19.Anbuudayasankar S. P., Ganesh K., Mohapatra Sanjay, “Models for Practical Routing Problems in...
Transcript of 19.Anbuudayasankar S. P., Ganesh K., Mohapatra Sanjay, “Models for Practical Routing Problems in...
CHAPTER 2
SURVEY OF METHODOLOGIES FOR TSP and VRP
2.1 SOLUTION APPROACHES OF TSP and VRP
Solution approaches of TSP and VRP are detailed here
2.1.1 TSPThe approaches for solving the TSP range from simple
heuristics to algorithms based on the working of the human
mind (Ganesh and Narendran, 2008). Many heuristics have been
proposed to find near-optimal solutions in reasonable time
(Flood, 1956; Croes, 1958; Lin and Kernighan, 1973). Apart
from the special purpose heuristics applicable only to the
TSP, there are meta-heuristics such as Simulated Annealing,
Thershold Accepting, Genetic Algorithm, Tabu Search, and Ant
Colony Optimisation, are useful in a variety of optimisation
problems. Several authors have proposed Branch and Bound
algorithms based on the assignment problem relaxation
(Christofides et al. 1981). Heuristic approaches for TSP mostly
fall into three broad classes:
Route construction procedures Route improvement procedures and Composite procedures (Bodin et al. 1983)Route constructive procedures generate an approximate
optimal route from the distance matrix. They consist of
nearest neighbour procedure (Rosenkrantz et al. 1977), savings
procedure (Clark and Wright, 1964), insertion procedures
(including nearest insertion, cheapest insertion, arbitrary
insertion, quick insertion, greatest angle insertion),
minimal spanning tree approach (Christofides, 1976). Route
improvement procedure attempts to find a better route from a
given initial route, generally using branch exchange
approach. The 2-opt and 3-opt heuristics were introduced by
Lin (1965) while the k-opt procedure was presented by Lin
and Kernighan (1973). Recently for many variants of TSP,
meta-heuristics have been used for improvement of given
initial solutions (Goldberg, 1989). Composite procedures
construct a route using an existing procedure and then
attempt to find a better route using one or more of the
route improvement procedures (Russell, 1995; Baraglia et al.
2000).
2.1.2 VRPDaskin (1985) formulated a conceptual model of logistics
incorporating consumers, producer/shippers, carriers and
governments. Private and public sector problems including
vehicle routing, inventory management, fleet selection and
facility location were identified. Bodin et al. (1983)
presented a comprehensive review of VRP. Bott and Ballou
(1986) discussed various approaches to the vehicle
scheduling and routing problems and identified the
restrictions and extensions that should be incorporated in a
generalised vehicle routing and scheduling methodology.
Laporte and Nobert (1987) presented an extensive survey
which was entirely devoted to exact methods for the VRP and
gave a complete analysis of the state of the art up to the
late eighties. Other surveys on the use of exact algorithms
and heuristic methods were presented by Laporte (1992) and
Fisher (1995). Laporte (1992) surveyed some of the well-
known results related to the VRP. Useful techniques for the
general VRP have been outlined by Golden and Assad (1988),
Ball et al. (1995) and Aarts and Lenstra (1997). Many
references on meta-heuristics have been reported by Laporte
and Osman (1995), Osman and Laporte (1996) and Gendreau et al.
(1998) for applications on routing problems. Laporte et al.
(2000) presented a survey of heuristics divided into two
parts, classical and modern heuristics. The first part
contains well-known schemes such as the savings method, the
sweep algorithm and various two-phase approaches. The second
part is devoted to TS heuristics which has been proved to be
successful. Tarantilis et al. (2005) surveyed the recent
research efforts on meta-heuristic solution methodologies
for the standard version of the VRP.
Literature abounds with the solution approaches for TSP
and VRP. These can be classified as follows (Figure 2.1)
(Ganesh et al. 2007a):
1. Mathematical Modeling
2. Heuristics
3. Meta-heuristic
4. Interactive approaches
5. Hybrid Approaches
2.2 MATHEMATICAL MODELLING
This approach one is inclined to think, is capable providing
exact solutions. Unfortunately, even for a TSP and VRP of
modest size, it is computationally too complex to solve.
Branch and Bound algorithms (Radharamanan and Choi,
1986; Laporte et al. 1987; Ralphs, 2003), Branch and Cut
algorithms (Bard et al. (1998); Lysgaard et al. (2004); Fukasawa
et al. (2006) Dynamic programming (Magnanti, 1981; and Kolen et
al. 1987) and Lagrangean relaxation procedure (Kallehauge et al.
