr et al. (1995) presented a detailed discussion of the how to properly test heuristics methods and provide useful test-
ing result. These authors emphasized reproducibility as an ‘‘. . .essential ingredient of scientific research’’ further indicating
that ‘‘experimental results that cannot be independently verified are given little credence in the scientific community’’.
Although this may seem too obvious for many researchers in the physical sciences, it is unfortunately not followed by many
papers dealing with vehicle routing problems. Clear instructions regarding computing environment, testing, quality of the
solution, parameter selection, statistics, variability, analysis and interpretation, and reporting were provided by the authors.
A few years later, Golden et al. (1998) expanded on some of the key problems and emphasized the need for parsimonious,
efficient, robust and simple algorithms indicating that ‘‘In fact, some current heuristics suffer from inelegant design: they have
more parameters than the number of benchmark problems! Furthermore, we suspect that heuristics are being over fit on the
benchmark problems. In the future, the goal should be to design VRP heuristics that are capable of producing high-quality
nearly optimal solutions. Heuristics should be lean and parsimonious and contain few parameters. They should be compu-
tationally efficient, robust, and simple’’. Unfortunately, to the best knowledge of the author, there has not been a change in
the way results are presented and algorithms are tested; simplicity, robustness, and parsimoniousness are not properly dis-
cussed or analyzed in VRP related papers.
As later also indicated by Cordeau et al. (2002), results presented in the VRP literature usually present better results on
benchmark problems at the expense of (a) too many parameters or complicated coding that lacks flexibility to accommodate
real-life constraints, (b) too many parameters that are difficult to calibrate or even understand, and (c) solution approaches
that are markedly tailored to perform well on the benchmark problems but that may lack generality and robustness in real-
life problems. Furthermore, for solutions approaches that are not deterministic because they employ randomization, e.g.
tabu search or GRASP, the testing and presentation of the results must be more demanding and include the variability of
the solution quality and statistical testing instead of simply reporting ‘‘best results’’ or ‘‘averages’’ after the algorithm is fine
tuned (Taillard, 2001; Golden et al., 1998).
As mentioned in Section 2, results provided in previous research efforts cannot be compared in terms of solution quality
or computational time. This is revealing of a still incipient body of work for the TDVRP. The proposed set of benchmark prob-
lems are based on the classical instances of the VRP with time windows proposed by Solomon (1987). The Solomon instances
include distinct spatial customer distributions, vehicles’ capacities, customer demands, and customer time windows. These
problems have not only been widely studied in the operations research literature but the datasets are readily available.3
The well-known 56 Solomon benchmark problems for vehicle routing problems with hard time windows are based on six
groups of problem instances with 100 customers. The six problem classes are named C1, C2, R1, R2, RC1, and RC2. Customer
locations were randomly generated (problem sets R1 and R2), clustered (problem sets C1 and C2), or mixed with randomly
generated and clustered customer locations (problem sets RC1 and RC2). Problem sets R1, C1, and RC1 have a shorter sched-
uling horizon, tighter time windows, and fewer customers per route than problem sets R2, C2, and RC2 respectively.
This section proposes new test problems that capture the typical speed variations of congested urban settings which gen-
erally include slower travel speeds during morning and evening rush hours. The problems are divided into three categories of
study: (1) constant speed Solomon instances, (2) time dependent problems with hard time windows, and (3) time dependent
problems with soft time windows. Some previous research efforts may have used standard problems but they allocated tra-
vel speed distributions randomly to customer arcs or it is ambiguous the type of time dependency allocated to each arc. In
order to provide readily replicable instances, the travel speed distributions apply to ALL arcs among customers, i.e. in the arc
set:
A ¼ fðv i ; v j Þ : i–j ^ i; j 2 Vg:
Most recent research efforts, as stated in Section 2, have used constant speed intervals. The same approach is adopted in
this research because constant speed intervals guarantee the FIFO property and can be readily replicated. The algorithm used
to calculate travel times is presented in Appendix A.
6.1. Constant speed problems with hard time windows
Constant travel speed is a special case of the general time dependent problem. These instances are the classical Solomon
problems that have been widely studied and provide an indication of the performance of the algorithm with constant travel
speed.
6.2. Time dependent problems with hard time windows
There could be many potential speed distributions. The goal is to present meaningful speed distributions while minimiz-
ing as much as possible the number of speed distributions and test problems. Solomon’s instances represent the archetypical
and ubiquitous case where a central depot services a set of surrounding customers with delivery time windows during the
depot open time. Assuming a typical urban area with morning and evening congested periods and also assuming that the
depot open time represents the duration of the vehicles/drivers workday, four distinct cases are proposed and interpreted
as follows:
(a) The trucks leave the depot very early in the day and can travel without congestion during the first 1/5 of the workday
before morning congestion takes place. The following 1/5 of the workday coincides with morning congestion followed
by the somewhat less congested midday period. Another 1/5 of the workday coincides with evening congestion and
the last 1/5 of the workday takes place after congestion has subsided.
3
Several websites maintain downloadable datasets of the instances including Solomon’s own website: http://web.cba.neu.edu/~msolomon/problems.htm.
(b) The opposite of the case (a); the trucks leave the depot just as morning congestion starts for 1/5 of the workday, the
following 1/5 of the workday coincides with no congestion followed by the somewhat congested midday period.
Another 1/5 of the workday takes place before evening congestion and the last 1/5 of the workday coincides with
the heavy evening congestion.
(c) The trucks leave the depot early in the day so that the first 1/2 of the workday is uncongested followed by the second
1/2 workday that takes place during the heaviest morning congested period.
(d) Due to time window constraints, the trucks leave the depot just as morning congestion begins; hence, the initial 1/2
workday takes place under heavy morning congestion and the second 1/2 of the workday takes place after congestion
has subsided.
Because the Solomon problems are tightly constrained, i.e. a small decrease in travel speed results in infeasible problems
due to the hard time windows, the only approach to leverage the existing Solomon instances to evaluate time dependent
problems is to introduce speed variability with higher travel speeds (the speed in the original Solomon problems is constant
and equal to one). The maximum ratio in travel speeds is 2.5–1. This ratio is chosen because it is realistic to assume that in an
urban freeway the free-flow or uncongested travel speed is around 60 miles per hour and with heavy congestion the speed
drops to 24 miles per hour. A time dependent problem is generated by assigning the speed distribution to all arcs to facilitate
reproducibility.
Type (a)
These instances introduce fast periods between depot opening and closing times. The depot working time [e0, l0] is di-
vided into five time periods of equal durations:
[0, 0.2l0); [0.2l0, 0.4l0); [0.4l0, 0.6l0); [0.6l0, 0.8l0); and [0.8l0, l0].
and the corresponding travel speeds are:
TD1a = [1.00, 1.60, 1.05, 1.60, 1.00],

