Omega 55 (2015) 24–37

Contents lists available at ScienceDirect

Omega
journal homepage: www.elsevier.com/locate/omega

Vehicle routing–scheduling for municipal waste collection system

under the “Keep Trash off the Ground” policy$
Shan-Huen Huang a,1, Pei-Chun Lin b,n
a
Department of Logistics Management, National Kaohsiung First University of Science and Technology, No.1, University Road, Yanchao District, Kaohsiung
City 824, Taiwan
b
Department of Transportation and Communication Management Science, National Cheng Kung University, No. 1, University Road, Tainan 701, Taiwan

art ic l e i nf o a b s t r a c t

Article history: This paper investigates the waste collection problem and characterizes the problem as a set-covering and
Received 13 December 2012 vehicle routing problem (VRP) complicated by inter-arrival time constraints. The study proposes a bi-level
Accepted 14 February 2015 optimization formulation to model the split delivery VRP with multiple trips to determine the minimum-
Available online 28 February 2015
distance route. The first stage optimally plans the collection points that cover all residential blocks. The
Keywords: second stage applies a heuristics method to solve the minimum vehicles used and minimum distance
Waste collection problem traveled for collecting residential waste. This research contributes to model this period VRP and to introduce
VRP the heuristics method to solve the problem efficiently. The study is important in laying the groundwork for
Split delivery understanding the possibility of improving the service level of municipal solid waste collection.
& 2015 Elsevier Ltd. All rights reserved.

1. Introduction block collection is the elimination of permanent containers and

Solid waste management is among the mandatory and essential
services provided by municipal authorities in most countries to keep
urban centers clean [1]. However, it is often among the most poorly
rendered services—the systems applied are unscientific, outdated and
inefficient [2]. Most developing countries have to wrestle with the
problem of efficient solid waste management in the face of increasing
waste-generation rates, high collection costs and dwindling financial
resources [3]. In some cities, waste is dumped randomly and litters the
streets [4]. Public waste collection points are usually extremely smelly
and create a sight that largely affects the hygiene and image of a
modern society. This once was the situation in Taiwan, the second
most densely populated country in the world, where 23 million people
crowd 35,800 km2 of which two-thirds area is rugged mountains. The
central government, therefore, decided to change the waste collection
system and employ several policies. Among these policies, those
having the most direct influence on residents' lifestyle are the mand-
atory garbage-sorting policy and the “Keep Trash off the Ground”
policy, which requires residents to personally bring their waste from
their house to dump it into collection vehicles. One advantage of
changing the collection system from designated dumping locations to
☆
This manuscript was processed by Associate Editor Yagiura.
n
Corresponding author. Tel.: þ 886 62757575x53222; fax: þ 886 62753882.
E-mail addresses: shanhuen@nkfust.edu.tw (S.-H. Huang),
peichunl@mail.ncku.edu.tw (P.-C. Lin).
1
Tel.: þ886 76011000x3218.
storage sites. However, if family members are out when collectors
pass by, the waste must be left at home for the next scheduled
collection day, causing a considerable nuisance.
This paper examines the problem of efficiently routing and
scheduling collectors for municipal solid waste collection in Taiwan.
Municipal solid waste management includes a heterogeneous collec-
tion of waste produced in urban areas, the nature of which varies
from region to region [5]. In the United States and most European
countries, residents place trash and household waste in garbage cans
for curbside collection once-a-week. The resulting problem of collect-
ing municipal waste involves visiting each street in a residential
network, and studies have typically modeled the problem as a
capacitated arc routing problem (CARP), and sought to solve it
accordingly [6–13]. Under Taiwan's “Keep Trash off the Ground”
policy, the waste collection problem is modeled as a vehicle routing
problem (VRP), involving point-to-point collection with stop points
for collection trucks scattered throughout the area.
Due to the hot climate of Taiwan, most residential blocks require
four visits per week (e.g. Monday–Tuesday–Thursday–Saturday) and
the same collection schedule is repeated weekly. To allow residents
with daytime jobs to dump their refuse, collection trucks must visit
neighborhoods at different times on each collection day, particularly
in the evening. Therefore, establishing vehicle routes requires that
each collection point is first scheduled [14]. Russel and Igo [15]
examined a refuse collection routing problem in which the objective
was to assign customer demand points to days of the week and
to solve the resulting node routing problems for the entire week.

http://dx.doi.org/10.1016/j.omega.2015.02.004
0305-0483/& 2015 Elsevier Ltd. All rights reserved.
 S.-H. Huang, P.-C. Lin / Omega 55 (2015) 24–37
Because residential refuse collection involves regularly scheduled
collection [16,17], several researchers have modeled the problem of
collecting waste as a periodic VRP [15,18–21].
This paper applies management science to the collection of waste
in Taiwan, to minimize collection costs while providing adequate
service for the entire residential network with fixed charge [22]. The
current study characterizes this real-world vehicle routing–scheduling
problem as a set-covering and routing problem [23,24] complicated by
inter-arrival time constraints at each collection point. To solve a set
covering problem [25], we must first determine the collection points
at which trucks temporarily stop to collect waste. This problem is a
very rich and challenging VRP [7]. Several heuristic algorithms with
stochastic techniques were employed to solve complex VRPs and were
found to be powerful form the viewpoints of management science
[26–29]. We referred to the urban waste collection problem solving
with ants heuristics presented by Bautista et al. [6], in which the
objective was to minimize the total traveling cost. However, our
problem differed from that of Bautista et al. [6] in two respects. First,
under Taiwan's “Keep Trash off the Ground” policy, the waste
collection problem is directly modeled as a VRP that involves point-
to-point collection with stop points. In comparison; Bautista et al. [6]
considered the problem of collecting municipal waste as involving
visiting each street in a network, and modeled the problem as a CARP,
then they transformed the problem into a node routing one. Second,
we dealt with scheduling collectors and the number of stops along a
route. In contrast, Bautista et al. [6] considered forbidden turns, which
implied important constraints for defining feasible routes and speci-
fied the routing plan while minimizing collection costs. This study
introduces a heuristic based on ant colony optimization (ACO) to
overcome the routing–scheduling problem of municipal solid waste
collection. Dorigo et al. [30] was the first to propose the ACO based on
the food-foraging behavior of ant colonies and researchers have
advocated the ACO algorithm for solving VRPs [32,33].
Because of the successful application of the ACO, several versions of
ant-inspired algorithms have been proposed for routing problems
[34–36]. The two most popular ones are ant system (AS) [30] and the
ant colony system (ACS) [31]. Ant system (AS) was the first ACO
algorithm to be proposed in the literature [30]. Its main characteristic
is that the pheromone values are updated by all the ants that have
completed the tour. Once all ants have computed their tour (i.e. at the
end of each iteration), the AS updates the pheromone trail using all
the solutions produced by the ant colony. Each edge belonging to one
of the computed solutions is modified by an amount of pheromone
proportional to its solution value. At the end of this phase the
pheromone of the entire system evaporates and the process of
construction and update is iterated. In the ACS only the best solution
computed since the beginning of the computation is used to globally
update the pheromone. As was the case in AS, global updating is
intended to increase the attractiveness of promising route but the ACS
mechanism is more effective because it avoids long convergence time
by directly concentrate the search in a neighborhood of the best tour
found up the current iteration of the algorithm [37]. The main idea of
the algorithm developed in this research was inspired by the ACS,
which accounts for exploration and exploitation during the transition
state. The ants cooperate using an indirect form of communication
mediated by pheromone trails of scent and find the best solution to
their tasks guided by both information (exploitation) which has been
acquired and search (exploration) of the new route.
Scheduling routes for the collection of solid waste involves
several factors, including the number of stops along a route, the
amount of solid waste accumulated at each refuse-collection point,
and road access to the refuse-collection points [38]. The schedule of
workers is limited each workday, and full garbage trucks must
travel to the nearest available landfill. Therefore, each truck typi-
cally delivers the collected waste to a landfill, returning to the depot
for a rest several times per workday. Vehicle capacity dictates the
25
number of landfill trips each truck must make. The service demand
associated with each collection point is estimated using demo-
graphic data to ensure that routing results fulfill the capacity
constraints of designated collection trucks. The main objectives of
this study are to (1) minimize the total frequency with which
collection points are visited, while providing service of a higher
level than that currently provided and (2) determine efficient routes
in terms of travel distance required to collect waste.
The remainder of this paper is organized as follows. Section 2
provides a detailed description of the set-covering problem used
to determine collection points and the routing–scheduling pro-
blem of municipal waste collection, as well as formulations for the
trucks and distance minimization model. Section 3 discusses the
ACO algorithm in detail and describes the procedures involved in
applying the ACO to resolve routing–scheduling problems. Section
4 presents the results, which demonstrate the benefits of incor-
porating the minimum allowable inter-arrival time between two
consecutive collections on any block. The final section summarizes
the findings and suggests avenues for further research.
2. Model formulation
This study proposes a bi-level optimization model that first
plans collection points by solving a set covering problem, followed
by route planning based on solving a VRP with pickup and delivery.
The first stage reviews the municipality made up of a number of
neighborhoods for waste collection trucks to stop temporarily and
designates collection points at street intersection of the residential
network to collect the waste of any adjacent neighborhood. Fig. 1
illustrates that intersection m is a potential collection point which is
adjacent to blocks 5, 8, and 9. If the municipality designates a
collection point at intersection m, then the collectors will collect
waste produced by the residents of blocks 5, 8, and 9. The objective
of the model is to minimize the frequency of collection over the
entire residential network, by providing residents with convenient
services for the disposal of waste by individuals. However, many
residents dislike collection points near their front door, not only
because of objectionable odors, but also because of physical
obstructions. Although this study permits trucks to stop at any
street intersection (node) more than once during a single trip, it
avoids the noise of trucks hovering around neighborhoods. More-
over, a reduction in the number of stops facilitates routing and
reduces transportation and handling costs. Given a connected graph
G ¼ ðV; A; BÞ with V as the set of vertices representing the intersec-