2001; Magnanti, 1981; Stewart and Golden, 1984) are some of
the exact approaches that have been used for solving VRPs.
TSP was also approached with Branch-and-bound by Diderich and
Gengler (1996), Cotta et al. (1995) and Tschöke et al. (1995).
NeuralNetwork Ant
Colony
ParticleSwarm
LagrangeanRelaxation
DynamicProgramming
Tabu
SimulatedAnnealing
GeneticAlgorithm
s
Branch & Bound
MathematicalProgramming
PreferenceBased Approach
Intuitive
Simulation
Graphics
Meta -Heuristic
s
ExactTechniques
Heuristics Interactive TSP & VRP
MemeticAlgorithm
s
Figure 2.1 Solution Methodologies of TSP and VRP
SavingsProcedure
InsertionProcedure
Improvement(2-opt, 3-
Cluster-FirstRoute-Second
Route-FirstCluster-Second
Combination Space-Filling
Heuristic
Cross-EntropyMethod
2.3 HEURISTICS
As in the case of other combinatorial problems, heuristics
procedures are widely used for solving TSP and VRP.
Heuristics limit their exploration of the search space but
aim at producing a good solution in a reasonably short time.
Funke et al. (2005) provided a review of both classical and
modern local search neighborhoods for VRP. Ball (2011) has
made a broad survey on the heuristics based on mathematical
programming models and methods.
Three basic categories of heuristics have been proposed
for solving the VRP (Laporte, 1992)
1) Constructive heuristics
a) Nearest neighbour
b) Savings procedure
2) Two-phase heuristics
a) Cluster-first, route-second procedure
b) Route first, cluster-second procedure
3) Local search improvement heuristics
a) Insertion procedure
b) Improvement procedure
2.3.1 Constructive Heuristics
Constructive heuristics use the data of a problem to
construct a solution step by step. (Tarantilis et al. 2005)
stated that no solution is obtained until the procedure is
complete and a special constructive approach is the greedy
method, where, at each step, the node selected for insertion
in a route is based on the lowest contribution to cost.
2.3.1.1 Nearest Neighbour
It starts with any node at the beginning, finds the closest
to the last-added node at each step and completes the route
with all the nodes included. It requires high computational
time (Rosenkratz et al. 1977).
2.3.1.2 Savings procedure
This procedure builds a solution, which may be infeasible,
by calculating savings generated by a new route
configuration. But the savings procedure may produce sub-
optimal routes (Clark and Wright, 1964; Dror and Trudeau,
1986; Altinkemer and Gavish, 1991; Reimann et al. 2004). A
robust enhancement to the Clarke-Wright CW savings
formulation was proposed by Doyuran and Çatay (2011).
2.3.2 Two-Phase Heuristics
Laporte et al. (2000) indicated that the two-phase heuristics
cluster the nodes into feasible routes and then constructs
actual routes, using feedback loops between these two
stages. Two-phase heuristics do not employ a unified
approach for combining the clustering and routing phases;
each of the proposed methods uses a unique recursive loop
for implementing the two-phase method and obtaining the best
possible results.
2.3.2.1 Cluster-first, route-second procedure
Cluster the nodes, and determine feasible routes for each
cluster -- this is the principle of this approach. It is
difficult to use this procedure when vehicles have different
capacities (Renaud and Boctor, 2002). Some of the methods
are sweep algorithm (Gillet and Miller, 1974) and Fisher and
Jaikumar Algorithm (Fisher and Jaikumar, 1981). Hiquebran et
al. (1993) applied Cluster-First Route-Second Algorithm for
VRP. A recent approach can be seen with Bräysy and Hotokka
(2007).
2.3.2.2 Route first, cluster-second procedure
This starts with a large route, which is often infeasible,
and partitioned it to smaller clusters (Beasley, 1983; Mole
et al. 1983; Hachicha et al. 2000). This approach is not suitable
for small problems.
2.3.3 Local search improvement heuristics
(Tarantilis et al. (2005) stated that the local search
improvement heuristics are iterative search procedures that
start from an initial feasible solution (often the result of
a constructive heuristic), and then progressively improve
the solution by applying a series of local modifications
called moves. At each iteration of a local search heuristic,
the moves applied to the current solution, define a set of
neighbouring solutions in the search space. The simplest
neighbourhood structures for the VRP involve moving, within
each iteration cycle, a single node from its current route
and then the selected node is inserted in the same route or
in another route with sufficient residual capacity. An
important feature of neighbourhood development is the way in
which insertions are performed: one could use random
insertion or insertion at the best position in the target
route; alternatively, one could use more complex insertion
schemes that involve a partial re-optimisation of the target
route (Gendreau et al. 1997). Zeng et al. (2005) also proposed an
assignment-based local search method for VRP.