TD2a = [1.00, 2.00, 1.50, 2.00, 1.00],
TD3a = [1.00, 2.50, 1.75, 2.50, 1.00].
Type (b)
The opposite to type (a), higher travel speeds are found at the extremes of the working day.
TD1b = [1.60, 1.00, 1.05, 1.00, 1.60],

TD2b = [2.00, 1.00, 1.50, 1.00, 2.00],
TD3b = [2.50, 1.00, 1.75, 1.00, 2.50].
Type (c)
Higher travel speeds are found at the beginning of the working day.
TD1c = [1.60, 1.60, 1.05, 1.00, 1.00],

TD2c = [2.00, 2.00, 1.50, 1.00, 1.00],
TD3c = [2.50, 2.50, 1.75, 1.00, 1.00].
Type (d)
The opposite to type (b), higher travel speeds are found at the end of the working day.
TD1d = [1.00, 1.00, 1.05, 1.60, 1.60],

TD2d = [1.00, 1.00, 1.50, 2.00, 2.00],
TD3d = [1.00, 1.00, 1.75, 2.50, 2.50].
In all cases, types (a)–(d), if the vehicles were to travel non-stop in the interval [e0, l0] the vehicle would travel an extra
25%, 50%, and 75% more for speeds TD1, TD2, and TD3 respectively than in the original Solomon instances.
6.3. Time dependent problems with soft time windows
These instances introduce two congested periods between depot opening and closing times. The depot working time is
divided into the same five periods and the corresponding travel speeds are:
TD4 = [1.10, 0.85, 1.10, 0.85, 1.10],