in this study, A ¼ ði; jÞ : i; j AV; i a j is the set of arcs,
tion points
and B ¼ ði; j; k; …Þ : i; j; k; … A V; i a ja k represents the set of
blocks in which the tuple of street intersections ði; j; k; …Þ defines
block b. Referring to Fig. 1, block 5 is characterized by an intersec-
tion tuple ði; j; n; m; l; kÞ.
Fig. 1. Adjacency of blocks and intersections.
26 S.-H. Huang, P.-C. Lin / Omega 55 (2015) 24–37

This study then creates a integer decision variable xi for each

intersection i to represent the frequency of collection at point i,
selected to service adjacent blocks. This variable will be at least
one, if the municipality designates a collection point at intersec-
tion i; otherwise, it will be zero. The municipality may designate
any street intersection as a collection point for multiple stops in a
single day to minimize the total frequency of collection points
visited while fulfilling the demand for collection. This leads to the
following integer programming model:
P
Minimize xi
i
P 
subject to xi U adjacent i;b Z required_serviceb 8 b ð1Þ
i
xi A Z þ [ f0g 8i
Notice that the constraint coefficient adjacent i;b is 1 if street
intersection i is adjacent to block b; otherwise, it is 0. The ith column
of the constraint matrix in Eq. (1) represents the set of blocks served
by a collection point at intersection i. We need to find a set of subsets
for i that covers the set of all blocks to ensure that every intersection
appears in the service subset associated with at least the number of
collection services for block b ðrequired_serviceb Þ based on the
demand for collection. For instance, block 5 in Fig. 2 requires four
visits on a waste collection day and the results of solving the integer
programming model (1) suggest selecting intersections i; j; k as
collection points, in which intersection k will be visited twice;
intersections i and j will each be visited once, respectively.
After determining the location of stops to collect waste according
to the service frequency required by each block, the model then
determines the routes taken by the collection trucks and the waste
collection schedule. A collection truck makes several trips departing
from and ending at the depot each workday. Each trip consists of
visiting a sequence of collection points, where waste is picked up, and
one destination, where waste are delivered. In Fig. 3, {depot, i1, i2, i3,
i4,destination} and {depot,i1, j1, j2, j3, j4,destination} are both
feasible routes for collection trucks. If the route {depot, i1, i2, i3,
i4,destination} is assigned to the first trip of collection truck k and
{depot, i1, j1, j2, j3, j4,destination} is assigned to the second trip of
collection truck k, then collection point i1 will be served twice by truck
k. All truck routes in the planning horizon constitute a complete
routing plan, ensuring that all of the waste at each collection point is
collected. Unlike the classical VRPs using a fleet of capacitated vehicles
to service a set of customers with known demand, this study removes
the constraint of having to visit each customer exactly once. In other
words, split collections are allowable and necessary. This is analogous
to split delivery VRP (SDVRP) [39–43]. Compared to traditional SDVRP,
the collection service in this study adds a number of visits to service
residents with a variety of work schedules, involving time constraints
that establish time intervals during which pickup service should avoid
taking place at each collection point. The model then formulates a
Fig. 3. Routes of the multi-trip SDVRP with pickup and delivery.
multi-trip SDVRP with pickup and delivery to determine routes,
consisting of a set of multiple trips and the minimum total distance
served by a fleet of capacitated homogeneous vehicles. This formula-
tion is a combination of the SDVRP presented by Dror et al. [39] and
the VRP with pickup and delivery presented by Desaulniers et al. [44],
where deliveries occur only at the depot/destination.
Let N be the set of selected collection points based on the optimal
solution of the integer programming model (1), and o and d be the
origin and destination for trucks, respectively. The set E ¼ fði; jÞ :
i; jA N [ fo; dg; i a jg comprises all feasible arcs joining the collection
points. In this manner, the multi-trip SDVRP with pickup and delivery
is defined on a complete undirected graph. To build the vehicle and
distance minimization model for the multi-trip SDVRP with pickup and
delivery, we defined the following parameters and variables:
K ¼ the set of collection trucks owned by a local municipality;
xni ¼the optimal frequency of collection point i to serve its
adjacent blocks every workday;
qif ¼the amount of waste to be collected for the fth collection of
point i, i A N, f A Z þ ;
if ¼the decision variable representing whether the fth collec-
zkt
tion of point i is assigned to truck k on its tth trip, iA N and
t A Z þ , f ¼ 1…xni , f A Z þ , 8 k A K; zkt
if ¼ 0 or 1;
ykt 0 ¼ the decision variable representing the flow from collec-
ijf f
tion point i headed to collection point j; ykt 0 ¼ 1 if arc ði; jÞ is
ijf f
assigned to collection truck k on its tth trip for the fth collection
at point i and f'th collection at point j, ykt ijf f ¼ 0; otherwise, for all
0 0
i a j, i; j A N, t A Z þ , f ¼ 1…xni , f ¼ 1…xnj , f ; f A Z þ , 8 k A K;
H kt ¼ the decision variable representing the time at which
if
collection truck k on its tth trip starts serving point i for the
fth collection;
dij ¼the distance from collection point i to collection point j,
i; jA N;
Q ¼the maximum service capacity for each waste coll-
ection truck;
v ¼ the average travel speed of trucks;
si ¼ the time duration of waste collection at point i for each
collection;
W ¼the regular working time for vehicle drivers per day.