2.3.3.1 Insertion procedure
This procedure builds a solution by determining the least
expensive insertion of a node into a route. It may end up
with sub-routes. Some of the examples are nearest
insertion, cheapest insertion, farthest insertion, quick
insertion and the convex hull insertion algorithms (Chung
and Norback, 1991; Gendreau et al. 1992; Foisy and Potvin,
1993). The push-forward insertion heuristic (Jee 2000,
Ghoseiri and Ghannadpour 2009) and virtual vehicle heuristic
(Kilby et al. 1997) are two useful heuristics for search
initialisation in solving difficult VRPs.
2.3.3.2 Improvement procedure
Given a route, the algorithm examines all the routes that
are neighbouring to it and tries to find a short route.
Starting with a small initial route, chosen arbitrarily or
by some other method, if there is no neighbouring route
which is shorter than the original one, the process stops.
This modifies the routes step by step and maintains
feasibility of the solution. It requires long computational
time (Psaraftis, 1983; Sule et al. 1991; Mak and Morton, 1993;
Bianchi, 2005). Some of the examples are 2-opt method, 3-opt
method (Lin, 1965, Alfa et al, 1991), Lin-Kernighan algorithm
(Lin and Kernighan, 1973, Papadimitriou (1992) and Helsgaun
(2000) and Or-opt (Or, 1976; Taillard et al. 1997). Kytöjoki et
al, (2007) presented an efficient variable neighborhood
search heuristic and Goel and Gruhn (2008) proposed
iterative improvement approaches based on the idea of
changing the neighbourhood structure during the search for
near optimal solutions for VRP.
2.4 META-HEURISTICS
Gendreau and Potvin (2005) highlighted that the emergence of
metaheuristics for solving difficult combinatorial
optimisation problems is one of the most notable
achievements of the last two decades. Notable among them are
the emergence of TSP and VRP.
The most promising and effective solution methods for
the TSP and VRP are meta-heuristics (Gendreau et al. 2002),
which are general-purpose mechanisms for solving hard
optimisation problems. In meta-heuristics, the emphasis is
on performing a deep exploration of the most promising
regions of the solution space. These methods typically
combine sophisticated neighbourhood search rules, memory
structures, and recombination of solutions. The quality of
solutions produced by them is usually much higher than those
obtained by classical heuristics. Nevertheless the increased
computing time is the only problem. The procedures are
usually context-dependent and require finely tuned
parameters for effective search. Each meta-heuristic has one
or more adjustable parameters. This permits flexibility, but
for any application to a specific class of problems,
requires careful calibration on a set of numerical instances
as well as testing on an independent set of instances
(Ganesh and Narendran, 2008). Meta-heuristics are classified
as memory-less and memory-based, according to the use of
previously exploited areas of the solution space (Blum and
Roli, 2003). Gendreau and Potvin (2005) provided an account
of the most recent developments in the field metaheuristics
and identified some common issues and trends with respect to
VRP. nevertheless
2.4.1 Memory-less meta-heuristics
The two most important memory-less meta-heuristics are
Simulated Annealing (SA) (Kirkpatrick et al. 1983) and
Threshold Accepting (TA) (Dueck and Scheuer, 1990).
2.4.1.1 Simulated annealing (SA)
SA is inspired from the physical annealing process emanating
in statistical mechanics. It is a local search meta-
heuristic, in the sense that it conducts local search while
guiding the overall exploration process intelligently,
offering the possibility of accepting, in a controlled
manner, solutions that do not descend along the path of
search. This feature allows SA to escape from a low quality
local optimum (Malairajan et al., 2009). Osman (1993) proposed
an SA algorithm whose neighbourhood structure uses λ-
interchanges, in which exchanges of up to λ nodes between
two routes take place. Lin et al, (2006) applied SA for
capacitated VRP. Janakiram et al, (1996) and Pepper et al,
(2002) used annealing based heuristics to solve TSP.
2.4.1.2 Threshold accepting (TA)
TA is a modification of the SA. Specifically, it leaves out
the stochastic element in accepting worse solutions by
introducing a deterministic threshold. During the
optimisation process, the threshold level is gradually
lowered like the temperature in SA (Tarantilis et al. (2005).