TD5 = [1.20, 0.80, 1.00, 0.80, 1.20],
TD6 = [1.20, 0.70, 1.20, 0.70, 1.20].
If one vehicle were to travel non-stop in the interval [e0, l0], this vehicle would travel the same distance as in the original
Solomon instances but with increasing travel speed variability, i.e. same average speed but with increased variability. How-
ever, soft time windows are required because some Solomon problems would be infeasible otherwise (Ichoua et al., 2003;
#
Donati et al., 2008). An allowable time window violation per customer equal to: P max ¼ 0:1ðl0 e0 Þ ¼ e# i ei ¼ li li is al-
lowed. However, the depot working time [e0, l0] is not relaxed. The penalty cost for an early or late delivery is one unit of
cost per unit time which is the same value used in constant speed Solomon instances with soft time windows (Balakrishnan,
1993; Chiang and Russell, 2004).
6.4. Time windows elasticity
As stated previously, Barr et al. (1995) and Golden et al. (1998) indicated that it was necessary to test in such a way that
‘‘overfitting’’ is avoided. Algorithmic parameters are always fine tuned to the benchmark instances but researchers do not
publish the time needed to properly fine tune the parameters (Golden et al., 1998).
To better understand the response of the algorithm by problem type, a small perturbation is introduced. Given the impor-
tance of time windows, a small perturbation is introduced as follows:
ei ei þ uniform½1:0; 1:0 0:02 ðl0 e0 Þ; 8i 2 C
The magnitude of the perturbation cannot exceed 2% of the depot working time and the perturbation is randomly applied to
each customer. The expectation of the sum of the perturbations equals zero or ‘‘neutral’’ since the expectation of the pertur-
bations is:
Eðuniform½1:0; 1:0Þ ¼ 0
However, the expectation of the sum of the absolute perturbations is 0.01 (l0 e0) or 1% of the depot working time; i.e.
conditional on being positive (or negative), the perturbations are on average 1% (1%) of the depot working time.
There are two reasons for choosing absolute perturbations that are 1% on average:
(a) conceptual: a elasticity is conceptually described in the economic literature as the % change in the dependent variable
when the independent variables is changed 1%, hence, the results of the perturbation analysis can be interpreted as the
elasticity of the solution quality to a 1% change in the window start time, and
(b) practical: some problems or instances are so tight that they become infeasible if time window perturbations are larger
than 1% on average.
7. Experimental results
Proper benchmarking of algorithms, solution quality and computation times can be performed using standardized in-
stances and computers. However, computation times can be difficult to compare if there are significant differences in com-
puter processing power or equipment. Detailed information regarding computer equipment (brand, model, processor, RAM)
can be used to estimate relative computer power using Dongarra (2007) and SPEC4 results. All the results presented in this
section were obtained with a laptop Dell Latitude D430, with an Intel Core CPU 1.2 GHz and 1.99 GB of RAM. Even after stan-
dardizing problems and equipment there may be differences in running time due to different compilers, programming language,
or code efficiency and implementation.
Golden et al. (1998) indicates that algorithms should be compared not only by the number of parameters but also by how
intuitive and reasonable these parameters are from a user’s perspective. To avoid excessive ‘‘tailoring’’, all the results pre-
sented in this research use the exact same procedure and parameter values in all cases, i.e. the same code, with the param-
eters described in Section 4, and the same parameter values. The exact same parameters are used not only in different types
of problems, e.g. soft vs. hard, but also in different types of instances, e.g. R1 and C2. Travel time calculations were performed
using the algorithm presented in Appendix A. It is also assumed that the algorithm does not ‘‘know’’ anything regarding the
type of problem or its characteristics, e.g. average number of customers per route, binding constraints, or lower bounds. This
type of information can be exploited to reduce computational times, e.g. usage of lower bounds (Figliozzi, 2008), but if new
parameters, steps, or lines of code are needed they have to be explicitly stated to provide a level playing field when it comes
to comparisons among algorithms. The hard time window results provided in this paper are found using a search space
bounded by:
0:00 6 d1 6 0:65; 0:25 6 d2 6 1:00; 0:00 6 d3 6 0:20; 0:00 6 d4 6 0:03
and d1 + d2 + d3 = 1; d0 is a sufficiently large number (primary objective is to minimize fleet size).
4
Comparison among computers can be found at http://www.specbench.org/.
7.1. Constant speed problems with soft time windows
The first set of results corresponds to the extensively studied Solomon instances with constant travel speeds; results are pre-
sented in Table 1. In these instances the primary objective is to minimize the number of vehicles and the secondary objective to
minimize travel distance. The first row presents the combination of the absolute best solutions found to date which have been
obtained by different researchers, algorithms, machines, and computational times (Donati et al., 2008; SINTEF, 2008).
The second row presents the results of Taillard et al. (1997) using the tabu search algorithm for soft time window prob-
lems and constant travel speed that was implemented by Ichoua et al. (2003). Taillard et al. reported better results but at the
expense of significantly longer computational times. The results reported by Taillard et al. and Donati et al. are average re-
sults and computation times over independent runs.
The performance of the IRCI algorithm, in relation to other approaches that can solve problems with both soft and hard time
windows have been used in time dependent problems, is somewhat comparable. The IRCI solutions have relatively low com-
putational times – an average of 21.3 s for each 100 customer problem but comparisons in terms of speed with Taillard et al.
(1997) are difficult. Computers and their architecture have evolved significantly in the last 10 years. However, the IRCI is faster
than the method presented by Donati et al. (2008). In terms of solution quality, the IRCI is outperformed by the best local search
approaches (Braysy and Gendreau, 2005b). The IRCI solutions are, on average, slightly less than 4% from the best results ever
obtained for the Solomon instances with constant travel times. The IRCI can obtain slightly better performances, around 3%,
in terms of number of vehicles with longer computational times or by tailoring some parameters to each problem type. How-
ever, to avoid any kind of ‘‘distortion’’, the same general code is utilized to obtain all the results presented in this section.
7.2. Time dependent problems with hard time windows
The second set of results corresponds to the Solomon instances with time dependent travel speeds and hard time win-
dows. In these instances the primary objective is to minimize the number of vehicles, the secondary objective is to minimize
time and distance traveled.
To the best of the author’s knowledge this is the first reporting of Solomon instances with hard time windows and time
dependent speeds; results are presented in Tables 2–5. As expected, with increased travel speeds, the number of vehicles is
reduced significantly. However, there is relatively minimal change in the distance traveled. Time traveled decreases as aver-
Table 1
VRPTW results for classical Solomon instances – constant speed.
Method R1 R2 C1 C2 RC1 RC2

Average number of vehicles by problem class
(1) Best Ever (1987–. . .) 11.92 2.73 10.00 3.00 11.50 3.25
(2) Taillard et al. (1997) 12.64 3.00 10.00 3.00 12.08 3.38
(3) Donati et al. (2008) 12.61 3.09 10.00 3.00 12.04 3.38
(4) IRCI 12.58 3.00 10.00 3.00 12.12 3.38
Average distance
(1) Best Ever (1987–. . .) 1210 952 828 590 1384 1119
(2) Taillard et al. (1997) 1220 1013 828 591 1381 1199
(2) Donati et al. (2008) 1199 967 828 590 1374 1156
(4) IRCI 1248 1124 841 626 1466 1308
Computation time for all 56 problems: (1) different authors, machines and computation times; (2) Sun Sparc 10, 261 min; (3) Pentium IV 2.66 GHz, 168 min;
(4) Dell Latitude D430, 1.2 GHz, 19.0 min.
Table 2
VRPTW results – hard time windows – type (a).
Travel time distribution R1 R2 C1 C2 RC1 RC2

(1) TD1 11.67 2.82 10.00 3.00 11.38 3.25
(2) TD2 10.75 2.55 10.00 3.00 10.50 2.88
(3) TD3 9.92 2.27 10.00 3.00 10.00 2.75
Average distance
(1) TD1 1295 1216 879 657 1405 1444
(2) TD2 1258 1244 864 654 1395 1454
(3) TD3 1237 1269 880 697 1362 1434
Average travel time
(1) TD1 1080 990 729 563 1164 1177
(2) TD2 897 861 644 495 989 993
(3) TD3 793 774 608 485 860 867
Computation time for all 56 problems: (1) TD1, 19.1 min; (2) TD2, 17.7 min; (3) TD3, 17.3 min – in all cases using Dell Latitude D430, 1.2 GHz.
Table 3
VRPTW results – hard time windows – type (b).