Next, we formulate a multi-trip SDVRP with pickup and

delivery as a multi-objective programming model to minimize
the number of the required fleet ðjK jÞ and total distance of trips
 
PPP P P P
dij Uykt
ijf f 0 :
t f f 0 k A K i A N [ fogj A N [ fdg
8
< jK j
PPP P P P
minimize dij U ykt 0:
ð2Þ
ykt 0 ;zkt ;Hkt : t 0
ijf f
Fig. 2. A possible way to fulfill number of collection services for block 5. ijf f ifif f f k A K i A N [ fogj A N [ fdg
 S.-H. Huang, P.-C. Lin / Omega 55 (2015) 24–37 27

s.t. the proven optimal solution up to 10 collection nodes. These results

! are consistent with the results of ACO.
X X
qif zkt
if  1 ¼0 8 i A N; f ¼ 1…xni ð3Þ
þ
tAZ kAK

X X 3. ACO heuristic for the SDVRP

zkt
if r ykt
ijf f
0 8 i A N [ fog; 8 k A K; 8t AZ þ ; f ¼ 1…xni
0
f j A N [ fdg Two major components of the proposed solution scheme are
ð4Þ described in this section: the ACO method and the route improvement
X X X X algorithm. This solution procedure adopts an ACO algorithm to solve
ykt
ijf f 0
¼ ykt
jif 0 f
8 j A N; 8 k A K; 8t AZ þ ; for the multi-trip SDVRP (Eqs. (2)–(14)). The local search and the
f i A N [ fog f i A N [ fdg problem-specific route improvement heuristic explores the neighbor-
0
f ¼ 1…xnj ð5Þ hood of the solution identified. The ACO is a probabilistic technique for
solving computational problems, which can be reduced to the identi-
!
X 0
fication of good routes through graphs. Provided an undirected edge-
qif ykt
idf f
0 1 ¼0 8 k A K; 8t AZ þ ; f ¼ 1…xnd ; weighted graph and distinct depot node, the problem involves the
i A N [ fog
designation of trips, where a trip is a closed route starting and ending
f ¼ 1…xni ð6Þ at the same depot node, such that the required fleet size and the total
0 1 travel distance are minimized. To fulfill these objectives, the ACO is
X used to determine both the time that designated trucks collect waste at
qjf @ 1A ¼ 0
0
ykt 0
ojf f
8 k A K; 8t AZ þ ; f ¼ 1…xno ;
j A N [ fdg
each point and the arc from point i to point j forming the route. The
notations required for the ACO algorithm are provided below:
f ¼ 1…xnj ð7Þ

n
q0 ¼a predefined real number where 0 r q0 r1;
xi
X X þ Ωi ¼the set of all arcs connected to the unvisited node j, such
qif zkt
if r Q 8 k A K; 8t AZ ð8Þ
f ¼ 1 iAN
that the ants in node i passing through arc ði; jÞ will not violate
any constraint;

H kt kt τi;j ¼the pheromone value allocated on arc ði; jÞ;
ykt
ijf f 0 if þ si þ dij =v  H jf 0 r 0
wi;j ¼the length of arc ði; jÞ;
8 i A N; 8 jA N; 8 k A K; 8 t A Z þ ; f ¼ 1…xni ; ηi;j ¼the reciprocal of the length of arc ði; jÞ, representing the
0
f ¼ 1…xj n
ð9Þ desirability of arc ði; jÞ; ηi;j ¼ 1=wi;j ;
ψ NN ¼the initial solution;
J ψ NN ¼the weight of the initial solution;
H kt
if þ si þ dij =v r W 8 k A K; 8 i A N [ fog; 8 j A N
ψ gl ¼the best solution found so far;
þ
[ fdg; 8 t A Z ; f ¼ 1…xni ð10Þ J ψ gl ¼the weight of ψ gl ;
ant ¼ the predefined number of ants dispatched in each
0
H kt k0t0 kt k0t0
if þ 30  H i;f þ 1 r 2  zif  zi;f þ 1 8 i A N; 8 k; k A K; 8 t; t 0 A Z þ ; iteration;
0
t 4 t; f ¼ 1…xi n
ð11Þ Iter ¼the predefined number of iterations.

0
ykt
ijf f
0 A f0; 1g 8 i; j A E; 8 k A K; 8 t A Z þ ; f ¼ 1…xni ; f ¼ 1…xnj ð12Þ
3.1. The construction process
zkt
if A f0; 1g 8 iA N; 8 k A K; 8t AZ þ ; f ¼ 1…xni ð13Þ
Ants explore a given graph in a haphazard order until they form a
H kt
if Z 0 8 i A N [ fo; dg; 8 k A K; 8t AZ þ ; f ¼ 1…xni ð14Þ route. Each ant deposits a specific amount of pheromone on trails so
that following ants can refer to the wisdom of previous ants in further
Objection function (2) minimizes the number of collection exploring the graph. The higher the density of pheromones in an arc,
trucks used and the total distance traveled. Constraints (3) and the higher the probability that the following ants will be attracted to
(4) impose that each collection request is served exactly once. the arc. A random number q is extracted from a uniform distribution
Constraints (5)–(7) characterize a multi-commodity flow ensuring over interval (0,1) and the predefined parameter q0 directs the ants'
that every trip t of truck k starts from the origin depot o and behavior toward either the exploration of new path or the exploitation
terminates its route at the destination depot d. Constraints (8) of the best path that have previously emerged. If q is less than the
specify that vehicle capacities are never exceeded, where xni is the predefined parameter q0 ðq rq0 Þ, then β the next arc chosen is the arc
 α
result of solving Eq. (1), representing the optimal frequency of with the highest attraction τi;j ηi;j (exploitation). The best state s
collection at point i to service the neighborhood. The requirements is chosen with probability q0 , leading to the following formula:
of compatibility between routes and schedules are handled by

 α  β
constraints (9) and (10), requiring that the time spent traveling s ¼ max τi;j ηi;j if q r q0 ðexploitationÞ ð15Þ
j A Ωi
back to the depot after each collection by vehicle k on its tth trip
should not exceed the regular working time for a vehicle driver per Otherwise, an ant will move from node i to node j with
day. Next, constraints (11) express that an inter-arrival time probability pi;j (exploration) given by:
between two consecutive collections for any point i should be 8 α β
> ðτi;j Þ ðηi;j Þ