(1) TD1 12.42 3.00 10.00 3.00 12.13 3.38
(2) TD2 11.50 2.73 10.00 3.00 11.25 3.25
(3) TD3 11.42 2.73 10.00 3.00 11.00 3.00
Average distance
(1) TD1 1289 1212 892 670 1454 1403
(2) TD2 1279 1218 883 667 1429 1433
(3) TD3 1265 1245 866 714 1442 1483
Average travel time
(1) TD1 1064 1027 732 545 1180 1200
(2) TD2 905 893 650 467 1010 1053
(3) TD3 808 831 584 446 916 981
Table 4
VRPTW results – hard time windows –type (c).

(1) TD4 11.67 2.73 10.00 3.00 11.50 3.25
(2) TD5 10.83 2.55 10.00 3.00 10.75 2.75
(3) TD6 10.17 2.36 10.00 3.00 10.13 2.75
Average distance
(1) TD4 1302 1245 865 683 1435 1407
(2) TD5 1266 1238 863 658 1413 1472
(3) TD6 1272 1243 862 665 1409 1438
Average travel time
(1) TD4 1066 1003 697 573 1186 1147
(2) TD5 881 843 618 483 1012 1027
(3) TD6 801 760 565 451 904 886
Table 5
VRPTW results – hard time windows – type (d).

(1) TD4 12.25 3.00 10.00 3.00 12.00 3.38
(2) TD5 11.58 2.73 10.00 3.00 11.25 3.25
(3) TD6 11.08 2.64 10.00 3.00 10.75 3.25
Average distance
(1) TD4 1311 1218 872 666 1425 1394
(2) TD5 1272 1216 856 679 1404 1412
(3) TD6 1293 1215 867 690 1436 1424
Average travel time
(1) TD4 1114 1045 731 552 1192 1192
(2) TD5 943 915 652 494 1035 1053
(3) TD6 871 846 612 461 964 975
age travel speed increases though not at the same rate. Results for problem sets C1 and C2 are largely unchanged due to the
binding constraint of the vehicle capacity. In the four types, the computation time for TD2 and TD3 is less than TD1 because
when vehicles travel faster the number of time periods utilized (on average) to travel between customers is reduced.
Although in all cases, types (a)–(d), if the vehicles were to travel non-stop in the interval [e0, l0] the vehicles would travel
the same distance (extra 25%, 50%, and 75% more for speeds TD1, TD2, and TD3 respectively than in the original Solomon
instances), the results in terms of fleet size are significantly different. Types (a) and (c) have significantly less vehicles than
types (b) and (d).
Table 6
Distribution of time windows.
Fraction depot working time 0.0–0.2 0.2–0.4 0.4–0.6 0.6–0.8 0.8–1.0

Overlap counts between time windows and time periods
R101 15 41 37 25 7
R201 26 48 33 29 18
C101 30 33 35 17 3
C201 24 27 27 27 16
RC101 17 60 53 31 5
RC201 26 50 39 29 16
To explain the differences in fleet size it necessary to understand how time windows are distributed in the Solomon prob-
lems. Table 6 presents the distribution of time window start, end times, and overlaps for the first instance in each problem
type. Time window starts and ends are more frequent in the second and third period and less frequent at the extremes of the
depot working time. Hence, speed distributions with the fasted periods at the beginning and end of the depot working time,
type (b), or in the second half the working time, type (d), are less efficient that types (a) or (c) respectively.
7.3. Time dependent problems with soft time windows
The third set of results corresponds to the Solomon instances with time dependent travel speeds and soft time windows;
results are presented in Table 7. In these instances the primary objective is to minimize number of vehicles, the secondary
objective is to minimize time window violations, and the tertiary objective is to minimize the soft time window penalties
and distance traveled. Table 7 presents the results in terms of fleet size, distance, and travel time.
The travel speed distributions TD4, TD5, and TD6 are listed in increasing order of travel speed variability. Without chang-
ing overall average speed, travel speed variability worsens the results in terms of number of vehicles for R1 and RC1 prob-
lems. Results in terms of distance traveled have little variation. Travel time slightly increases. Problem sets C1 and C2 are
mostly unchanged because the binding constraint is vehicle capacity.
Although in all cases, types (a)–(d), if the vehicles were to travel non-stop in the interval [e0, l0] the vehicles would travel
the same distance (extra 25%, 50%, and 75% more for speeds TD1, TD2, and TD3 respectively than in the original Solomon
instances), the results in terms of fleet size are significantly different. Types (a) and (c) have significantly less vehicles than
types (b) and (d).
As customary in the VRP with time windows literature. Table 8 reports the number of soft time windows used, broken
down into early and late service times as well as the penalty paid for early or late services. Usage of early soft time windows
is more prevalent than the usage of late time windows. As expected, time window violations and penalties decrease as the
number of vehicles used increases.
7.4. Time window elasticity
To better understand the response of the algorithm by problem type, a small perturbation is introduced. Given the impor-
tance of time windows, a small perturbation is introduced as follows:
ei ei þ uniform½1:0; 1:0 0:02 ðl0 e0 Þ; 8i 2 C
The perturbation is applied to problems TD1 type (a). The time window duration is not changed; hence, the difference (li ei)
is maintained for all customers as well as all the other customer characteristics such as demand level and location. It was
Table 7
VRPTW results – soft time windows.

(1) TD4 10.42 2.82 10.00 3.00 10.50 3.00
(2) TD5 10.42 2.64 10.00 3.00 10.63 3.00
(3) TD6 10.58 2.73 10.00 3.00 10.75 3.00
Average distance
(1) TD4 1142 1010 856 666 1241 1135
(2) TD5 1131 1016 860 665 1226 1156
(3) TD6 1127 1016 869 660 1236 1149
Average travel time
(1) TD4 1139 1023 871 669 1237 1150
(2) TD5 1134 1039 884 672 1220 1184
(3) TD6 1143 1061 938 685 1253 1213
Table 8
VRPTW results – soft time windows.

Average number of soft time windows (early)
(1) TD4 20.5 20.1 15.8 18.6 21.1 21.3
(2) TD5 20.4 19.9 18.6 14.9 22.1 21.1
(3) TD6 20.4 20.1 16.1 13.9 21.3 21.5
Average number of soft time windows (late)
(1) TD4 18.0 13.6 8.2 17.0 16.3 14.1
(2) TD5 17.5 12.5 8.7 10.4 14.5 14.8
(3) TD6 15.8 12.5 6.2 8.4 15.1 15.1
Soft time window penalties (early)
(1) TD4 386.6 1516.3 516.5 3025.5 381.7 1718.9
(2) TD5 425.1 1609.4 861.5 1467.7 448.5 1664.4
(3) TD6 419.3 1559.1 657.2 1697.8 446.2 1508.2
Soft time window penalties (late)
(1) TD4 210.4 681.2 480.5 3267.1 197.8 695.9
(2) TD5 208.2 637.6 547.6 1797.7 173.2 692.1
(3) TD6 187.6 629.5 363.2 1708.8 189.2 787.8
mentioned in the previous section that the sum of perturbations are expected to be in absolute value equal to 1% of the depot
working time and the expectation of the sum of all the perturbations is expected to be zero or ‘‘neutral’’ on average. To exem-
plify the magnitude of the change, for a 5-h depot working time a 1% change is equal to a 3 min change.
As an example, six different random seeds are employed to draw real numbers from the uniform[1.0, 1.0] distribution;
clearly, more seeds can be used but it is essential that the researcher publishes the results of all the seeds to avoid bias. The
results, presented in Table 9, show that a small perturbation can have a significant impact even on the average number of
Table 9
Solution quality elasticity – TD1 type (a).
R1 R2 RC1 RC2
Average number of vehicles
Random seed 1 11.67 2.73 11.38 3.25
Random seed 2 11.58 2.82 11.38 3.25
Random seed 3 11.50 2.82 11.38 3.25
Random seed 4 11.92 2.91 11.25 3.25
Random seed 5 11.50 2.73 11.25 3.25
Random seed 6 11.58 2.82 11.38 3.25
Average distance
Random seed 1 1261 1218 1410 1383
Random seed 2 1299 1225 1413 1427
Random seed 3 1285 1202 1424 1411
Random seed 4 1271 1196 1395 1425
Random seed 5 1276 1240 1411 1412
Random seed 6 1264 1227 1406 1435
Average travel time
Random seed 1 1045 992 1161 1112
Random seed 2 1079 998 1170 1148
Random seed 3 1069 977 1171 1142
Random seed 4 1059 977 1160 1154
Random seed 5 1061 1003 1170 1148
Random seed 6 1052 999 1164 1167
VRPTW results – hard time windows – type (a)
TD1 – unperturbed 11.67 2.82 10.00 3.00 11.38 3.25
Average distance
TD1– unperturbed 1295 1216 879 657 1405 1444
Average travel time
TD1– unperturbed 1080 990 729 563 1164 1177
Table 10
Key statistics number of vehicles.
Statistic R1 R2 RC1 RC2

Min 11.50 2.73 11.25 3.25
Max 11.92 2.91 11.38 3.25
Average 11.63 2.80 11.33 3.25
Mean 11.58 2.82 11.38 3.25
St. dev. 0.16 0.07 0.06 0.00
Coef. of variation 1.3% 2.4% 0.6% 0.0%
vehicles for problem type R1. The impact or magnitude of the changes is less for problems RC1 and R2 and problems RC2 are
not affected on average. The comparable unperturbed results from Table 2 are reproduced at the bottom to facilitate the com-
parison. Problems C1 and C2 are not included because in these problems fleet size is not sensitive to changes in travel speed
or time windows.
The results obtained in Table 9 are obtained with exactly the same algorithm, parameters, and running time employed to
obtain the results presented in Table 2. For problem types R1, R2, and RC1 small perturbations can lead to outcomes that are,
on average, roughly ±1% to 3% in terms of fleet size. Unfortunately, these values cannot be compared to other solution algo-
rithms because, to the best of the author’s knowledge, this type of robustness or elasticity analysis has never been performed
for VRP problems with time windows or time dependent travel times. Clearly, performance on individual instances is impor-
tant to measure the overall quality of the algorithm. But, as expressed by Barr et al. (1995) and Golden et al. (1998) the gain
in solution quality should not come at the expense of over fitting the solution approach to the known instance since in real-
world applications instances do change regularly or extremely often in real-time applications. A balanced approach to test-
ing VRP algorithms should include absolute solution quality as well as elasticity to changes in small data inputs.
Table 10 shows some descriptive statistics including minimum, maximum, median, average, standard deviation, and coef-
ficient of variation (CV). Although problems R1 show the highest absolute change, percentage wise the problems R2 show the
greater change in terms of CV. Only time window results are shown since small perturbations to customer demand did not
produce any change in average fleet size (zero elasticity). The elasticity to perturbations is also important when comparing
algorithms and it will facilitate a better understanding of VRP algorithms. For example, a small difference in average RC2
performance can be statistically significant whereas the same absolute different for R1 problems may not be statistically
significant.
Some researchers have already suggested the use of statistical tests to compare algorithms. For example, Taillard
(Taillard, 2001; Golden et al., 1998) suggested the use of the non-parametric Mann–Whitney test to estimate whether the
distribution of the results produced by each algorithm are of comparable quality. For paired results, e.g. same instances,
the Wilcoxon test is recommended. To have a fair comparison, the results to be compared must have similar running times
and in similar computers. Ideally, the algorithms would be tested on a meaningful range of running times (short and long).
8. Computational complexity
The relative simplicity of the IRCI allows for a straightforward algorithmic analysis. The auxiliary heuristic Hr is called by
the construction algorithm no more than nW|D| times; where n is the number of customers. Hence, the worst case number of
operations of the construction algorithm is of order (nW|D|O(Hr(n))) where O(Hr(n)) denotes the computational complexity
of the auxiliary algorithm to route n customers. Hence, the complexity and running time of the auxiliary heuristic Hr will
have a substantial impact on the overall running time.
The improvement procedure calls the construction procedure a finite number of times. The number of calls is bounded by
the number of routes |R| = m. Let ni be largest number of customers contained a subset of routes u that is improved in each
iteration of Hi. The computational complexity of a call to the construction algorithm is then (niW|D|O(Hr(ni))). The complex-
ity of the Hi algorithm is then of order O(mniW|D|O(Hr(ni))) where ni < n if u < m.
If constant speed intervals are used to represent time dependent speeds and the depot working time [e0, l0] is partitioned
into p time periods, the computational complexity of the service start time algorithms, Hyb and Hyf is of order O(np). Each
travel time calculation between any two customers has a computational complexity O(p) – see Appendix A.
Table 11
VRPTW average run time ratios – TD3.
n Ratio O(n2) O(n3) % O(n3)