more than 30 min. Finally, binary and nonnegativity requirements >
< P α β 8 jA Ωi
ð Þ ð i;k Þ
τ η
are given by (12)–(14). pi;j ¼ k A Ωi i;k ð16Þ
>
>
In order to validate the effectiveness of the ACO algorithm, we : 0 =Ω
8 j2
i
compared the results of solving problems less than eleven collection
nodes using optimization-based MIP with using the ACO. Because the where α and β are parameters controlling the influence of the
32-bit version of the Windows operating system limits any GAMS or pheromone value allocated on arc ði; jÞ and the desirability of arc
solver process to 1.7 Gb, the optimization-based MIP only presents ði; jÞ, respectively. The route construction steps are as follows:
28 S.-H. Huang, P.-C. Lin / Omega 55 (2015) 24–37
Step (1) Dispatch an ant to search the graph for feasible nodes
that are unvisited and have required visits of nonzero. Select
the next node according to the selection mechanism s referring
to Eq. (15), and then move the ant to the selected node j. Record
the arrival time at node j and its adjacent blocks. Reduce the
required visits at node j by one and mark node j as infeasible.
Step (2) If there are no more feasible nodes that still require
services and do not violate any constraint, then move the ant
back to the depot and a trip is completed. Otherwise, go to
Step (1).
Step (3) If there is time available for another trip, resume with
the truck capacity as empty and the nodes with non-zero
required frequency to stop as feasible. Otherwise, increase the
number of trucks by 1.
Step (4) Repeat steps (2) through (3) until all nodes are served
as requested. One daily routing plan is completed.
Step (5) Each routing plan is compared to the current best
routing plan. The current best routing plan is updated if the
new solution is better than it.
3.2. Trail updating
The local pheromone-updating rule is applied to the network
whenever an ant finishes constructing a route. The effect of local-
updating to cause dynamic changes in the desirability of edges:
every time an ant uses an edge, this edge becomes slightly more
desirable, and the pheromone value remains the same as the
initial pheromone only on the edges that have never been
traversed. Once all of the ants have completed their tour (i.e., at
the end of all iterations), the global pheromone-updating proce-
dure is triggered. Global pheromone-updating is intended to
increase the attractiveness of promising routes by directly con-
centrating the search in the neighborhood of the best route found
up to the current iteration of the algorithm.
Each time an ant moves from the current node i to the next
node j the pheromone associated with the edge is modified in the
following way:
 
τi;j ¼ 1  ρ U τi;j þ ρ U τ0 ð17Þ
where ρ is the rate of pheromone evaporation and 0 o ρ o 1. jV j
denotes the number of nodes. τ0 is the initial pheromone value,
and is defined as τ0 ¼ 1=ðjV j U J ψ NN Þ; J ψ NN is the weight of the initial
solution.
Once all of the ants have completed their tour (i.e., at the end of
all iterations), the global pheromone-updating procedure is trig-
gered. Global pheromone-updating is intended to increase the
attractiveness of promising routes by directly concentrating the
search in the neighborhood of the best route found up to the
current iteration of the algorithm.
The ACO updates the pheromone allocation on the arcs that
constitute the current optimum route ψ gl after a predefined
number of iterations and the pheromone levels on the other arcs
are set to their current status. The global pheromone updating rule
is as follows:
τi;j ¼ ð1  ρÞ U τi;j þ ρ=J ψ gl ði; jÞ A ψ gl
3.3. Local search
The route construction procedures and pheromone updating
processes as described are typical for ACO as it is applied to the
vehicle routing problem [32–36]. However, research shows that
the attainment of improved solutions to the VRP is dependent on
route improvement strategies in the algorithm [45]. The strategies
involve the inclusion of a local exchange procedure to act as an
improvement heuristic within the routes found by individual
vehicles. The technique used for this purpose is the common 2-
opt heuristic where possible pairwise exchanges of locations
visited by individual vehicles are tested to see if an overall
improvement in the objective function can be attained. Note that
for each iteration, the ACO dispatches a predefined number of ants
and once the ACO procedure is completed, a set of collection truck
routes are determined, along with the intersections where each
truck should stop and pick up solid waste. In the local search, the
sequence of collection interactions is iteratively swapped and
subsequently generates new truck routes. If the resulting set of
truck routes yield feasible and lower traveling distances, the swap
is made and the process is repeated. The algorithmic details are
described below.
Step (a) Merging collection routes. Fig. 4 illustrates the
route merging procedure. Suppose that two trucks (collection
truck A and collection truck B in this example) are selected
to merge collection routes. The current routes of both trucks
({ ,1,2,3,4,5, ,6,7,8,9, ,10,11,12,13, } and { ,14,15,16,17, ,18,19,
20,21,22, ,23,24,25,26,27, }, respectively) are determined by
the solution procedure presented in the earlier section, where
represents the depot and other integers represent collection
points for trucks to stop. In this step, both routes are merged.
Step (b) Swapping the sequence of collection interactions and
calculating the objective value. In this step, one examines
whether swapping the selected pairs results in a better objec-
tive function. If a lower objective value (the transportation
distance) is obtained, the newly generated routes are retained
as a new neighborhood solution. Otherwise, the routes are
abandoned and the next combination is attemped. Fig. 5
illustrates the route swapping procedure.
3.4. Problem specific route improvement
After the local search attains an improved local optimal solu-
tion and lists all routes for collection trucks, the following route
improvement steps are implemented. A higher service rate in this
study indicates that more households are able to dump their waste
into collection trucks as a percentage of total households. Due to
Fig. 4. Swapping part of routes in the local search procedure.
ð18Þ
Fig. 5. The proposed ACO algorithm for solving the multi-trip SDVRP.
 S.-H. Huang, P.-C. Lin / Omega 55 (2015) 24–37 29

Fig. 6. Location of road map codes for Test Instance I.

Fig. 7. Assignment of the optimal collection frequency to every interaction for Test Instance I.
30 S.-H. Huang, P.-C. Lin / Omega 55 (2015) 24–37

Table 1
Computational results of objective functions.

α β

1 2 3 4 5 6 7 8 9 10

1 Number of trucks 4 5 5 4 4 5 5 4 4 5
Total distance 101.36 98.37 102.43 11.69 100.17 99.44 105.27 99.39 103.93 97.42
Execution time (min) 11 10 10 11 11 10 10 10 10 10
2 Number of trucks 4 4 4 5 5 4 4 4 4 5
Total distance 101.75 105.07 103.31 96.93 99.59 103.34 97.63 104.9 103.02 89.79
Execution time (min) 10 10 10 10 10 10 10 11 10 10
3 Number of trucks 5 5 4 5 5 5 4 4 5 4
Total distance 102.6 98.36 102.69 99.71 102.11 92.88 99.77 99.79 104.2 102.41
Execution time (min) 10 10 10 10 10 10 10 10 10 10
4 Number of trucks 4 5 5 4 4 5 4 4 5 4
Total distance 96.94 100.83 98.41 98.37 99.27 102.93 106.57 99.46 95.66 103.9
Execution time (min) 10 10 10 11 10 10 10 10 10 10
5 Number of trucks 4 5 4 4 4 4 5 5 5 5
Total distance 102.5 98.77 101.19 101.66 100.58 103.14 100.06 105.86 102.41 103.24
Execution time (min) 10 10 10 10 10 10 10 11 10 10
6 Number of trucks 4 5 4 4 4 4 5 4 5 4
Total distance 102.07 104.29 90.24 104.25 102.42 96.86 102.88 105.17 100.42 101.8
Execution time (min) 10 10 10 10 10 10 10 10 10 10
7 Number of trucks 4 4 5 4 5 4 5 4 4 5
Total distance 103.26 103.98 102.38 100.27 101.53 104.2 100.45 98.34 101.34 103.91
Execution time (min) 10 10 10 10 10 10 10 10 10 11
8 Number of trucks 4 4 5 4 5 4 4 4 4 5
Total distance 101.08 100.08 100.54 105.28 105.23 93.68 101.52 101.64 101.85 101.07
Execution time (min) 10 11 10 10 10 10 10 11 10 10
9 Number of trucks 4 5 5 4 5 4 5 4 5 5
Total distance 101.64 101.95 104.71 104.46 86.09 97.54 103.58 100.47 104.26 99.48
Execution time (min) 10 10 10 10 10 11 10 10 10 10
10 Number of trucks 5 4 4 4 5 5 4 5 5 5
Total distance 99.36 99.66 97.23 101.35 102.49 99.76 100.14 102.68 99.05 100.91
Execution time (min) 10 10 11 10 10 10 10 10 11 10
α β
1 2 3 4 5 6 7 8 9 10

Fig. 8. Optimal route depicted using TransCAD for Test Instance I.