25 1.0 1 1 100
50 3.3 4 8 41
100 17.4 16 64 27
200 90.5 64 512 18
To test the increase in computational running time, instances with different numbers of customers are run. Firstly, the
first 25 and 50 customers of each Solomon problem are taken to create instances with n = 25 and O(np) = 50 respectively.
Secondly, to create and instance with n = 200 customers, for each customer in the original Solomon problem a ‘‘clone’’ is cre-
ated. Each ‘‘clone’’ has the same customer characteristics of an ‘‘original’’ customer but multiplied by a random number
drawn from [0.95, 1.05] to avoid exact replicas. Hence, the clones are ‘‘similar’’ but not ‘‘identical’’.
The summary results for each problem size are shown in Table 11. The results are expressed as the ratio between each
average running time and the running time for n = 25. To facilitate comparisons, the corresponding increases in running time
ratios for O(n2) and O(n3) are also presented.
The results indicate that the average running time is increasing by a factor of O(n2). This is expected from the complexity
analysis as the complexity of the nearest neighbor heuristic Hr has a worst case of O(n2). As customer size n increases, the
ratio as a % of the n3 growth factor is decreasing – see last column of Table 11.
8.1. Travel time updates
Current location and communication systems allow the utilization of real-time information about vehicle locations and
demands to update routes whenever there are significant changes in travel times (Regan et al., 1996). For example, delays
due to non-recurrent events such as accidents can significantly alter travel times between customers. In these situations fast
algorithms may have a significant edge over significantly slower though higher quality approaches. The advantage of fast
algorithms is more significant when there are time windows or significant differences among customers’ time and price sen-
sitivity (Kim et al., 2004).
The route improvement algorithm proposed in this research can be easily extended to situations where a subset of arc
travel times is updated as follows:
Algorithm Hu
1. Find the subset of T = T1, T2, . . . , Tp time periods that are affected by the update.
2. Find the set of tij(bi) values that has been updated.
3. Select the subset of customers C(updated) that are affected by changes in travel times.
4. Form a subset of routes R(updated) that contains all customers in C(updated).
5. Apply the route improvement algorithm Hi to each route contained in R(updated).
The running time of the update algorithm Hu will be a function of the number of customers and routes affected by the
new travel times: O(mniW|D|O(Hr(ni))) where ni < n if u < m. An objective function with soft time windows guarantees that
a solution will be found after the first run of Hu. The evaluation of real-time strategies is beyond the scope of this research.
The interested reader is referred to research in real-time environments, e.g. Haghani and Jung (2005).
9. Conclusions
This is the first research effort to publish solutions to time dependent problems with hard time windows using standard
and replicable instances that simulate variations in travel times that can be found in congested urban areas. The results show
that the overlap between the temporal distribution of the congested periods and the distribution of customer time windows
has a significant impact on routing costs. Results also indicate that a constant travel speed assumption in congested urban
areas with customer hard time windows produce suboptimal or unfeasible solutions. In addition, a method to study the im-
pact of perturbations by problem type is introduced and results showed that problem types have significantly different levels
of time window elasticity to small perturbations in the data inputs.
The proposed solution algorithm can tackle constant speed or time-dependent speed problems with hard or soft time
windows without any alteration in their structure. This is the first research effort to published results that can be readily
benchmarked for TDVRP with hard time windows and it provides faster and higher quality results than already published
methods to solve the TDVRP with hard time windows in constant speed Solomon instances. The analysis and experimental
results of the computational complexity indicate that average computational time increases proportionally to the square of
the number of customers. The relative low computational complexity, simplicity, and generality of the IRCI are important
factors in real-world applications with constant and time dependent travel times. Future research efforts may explore alter-
native grouping methodologies, alternative route construction approaches, and real-time implementations.
Acknowledgements
The author gratefully acknowledges the Oregon Transportation, Research and Education Consortium (OTREC) and the
Department of Civil & Environmental Engineering in the Maseeh College of Engineering & Computer Science at Portland State
University for sponsoring this research. The author would like to thank Stuart Bain, at the University of Sydney, for his assis-
tance coding during the initial stages of this research and Myeonwoo Lim, Computer Science Department at Portland State
University, for assistance coding during the final stages of this research. The author gratefully acknowledges Brian Davis and
anonymous reviewers for their useful suggestions and comments. Any errors are the sole responsibility of the author.
Appendix A
Unlike the algorithms presented in Section 4, the calculation of travel times is dependent on the specific data format and
speed functions. Travel times from any two given customers i and j are calculated using an iterative forward calculation from
the arrival time at customer i. The depot working time [e0, l0] is partitioned into p time periods T = T1, T2, . . . , Tp; each period
Tk has an associated constant travel speed sk. The algorithm is adapted from Ichoua et al. (2003):
Data:
T = T1, T2, ..., Tp, and corresponding travel speeds
vi, vj, ai: Given any two customers and the arrival time to customer i
START
if ai < ei then
bi ei + gi
else bi ai + gi
end if
find k, t k 6 bi 6 tk
aj bi + dij/sk
d dij, t bi
while aj > tk do
d d ðtk tÞsk
t t k
aj t + d/sk+1
k k+1
end while
Output:
aj, arrival time at customer j
END Hr
The algorithm is guaranteed to find the arrival time in no more than p iterations.
References