 S.-H. Huang, P.-C. Lin / Omega 55 (2015) 24–37
Fig. 9. Optimal route depicted using TransCAD for Test Instance II.
the variety of work schedules among urban residents, collection
trucks need to service neighborhoods at various times throughout
each collection day, particularly in the evening. If family members
are away when the collector passes through, the waste must be
left at home for the next collection day, causing a considerable
nuisance. Ideally, the schedule should avoid short time gaps
between consecutive services for any blocks. To improve the rate
of service, an inter-arrival time between two consecutive collec-
tions for any block of less than the minimum allowable inter-
arrival time, such as 30 min, is considered “too close” and there-
fore treated as an imposed constraint to be fulfilled to the greatest
possible extent. This constraint is not considered an intrinsic limit
that must be satisfied during the routing process, because it often
leads to the designation of too many infeasible nodes and detains
ants at the depot.
Let τ be a predefined parameter as the minimum allowable
inter-arrival time between two consecutive collections for all
blocks, δζ;ζ þ 1 record the time gap between consecutive services
b
ζ and ζ þ 1 for block b, and ϕb record the frequency of the time gap
between two consecutive services for block b as less than τ min.
Thus, δζ;ζ þ 1 r τ should also imply an increase in ϕb by one. To
b
improve the service rate through efficient scheduling and routing,
31
this study incorporated a route improvement algorithm to reduce
the time gap frequency between two consecutive services for any
P
block of less than τ, that is to minimize ϕb . Suppose that Fig. 6
bAB
illustrates the local optimal solution and lists routes for collection
trucks A and B. The local optimal routing plan of truck A is
composed of trip 1, trip 2 and trip 3, and the plan of truck B is
composed of trip 4, trip 5 and trip 6. Search for blocks with too-
close visits and identify the pairs of trips (for instance, trip 1 and
trip 5) that resulted in this situation to swap. Because the whole
trips exchange, the total distance of the updated daily routing plan
keeps constant but two consecutive collections for any block of
less than the minimum allowable inter-arrival time might be
improved. Finally, the ACO for the multi-trip SDVRP can be
summed up by the schematic algorithm given in Fig. 7. The
process is terminated when the number of iterations reaches the
pre-specified number ant  Iter.
4. Test instances
To gain insight into whether the ACO is capable of efficiently
solving the multi-trip SDVRP with pickup and delivery, this research
32 S.-H. Huang, P.-C. Lin / Omega 55 (2015) 24–37

Table 2
The frequency of near visits.

α β

1 2 3 4 5 6 7 8 9 10

1 ACO 30 15 32 14 35 22 24 13 21 24
ACOþ Route improvement 9 2 3 1 13 8 11 4 4 8
Improving rate 70.00% 86.67% 90.63% 92.86% 62.86% 63.64% 54.17% 69.23% 80.95% 66.67%
2 ACO 31 32 30 37 35 30 22 45 26 34
ACOþ Route improvement 9 3 5 14 5 7 12 15 3 5
Improving rate 70.97% 90.63% 83.33% 62.16% 85.71% 76.67% 45.45% 66.67% 88.46% 85.29%
3 ACO 27 20 18 44 14 21 16 30 27 37
ACOþ Route improvement 5 2 9 14 4 3 4 12 15 6
Improving rate 81.48% 90.00% 50.00% 68.18% 71.43% 85.71% 75.00% 60.00% 44.44% 83.78%
4 ACO 38 37 11 15 48 15 12 34 18 15
ACOþ Route improvement 17 5 5 11 19 4 6 7 5 3
Improving rate 55.26% 86.49% 54.55% 26.67% 60.42% 73.33% 50.00% 79.41% 72.22% 80.00%
5 ACO 23 31 25 11 13 18 25 17 18 36
ACOþ Route improvement 4 11 8 6 9 8 7 4 2 7
Improving rate 82.61% 64.52% 68.00% 45.45% 30.77% 55.56% 72.00% 76.47% 88.89% 80.56%
6 ACO 13 35 38 40 25 23 23 17 38 35
ACOþ Route improvement 9 12 7 9 6 5 9 4 6 12
Improving rate 30.77% 65.71% 81.58% 77.50% 76.00% 78.26% 60.87% 76.47% 84.21% 65.71%
7 ACO 19 18 17 18 17 23 31 8 19 22
ACOþ Route improvement 7 9 2 6 6 8 15 3 7 8
Improving rate 63.16% 50.00% 88.24% 66.67% 64.71% 65.22% 51.61% 62.50% 63.16% 63.64%
8 ACO 45 34 27 40 27 29 14 53 23 15
ACOþ Route improvement 10 11 6 12 7 11 7 14 2 1
Improving rate 77.78% 67.65% 77.78% 70.00% 74.07% 62.07% 50.00% 73.58% 91.30% 93.33%
9 ACO 20 17 49 23 33 28 47 25 23 38
ACOþ Route improvement 4 13 14 3 11 12 16 10 3 8
Improving rate 80.00% 23.53% 71.43% 86.96% 66.67% 57.14% 65.96% 60.00% 86.96% 78.95%
10 ACO 42 41 50 17 34 21 23 27 14 36
ACOþ Route improvement 9 13 9 3 13 8 12 16 9 7
Improving rate 78.57% 68.29% 82.00% 82.35% 61.76% 61.90% 47.83% 40.74% 35.71% 80.56%
α β
1 2 3 4 5 6 7 8 9 10

Table 3
Computational results of objective functions versus τ.

τ 5 10 15 20 25 30 35 40 45 50 55 60

Number of trucks 4 4 4 4 4 4 4 4 4 4 4 5
Total distance 117.45 116.44 127.33 118.69 115.22 119.18 119.64 120.21 116.5 117.87 118.09 117.33
ACO 2 3 12 7 19 13 33 44 44 50 74 50
ACOþ Route improvement 0 0 1 2 5 6 7 12 16 26 30 33

tested two instance derived from a sub-network of Kaohsiung City in
Taiwan, Test Instance I comprising 262 nodes (four-digit or five-digit
code), 442 arcs, and 181 blocks (one-digit, two-digit or three-digit
code) as shown in Fig. 6; Test Instance II comprising 611 nodes, 1068
arcs, and 463 blocks. The data was mapped using available TransCAD
software tools.
blocks, scheduling collection times at road intersections and
requiring a minimum of 117 stops. The node with ID 8857 is the
depot. The nodes marked with a square indicate that one collec-
tion is required, the nodes with a circle indicate that two collec-
tions are required, and the nodes with a triangle denote nodes
requiring collection three times each day.

4.1. Collection points 4.2. Collection route

The first phase of the solution process involves determining the
stop points for waste collection trucks. The number of collections
for block b ðrequired_serviceb Þ in Eq. (1) is assumed to follow a
discrete uniform distribution in the set f1; 2; 3g. Appendix A lists
the number of collection services for each block of Test Instance I.
The integer programming model of the set covering problem was
run using GAMS/CPLEX in PC/Windows XP Professional with 2
quad-core CPUs and 1.96 gigabytes of RAM. Fig. 7 shows the
optimal solution for Test Instance I including 262 nodes and 181
The second phase of the solution process involves the deter-
mination of routes using the fewest trucks with the shortest travel
distances. In this study, the regular working time for each truck in
the test instance was 6 h, from 4 p.m. to 10 p.m., and vehicle
capacity was designated as 250 units. The quantity of garbage
generated each day from a block was assumed to follow a discrete
uniform distribution in the set f5; …; 22g and the waste generated
in any given block was assumed to be evenly distributed among
adjacent collection points. In this manner, we determined the
 S.-H. Huang, P.-C. Lin / Omega 55 (2015) 24–37
demand at each collection point. Each truck driver required a 30-
min break between every two consecutive routes; the average
truck speed was 10 miles per hour and the trucks remained at each
collection point for 3 min ðsi ¼ 3Þ. To apply the ACO algorithm to
solve the multi-trip SDVRP with pickup and delivery, parameters
q0 and the evaporation rate ρ were set to 0.1. The stopping criteria
for the algorithm included the number of epochs ðIter ¼ 32Þ
and the number of iterations for each epoch ðant ¼ 32Þ. This study
subsequently executed 1024 iterations for the test instance. For
ACO, Dorigo et al. [30] provided appropriate guidelines as
parameter settings for the traveling salesperson problem, with
recommended values of α ¼1 and β ¼5. However, optimal or
near-optimal settings of these parameters are heavily depen-
dent on the properties of a multidimensional objective function
surface associated with the optimization problem. Recommen-
dations for parameter guidelines can only be considered within
the context in which they were derived [46]. Therefore, we
conducted experiments to test the combinations with α and
β ¼ ½1; 2; 3; 4; 5; 6; 7; 8; 9; 10, and the output results are listed
in Table 1. Table 1 also illustrates that, on average, the execution
time required to apply the ACO to solve the test instance was
approximately 10 min. The average planned total distance ran-
ged from 86 to 106, indicating no significant difference among
settings of parameters α or β. The combination of α ¼ 9 and β ¼ 5
resulted in the best performance for the test instance according
to the objective functions. As shown in Figs. 8 and 9, this study
used TransCAD functions to depict the optimal route based on
the parameter combination of α ¼9 and β ¼5 for Test Instance I
and II. Appendix B shows an example of planned collection
times for each block of Test Instance I.
4.3. Solution improvement
Because the travel distance of each route is fixed, the computa-
tional results of total distances traveled remain unchanged. Table 2
lists the frequency of the inter-arrival time less than 30 min
between two consecutive collections for any block when applying
the ACO approach to solve the test instance with α and
β ¼ ½1; 2; 3; 4; 5; 6; 7; 8; 9; 10. Table 2 also lists the frequency
of the inter-arrival time less than 30 min between two consecutive
collections for any block after applying the route improvement.
Finally, Table 2 highlights the improvement and shows that after
applying the route improvement, the number of too-close visits
was reduced by an average of 68.97%.
This study also investigated the effects of changing the mini-
mum allowable inter-arrival time (τ) between two consecutive
collections on the solutions. On average, the execution time
required to apply the ACO approach to solve the test instance
remained at approximately 10 min, based on the parameter combi-
nation of α ¼ 9 and β ¼ 5. Table 3 shows the planned total distance
ranged from 116 to 128, which is not significantly different from
previous results shown in Table 1. Table 3 highlights the frequency
pattern of inter-arrival time between two consecutive collections
for any block less than τ min after applying route improvement
increase with τ. One possible explanation for this is that as the
minimum allowable inter-arrival time (τ) lengthened, the  constraint
of two consecutive services for block b less than τ min δζ;ζ þ 1 r τ
b
became increasingly difficult to satisfy. Therefore, incorporating the
route improvement algorithm into the solution process did not
significantly reduce the frequency of time interval between two
consecutive services for any block less than τ.
33
5. Conclusions
This paper formulates a minimum cost model for the municipal
waste collection problem based on Taiwan's “Keep Trash off the
Ground” policy; it also details the development of an ACO algorithm
that features route improvement to minimize vehicle use and
distance traveled, and to ensure that the time interval between
the two consecutive services for any block is not too short. In this
study, we compare the ACO and the ACO with route improvement
to confirm the superior performance of this adapted algorithm in
the multi-trip SDVRP with pickup and delivery. The results of
simulations and optimization reveal that the ACO algorithm with
route improvement provides considerable improvement over the
ACO alone. The contributions of this paper to the literature are the
modeling of a real period vehicle routing system and the introduc-
tion of the ACO to overcome this problem efficiently. This study may
be critically important in laying the groundwork for understanding
the possibility of improving the service level of municipal solid
waste collection.
Further research should conduct a more comprehensive investiga-
tion on large-scale SDVRP using the ACO algorithm and compare the
results to the currently best-known solutions. Future studies can
expand on the exploration of the present problem into a vehicle-
collection point dependency VRP [8] with a constraint prohibiting
vehicles of a certain class from servicing or traversing a street, such as
small alleys or bridges that only support a specific weight. Further
research could also consider potential collection points established in
the middle of streets instead of at intersections to shorten the walking
distance for residents on large-blocks. Researchers may also investi-
gate efficient combinations of small blocks with fewer residents to
reduce the number of collection points. In so doing, such research will
contribute to our growing understanding of how and to what extent
service policy influence on efficiently routing and scheduling collectors
for municipal solid waste collection before any new design is executed.
On 31 July 2014, deadly gas pipeline explosions occurred in Taiwan's
major industrial city, Kaohsiung. Thirty people were killed and 310
others were injured. Instead of delivering industrial gas via pipelines,
the other common method involves delivery in pressurized tank
trailers. For a tank-trailer routing and scheduling problem, the decision
maker must determine a distribution schedule and choose routes that
minimize risk and maximize efficiency. The routing and scheduling
results fulfill the goals of avoiding risks associated with transporting
industrial gases and attaining efficient delivery while conforming to
regulations and remaining consistent with good business practices.
Further study should develop a routing and scheduling decision
support system for industrial-gas deliveries that models this period
VRP and introduces the heuristics method to solve this problem
efficiently.
Acknowledgments
The authors gratefully acknowledges the helpful comments of
the editor and two anonymous reviewers who provided valuable
input and comments that have contributed to improving the
content and exposition of this paper. This research was supported
in part by Grant from the National Science Council of the Republic
of China in Taiwan under Grants NSC101-2221-E-006-215-MY2.
Appendix A
See Table A1.
34 S.-H. Huang, P.-C. Lin / Omega 55 (2015) 24–37

Table A1
The number of collection services for block b.