Ahn, B.H., Shin, J.Y., 1991. Vehicle-routing with time windows and time-varying congestion. Journal of the Operational Research Society 42 (5), 393–400.
Balakrishnan, N., 1993. Simple heuristics for the vehicle routeing problem with soft time windows. The Journal of the Operational Research Society 44 (3),
279–287.
Barr, R.S., Golden, B.L., Kelly, J.P., Resende, M.G.C., Stewart, W.R., 1995. Designing and reporting on computational experiments with heuristic methods.
Journal of Heuristics 1 (1), 9–32.
Braysy, I., Gendreau, M., 2005a. Vehicle routing problem with time windows, part 1: route construction and local search algorithms. Transportation Science
39 (1), 104–118.
Braysy, I., Gendreau, M., 2005b. Vehicle routing problem with time windows, part II: metaheuristics. Transportation Science 39 (1), 119–139.
Chiang, W.C., Russell, R.A., 2004. A metaheuristic for the vehicle-routeing problem with soft time windows. Journal of the Operational Research Society 55
(12), 1298–1310.
Cordeau, J.F., Gendreau, M., Laporte, G., Potvin, J.Y., Semet, F., 2002. A guide to vehicle routing heuristics. Journal of the Operational Research Society 53 (5),
512–522.
Dabia, S., Van Woensel, T., De Kok, A.G., 2010. A Dynamic Programming Approach to Multi-Objective Time-Dependent Capacitated Single Vehicle Routing
Problems with Time Windows. Working Paper 313, Eindhoven University of Technology, School of Industrial Engineering, The Netherlands.
Desrochers, M., Lenstra, J.K., Savelsbergh, M.W.P., Soumis, F., 1988. Vehicle routing with time windows: optimization and approximation. Vehicle Routing:
Methods and Studies 16, 65–84.
Desrosiers, J., Dumas, Y., Solomon, M.M., Soumis, F., 1995. Time constrained routing and scheduling. In: Ball, M., Magnant, T., Monma, C., Nemhauser, G.
(Eds.), Handbooks in Operations Research and Management Science Network Routing, vol. 8. North-Holland.
Donati, A.V., Montemanni, R., Casagrande, N., Rizzoli, A.E., Gambardella, L.M., 2008. Time dependent vehicle routing problem with a multi ant colony system.
European Journal of Operational Research 185 (3), 1174–1191.
Dongarra, J.J., 2007. Performance of Various Computers Using Standard Linear Equations Software. Technical Report CS-89-85, University of Tennessee,
November 20, 2007. <http://www.netlib.org/utk/people/JackDongarra/papers.htm> (accessed 23.11.07).
Figliozzi, M., 2008. An iterative route construction and improvement algorithm for the vehicle routing problem with soft and hard time windows. In:
Applications of Advanced Technologies in Transportation (AATT) 2008 Conference Proceedings, Athens, Greece, May, 2008.
Figliozzi, M.A., 2009. A route improvement algorithm for the vehicle routing problem with time dependent travel times. In: Proceeding of the 88th
Transportation Research Board Annual Meeting, Washington, DC, January, 2009. <http://web.cecs.pdx.edu/’maf/publications.html>.
Figliozzi, M.A., 2010a. The impact of congestion on commercial vehicle tours and costs. Transportation Research Part E: Logistics and Transportation 46E (4),
496–506.
Figliozzi, M.A., 2010b. An iterative route construction and improvement algorithm for the vehicle routing problem with soft-time windows. Transportation
Research Part C: Emerging Technologies 18 (5), 668–679.
Fleischmann, B., Gietz, M., Gnutzmann, S., 2004. Time-varying travel times in vehicle routing. Transportation Science 38 (2), 160–173.
Golden, B., Wasil, E., Kelly, J., Chao, I., 1998. Metaheuristics in vehicle routing. In: Craignic, T., Laporte, G. (Eds.), Fleet Management and Logistics. Kluwer,
Boston.
Haghani, A., Jung, S., 2005. A dynamic vehicle routing problem with time-dependent travel times. Computers and Operations Research 32 (11), 2959–2986.
Hill, A.V., Benton, W.C., 1992. Modeling intra-city time-dependent travel speeds for vehicle scheduling problems. Journal of the Operational Research
Society 43 (4), 343–351.
Ichoua, S., Gendreau, M., Potvin, J.Y., 2003. Vehicle dispatching with time-dependent travel times. European Journal of Operational Research 144 (2), 379–
396.
Jung, S., Haghani, A., 2001. Genetic algorithm for the time-dependent vehicle routing problem. Transportation Research Record 1771 (-1), 164–171.
Kim, Y.J., Mahmassani, H.S., Jaillet, P., 2004. Dynamic truckload routing, scheduling, and load acceptance for large fleet operation with priority demands.
Transportation Research Record 1882, 120–128.
Kok, A.L., 2010. Congestion Avoidance and Break Scheduling within Vehicle Routing. Ph.D. Thesis, University of Twente, Enschede, The Netherlands.
Malandraki, C., 1989. Time Dependent Vehicle Routing Problems: Formulations, Solution Algorithms and Computational Experiments. Ph.D. Dissertation,
Northwestern University, Evanston, Illinois.
Malandraki, C., Daskin, M.S., 1992. Time-dependent vehicle-routing problems – formulations, properties and heuristic algorithms. Transportation Science
26 (3), 185–200.
Malandraki, C., Dial, R.B., 1996. A restricted dynamic programming heuristic algorithm for the time dependent traveling salesman problem. European
Journal of Operational Research 90 (1), 45–55.
Regan, A., Mahmassani, H., Jaillet, P., 1996. Dynamic decision making for commercial fleet operations using real-time information. Transportation Research
Record 1537, 91–97.
Schrimpf, G., Schneider, J., Stamm-Wilbrandt, H., Dueck, G., 2000. Record breaking optimization results using the ruin and recreate principle. Journal of
Computational Physics 159 (2), 139–171.
SINTEF, 2008. Benchmarks – Vehicle Routing and Travelling Salesperson Problems. SINTEF Applied Mathematics, Department of Optimization, Norway,
<http://www.top.sintef.no/vrp/benchmarks.html>.
Soler, D., Albiach, J., Martínez, E., 2009. A way to optimally solve a time-dependent vehicle routing problem with time windows. Operations Research Letters
37 (1), 37–42.
Solomon, M.M., 1987. Algorithms for the vehicle-routing and scheduling problems with time window constraints. Operations Research 35 (2), 254–265.
Taillard, E., Badeau, P., Gendreau, M., Guertin, F., Potvin, J.Y., 1997. A tabu search heuristic for the vehicle routing problem with soft time windows.
Transportation Science 31 (2), 170–186.
Taillard, E.D., 2001. Comparison of Non-Deterministic Iterative Methods, Citeseer.
Van Woensel, T., Kerbache, L., Peremans, H., Vandaele, N., 2008. Vehicle routing with dynamic travel times: a queueing approach. European Journal of
Operational Research 186 (3), 990–1007.