Block required Block required Block required Block required Block required Block required
serviceb serviceb serviceb serviceb serviceb serviceb

1 3 31 2 61 3 91 3 121 3 151 2
2 3 32 3 62 2 92 2 122 1 152 1
3 1 33 1 63 1 93 1 123 1 153 1
4 1 34 1 64 2 94 1 124 2 154 1
5 2 35 3 65 1 95 1 125 3 155 2
6 1 36 3 66 2 96 3 126 1 156 2
7 1 37 2 67 3 97 3 127 1 157 3
8 3 38 3 68 1 98 1 128 2 158 1
9 3 39 2 69 2 99 1 129 3 159 1
10 3 40 3 70 1 100 3 130 2 160 2
11 2 41 3 71 2 101 1 131 1 161 1
12 2 42 1 72 3 102 3 132 2 162 3
13 2 43 3 73 3 103 2 133 3 163 3
14 1 44 1 74 3 104 3 134 3 164 3
15 1 45 2 75 2 105 2 135 1 165 1
16 2 46 2 76 1 106 3 136 1 166 2
17 3 47 1 77 2 107 1 137 1 167 1
18 1 48 1 78 1 108 3 138 3 168 2
19 1 49 3 79 2 109 3 139 3 169 2
20 2 50 3 80 3 110 1 140 1 170 3
21 2 51 1 81 3 111 1 141 2 171 1
22 1 52 1 82 3 112 1 142 3 172 3
23 1 53 2 83 3 113 2 143 1 173 3
24 3 54 2 84 2 114 3 144 1 174 2
25 2 55 1 85 3 115 3 145 2 175 2
26 3 56 3 86 1 116 2 146 3 176 3
27 1 57 2 87 2 117 2 147 1 177 3
28 1 58 3 88 2 118 2 148 2 178 3
29 1 59 3 89 2 119 3 149 2 179 3
30 3 60 2 90 2 120 3 150 3 180 1
181 2

Appendix B

See Table B1.

Table B1
Collection time for each block.

Block 1st 2nd 3rd 4th Block 1st 2nd 3rd 4th Block 1st 2nd 3rd 4th

1 16:23 19:09 21:22 31 17:1 21:15 61 16:42 19:23 21:21

2 16:56 19:16 21:05 32 16:9 18:33 21:01 62 18:11 20:06
3 19:39 33 17:3 21:12 63 21:27 21:38
4 16:56 21:05 34 19:16 21:15 64 18:15 19:56
5 16:23 19:09 35 16:9 18:38 21:01 65 20:6
6 17:15 36 16:9 18:33 21:01 66 18:15 19:56
7 19:28 37 19:16 21:09 67 18:26 20:00 21:46
8 17:15 19:36 21:22 38 16:36 18:38 19:16 68 21:21
9 17:8 19:28 21:17 39 16:9 18:38 21:01 69 18:26 20:00 21:46
10 17:6 19:45 21:22 40 19:28 21:09 21:33 70 16:23 17:41 21:15
11 17:8 19:23 21:17 41 16:4 18:20 20:51 71 18:11 19:50
12 16:55 19:39 42 16:4 20:51 72 16:23 17:41 21:15
13 17:6 19:45 43 16:36 19:16 21:09 73 18:21 19:31 20:57
14 19:23 21:22 44 18:20 74 17:11 18:11 19:50
15 17:6 19:45 45 16:36 18:38 75 16:19 21:11
16 16:56 21:05 46 19:28 21:33 76 17:11
17 16:23 19:09 21:22 47 20:56 77 16:19 21:11
18 16:56 21:05 48 21:9 21:33 78 18:58 19:36

S.-H. Huang, P.-C. Lin / Omega 55 (2015) 24–37

19 16:23 19:09 49 16:4 18:20 20:51 79 18:58 19:36
20 16:28 21:28 50 16:4 18:15 19:56 20:51 80 19:1 19:58 21:46
21 17:6 19:45 51 16:36 21:27 21:38 81 16:23 17:41 21:15
22 17:8 21:17 52 18:20 82 18:21 19:31 20:57
23 17:1 53 18:15 19:56 83 19:1 19:58 21:46
24 17:8 19:23 21:17 54 20:6 20:56 84 16:23 17:41 21:15
25 17:1 21:15 55 18:26 20:00 85 18:21 19:31 20:57
26 17:3 19:23 21:12 56 16:42 19:23 21:33 86 19:50
27 17:1 57 21:27 21:38 87 16:19 21:11
28 17:3 21:12 58 18:26 20:00 20:56 88 18:58 19:36
29 21:15 59 20:6 21:27 21:38 89 17:11 18:05 19:50
30 16:28 19:28 21:28 60 18:11 20:56 90 16:19 18:48 21:11
Block 1st 2nd 3rd 4th Block 1st 2nd 3rd 4th Block 1st 2nd 3rd 4th 5th
91 18:15 18:58 19:36 121 17:00 17:55 18:08 151 16:08 20:44
92 17:11 18:05 122 19:13 152 18:07
93 18:15 19:01 19:58 123 16:51 153 16:34 18:52
94 18:48 124 17:55 18:08 154 18:57
95 16:22 21:36 125 18:00 18:58 19:43 155 18:06 18:07
96 17:38 19:01 19:58 126 18:53 156 16:45 18:57
97 16:22 16:28 21:36 127 16:47 157 16:34 18:52 20:44
98 17:38 128 16:27 18:42 158 18:06
99 18:15 129 16:40 18:53 20:49 159 16:45
100 16:28 17:46 20:05 130 16:27 18:42 160 16:19 18:20
101 18:05 131 16:47 161 16:45 20:29
102 17:46 19:09 20:05 132 17:00 18:46 162 16:33 17:54
103 17:43 18:05 133 16:40 18:00 20:49 163 16:19 18:20 20:29
104 17:05 17:38 18:15 134 16:47 18:00 21:44 164 16:11 18:06 20:12
105 18:48 19:09 135 17:00 18:46 165 20:29
106 16:22 17:43 21:36 136 21:44 166 16:19 18:20
107 17:00 137 16:27 18:42 167 16:40
108 18:48 19:13 21:35 138 16:34 18:52 18:53 168 16:19 18:20
109 16:22 16:28 21:36 139 16:27 18:42 169 16:38 17:57
110 17:05 17:38 140 16:53 18:46 170 16:38 16:40 17:57
111 19:13 141 16:08 18:53 171 16:13 18:14
112 17:05 142 16:08 16:40 20:49 172 16:11 17:26 20:12
113 16:28 17:46 20:05 143 16:40 18:00 18:57 20:49 173 16:13 16:40 18:14 20:35
114 17:46 19:09 20:05 144 16:53 174 17:26 20:06

35
36 S.-H. Huang, P.-C. Lin / Omega 55 (2015) 24–37

References
5th

[1] Chang NB, Davila E. Minimax regret optimization analysis for a regional solid
waste management system. Waste Management 2007;27(6):820–32.
[2] Asnani PU. Solid waste management. Indian infrastructure report. 2006.
20:35
18:28
4th

[3] Li JQ, Borenstein D, Mirchandani PB. Truck scheduling for solid waste
collection in the City of Porto Alegre, Brazil. Omega 2008;36(6):1133–49.
[4] Agunwamba JC. Analysis of scavengers' activities and recycling in some cities
of Nigeria. Environmental Management 2003;32(1):116–27.
[5] Diaz, LF, Savage, GM, Eggerth, LL, & Rosenberg, L. Solid Waste Management:
20:06
20:35
18:28
17:57

CalRecovery Inc.; 2006.