1 s2.0 S1366554511001426 Main

Uploaded by

Copyright:

Available Formats

1 s2.0 S1366554511001426 Main

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

1 s2.0 S1366554511001426 Main

Uploaded by

Copyright:

Available Formats

Transportation Research Part E 48 (2012) 616–636

Contents lists available at SciVerse ScienceDirect

Transportation Research Part E

The time dependent vehicle routing problem with time windows:

E-mail address: ﬁgliozzi@pdx.edu

xki0 ¼ 0; xknþ1;i ¼ 0; 8i 2 V; 8k 2 K ð6Þ

xki;j ðyki þ g i þ t i;j ðyki þ g i ÞÞ 6 ykj ; 8ði; jÞ 2 A; 8k 2 K ð11Þ

xkij 2 f0; 1g; 8ði; jÞ 2 A; 8k 2 K ð12Þ

xei 2 f0; 1g; xli 2 f0; 1g ð19Þ

4.1. Construction phase

4.2. Improvement phase

4.3. The auxiliary routing algorithm Hr

4.4. The route construction algorithm Hc

4.5. The route improvement algorithm Hi

5. Service time improvement

pðyq ; aq Þ ¼ cd dq;qþ1 þ ct ðaqþ1 yq Þ þ ce ðe#q ykq Þþ þ cl ðykq e#l Þþ ð22Þ

where aq+1 = yq + gq + tq,q+1(yq + gq) and subject to lq P yq P aq .

5.1. Service time improvement algorithms

Both algorithms try to reduce the interval ½yki ; y

6. Proposed benchmark problems

6.1. Constant speed problems with hard time windows

6.2. Time dependent problems with hard time windows

and the corresponding travel speeds are:

 TD1a = [1.00, 1.60, 1.05, 1.60, 1.00],

 TD1b = [1.60, 1.00, 1.05, 1.00, 1.60],

 TD1c = [1.60, 1.60, 1.05, 1.00, 1.00],

 TD1d = [1.00, 1.00, 1.05, 1.60, 1.60],

6.3. Time dependent problems with soft time windows

 TD4 = [1.10, 0.85, 1.10, 0.85, 1.10],

6.4. Time windows elasticity

ei ei þ uniform½1:0; 1:0 0:02 ðl0 e0 Þ; 8i 2 C

0:00 6 d1 6 0:65; 0:25 6 d2 6 1:00; 0:00 6 d3 6 0:20; 0:00 6 d4 6 0:03

and d1 + d2 + d3 = 1; d0 is a sufﬁciently large number (primary objective is to minimize ﬂeet size).

7.1. Constant speed problems with soft time windows

7.2. Time dependent problems with hard time windows

Method R1 R2 C1 C2 RC1 RC2

Travel time distribution R1 R2 C1 C2 RC1 RC2

Travel time distribution R1 R2 C1 C2 RC1 RC2

Travel time distribution R1 R2 C1 C2 RC1 RC2

Travel time distribution R1 R2 C1 C2 RC1 RC2

Fraction depot working time 0.0–0.2 0.2–0.4 0.4–0.6 0.6–0.8 0.8–1.0

7.3. Time dependent problems with soft time windows

7.4. Time window elasticity

Travel time distribution R1 R2 C1 C2 RC1 RC2

Travel time distribution R1 R2 C1 C2 RC1 RC2

Statistic R1 R2 RC1 RC2

n Ratio O(n2) O(n3) % O(n3)

8.1. Travel time updates

You might also like

TD1a = [1.00, 1.60, 1.05, 1.60, 1.00],

TD1b = [1.60, 1.00, 1.05, 1.00, 1.60],

TD1c = [1.60, 1.60, 1.05, 1.00, 1.00],

TD1d = [1.00, 1.00, 1.05, 1.60, 1.60],

TD4 = [1.10, 0.85, 1.10, 0.85, 1.10],