3rd

[6] Bautista J, Fernández E, Pereira J. Solving an urban waste collection pro-

blem using ants heuristics. Computers & Operations Research 2008;35(9):
3020–3033.
[7] Golden BL, Assad AA, Wasil EA. Routing vehicles in the real world: applications
in the solid waste, beverage, food, dairy, and newspaper industries. In: Toth P,
20:35
20:12

16:40
18:14
18:14
17:57
2nd

Vigo D, editors. The vehicle routing problem. Philadelphia: SIAM; 2002.

p. 245–86.
[8] Maniezzo, V. Algorithms for large directed carp instances: urban solid waste
collection operational support. Italy: Department of Computer Science, Uni-
versity of Bologna; 2004.
16:38
16:38

18:28
18:28

[9] Sniezek J, Bodin L. Using mixed integer programming for solving the
16:13
16:13
16:11
1st

capacitated arc routing problem with vehicle/site dependencies with an

application to the routing of residential sanitation collection vehicles. Annals
of Operations Research 2006;144(1):33–58.
[10] Sniezek J, Bodin L, Levy L, Ball M. Capacitated arc routing problem with
vehicle-site dependencies: the Philadelphia experience. In: Toth P, Vigo D,
Block

editors. The vehicle routing problem. Philadelphia, PA: SIAM; 2002.

175
176
177
178
179
180
181

p. 287–308.
[11] Toth P, Vigo D. Models, relaxations and exact approaches for the capacitated
vehicle routing problem. Discrete Applied Mathematics 2002;123(1):487–512.
[12] Belenguer JM, Benavent E. A cutting plane algorithm for the capacitated arc
routing problem. Computers & Operations Research 2003;30(5):705–28.
20:44
4th

[13] Ralphs TK. Parallel branch and cut for capacitated vehicle routing. Parallel
Computing 2003;29(5):607–29.
[14] Beltrami E, Bodin L. Networks and vehicle routing for municipal waste
collection. Networks 1972;4(1):65–94.
[15] Russell RA, Gribbin D. A multiphase approach to the period routing problem.
21:44

18:52

Networks 1991;21(7):747–65.
3rd

[16] Hirsch WZ. Cost functions of an urban government service: refuse collection.
The Review of Economics and Statistics 1965;47(1):87–92.
[17] Kreith F, Tchobanoglous G. Handbook of solid waste management. New York:
McGraw-Hill Professional; 2002.
[18] Christofides N, Beasley JE. The period routing problem. Networks 1984;14
16:34
18:07

18:07
18:57

17:54
2nd

(2):237–56.
[19] Tan CCR, Beasley JE. A heuristic algorithm for the period vehicle routing
problem. Omega 1984;12(5):497–504.
[20] Chao IM, Golden BL, Wasil E. An improved heuristic for the period vehicle
routing problem. Networks 1995;26(1):25–44.
18:00
21:44
16:53

16:53
16:53
16:08

[21] Angelelli E, Grazia Speranza M. The periodic vehicle routing problem with
1st

intermediate facilities. European Journal of Operational Research 2002;137

(2):233–47.
[22] Adlakha V, Kowalski K, Lev B. A branching method for the fixed charge
transportation problem. Omega 2010;38(5):393–7.
[23] Bautista J, Pereira J. Modeling the problem of locating collection areas for
Block

urban waste management. An application to the metropolitan area of

145
146

148
149
150
147

Barcelona. Omega 2006;34(6):617–29.

[24] Overholts Ii DL, Bell JE, Arostegui MA. A location analysis approach for military
maintenance scheduling with geographically dispersed service areas. Omega
2009;37(4):838–52.
[25] Tung DV, Pinnoi A. Vehicle routing-scheduling for waste collection in Hanoi.
19:43
4th

European Journal of Operational Research 2000;125(3):449–68.

[26] Bektas T. The multiple traveling salesman problem: an overview of formula-
tions and solution procedures. Omega 2006;34(3):209–19.
[27] Day JM, Wright DP, Schoenherr T, Venkataramanan M, Gaudette K. Improving
routing and scheduling decisions at a distributor of industrial gasses. Omega
21:35

21:35
19:43
19:43
18:58
18:08

2009;37(1):227–37.
3rd

[28] Mujawar S, Huang S, Nagi R. Scheduling to minimize stringer utilization for

continuous annealing operations. Omega 2012;40(4):437–44.
[29] Karaoglan I, Altiparmak F, Kara I, Dengiz B. The location-routing problem with
simultaneous pickup and delivery: formulations and a heuristic approach.
Omega 2012;40(4):465–77.
18:00

19:09
18:58
18:58
19:13
17:55
2nd

[30] Dorigo M, Maniezzo V, Colorni A. Ant system: optimization by a colony of

cooperating agents. IEEE Transactions on Systems, Man, and Cybernetics, Part
B 1996;26(1):29–41.
[31] Dorigo M, Gambardella LM. Ant colony system: a cooperative learning
approach to the traveling salesman problem. IEEE Transactions on Evolution-
18:00
18:00
16:51

16:51
17:43
17:05

ary Computation 1997;1(1):53–66.

1st

[32] Gambardella LM, Taillard E, Agazzi G. MACS-VRPTW: a multiple colony system

for vehicle routing problems with time windows. In: David C, Marco D, Fred G,
Dipankar D, Pablo M, Riccardo P, Kenneth VP, editors. New ideas in optimiza-
tion. UK: McGraw-Hill Ltd; 1999. p. 63–76.
[33] Reimann M, Doerner K, Hartl RF. D-Ants: savings based ants divide and
Block

conquer the vehicle routing problem. Computers & Operations Research

120
115
116

119
118
117

2004;31(4):563–91.
 S.-H. Huang, P.-C. Lin / Omega 55 (2015) 24–37
[34] Bell JE, McMullen PR. Ant colony optimization techniques for the vehicle
routing problem. Advanced Engineering Informatics 2004;18:41–8.
[35] Tin CJ, Chen CH. A multiple ant colony optimization algorithm for the
capacitated location routing problem. International Journal of Production
Economics 2013;141:34–44.
[36] Vaisakh K, Srinivas LR. Evolving ant colony optimization based unit commit-
ment. Applied Soft Computing 2011;11:2863–70.
[37] Darshini EP, Kaur G. Implementation of ACO algorithm for EDGE detection and
sorting salesman problem. International Journal of Engineering Science and
Technology 2010;2(06):2304–15.
[38] Sculli D, Mok KC, Cheung SH. Scheduling vehicles for refuse collection. Journal
of the Operational Research Society 1987;38(3):233–9.
[39] Dror M, Trudeau P. Savings by split delivery routing. Transportation Science
1989;23(2):141.
[40] Dror M, Laporte G, Trudeau P. Vehicle routing with split deliveries. Discrete
Applied Mathematics 1994;50(3):239–54.
[41] Dror M, Trudeau P. Split delivery routing. Naval Research Logistics 1990;37
(3):383–402.
37
[42] Ho SC, Haugland D. A tabu search heuristic for the vehicle routing problem
with time windows and split deliveries. Computers & Operations Research
2004;31(12):1947–64.
[43] Wang K, Ye C, Tang G. A new vehicle routing problem and its two stage
algorithm. Operations Research Transactions 2010;14(3):55–63.
[44] Desaulniers G, Erdmann A, Solomon MM, Soumis F. VRP with pickup and
delivery. In: Toth P, Vigo D, editors. The vehicle routing problem. Philadelphia:
SIAM; 2002. p. 225–42.
[45] Bullnheimer B, Hartl RF, Strauss C. Applying the ant system to the vehicle
routing problem. In: Voß S, Martello S, Osman IH, Roucairol C, editors. Meta-
Heuristics. US: Springer; 1999. p. 286–96.
[46] Zecchin AC, Simpson AR, Maier HR, Nixon JB. Parametric study for an ant
algorithm applied to water distribution system optimization. IEEE Transac-
tions on Evolutionary Computation 2005;9(2):175–91.