Tarantilis et al. (2002a) presented a variant of a TA
algorithm, called Backtracking Adaptive Threshold Accepting
(BATA), in which the neighbourhood structure is defined in a
similar way to Osman’s (1993). Tarantilis et al. (2002b)
developed another variant of TA, called List-Based Threshold
Accepting (LBTA). LBTA expands the standard TA algorithm by
introducing a list of threshold values. Both BATA and LBTA
have been used in numerous real-life distribution operations
involving transportation of goods and materials (Tarantilis
and Kiranoudis, 2001b; Tarantilis and Kiranoudis, 2002a;
Tarantilis and Kiranoudis, 2002b; Tarantilis and Kiranoudis,
2002c; Tarantilis and Kiranoudis, 2002e).
Bräysy et al, (2003) pioneered TA for VRP with Time
Windows (VRPTW). Tarantilis et al, (2004) presented a
backtracking adaptive TA algorithm, for solving the
heterogeneous fixed fleet VRP. Nikolakopoulos and Sarimveis
(2007) proposed TA method, enhanced with intense local
search, while the candidate solutions are produced through
an insertion heuristic scheme. Liu (2007) developed the
hybrid scatter search by incorporating the nearest neighbor
rule, TA and edge recombination crossover into a scatter
search conceptual framework to solve the probabilistic TSP.
2.4.2 Memory-based meta-heuristics
Memory based meta-heuristics exploit the previously examined
area of the solution space through one or multiple lists of
solutions kept within a limited memory. The term memory was
used explicitly by the Tabu Search (TS) algorithms (Glover,
1989) and the Adaptive Memory Based Algorithms (AMBA)
(Rochat and Taillard, 1995). However, a number of other
meta-heuristics such as Genetic Algorithm (Holland, 1975),
Ant Colony Optimisation (Dorigo, 1992), Particle Swarm
Optimisation (Kennedy and Eberhart 1995) and Memetic
Algorithm (Moscato 1989), use mechanics and structures that
can be considered as memories. The literature pertaining to
routing problems of each major group of memory based meta-
heuristics is summarised here.
2.4.2.1 Tabu Search (TS)
TS is a local search meta-heuristic. TS explores the solution
space by moving at each iteration, from a current solution
to the best solution in its neighbourhood or in a subset of
neighbourhood for computational efficiency (Tarantilis et al.
(2005).
Osman (1993) presented one of the earliest successful
implementations of TS for solving VRP. Taillard (1993)
developed one of the most effective TS algorithms for
solving VRP. Gendreau et al. (1994) proposed a sophisticated TS
algorithm, called Tabu-route. Zachariasen and Dam (1995)
proposed TS for the Geometric TSP.
The TS in Rochat and Taillard (1995) exploits a
neighborhood based on the exchange of customers between
routes. Carlton (1995) described a reactive TS that
dynamically adjusts its parameter values based on the
current search status. Potvin et al. (1996) described a
standard TS heuristic based on 2-opt and Or-opt exchanges.
Xu and Kelly (1996) presented an approach for defining the
neighbour structure of a TS algorithm. Rego and Roucairol
(1995) and Rego (2001) developed a TS that defines its
neighbourhood structure by employing node-ejection chains.
Toth and Vigo (2003) developed a TS algorithm, called
Granular TS, based on the idea of excluding non-promising
areas of the search space. Cordeau et al. (2002) proposed a TS
algorithm, called Unified TS Algorithm (UTSA) which shares
some common features with Taburoute. Bräysy and Gendreau
(2002) applied TS for the VRPTW. Chiang and Russell (1997),
Cordeau et al. (2001), De Backer et al. (2000), Rochat and
Taillard (1995) and Taillard et al. (1997) have also made
levelheaded contributions to VRP using TS. Paraskevopoulos et
al, (2008) solved the heterogeneous fleet VRPTW using a two-
phase solution framework based upon hybridised TS, within a
new Reactive Variable Neighborhood Search metaheuristic
algorithm. Potvin and Naud (2011) proposed a tabu search
heuristic with a neighbourhood structure based on ejection
chains to solve VRP.
2.4.2.2 Adaptive Memory-Based Algorithms (AMBA)
The first AMBA was presented by Rochat and Taillard (1995).
The adaptive memory rationale constitutes one of the most
powerful tools for automatic diversification and
intensification of the search process (Tarantilis et al.
(2005).
Rochat and Taillard (1995) introduced the concept of
adaptive memory for the VRP, according to which a set of
high quality VRP solutions is stored in a pool that is
dynamically updated during the search process. Tarantilis
and Kiranoudis (2002d) developed an AMBA called BoneRoute.
Tarantilis (2005) presented a modified BoneRoute, called
SEPAS, generating an initial population of diversified
solutions in a systematic way. Tarantilis et al (2005) studied
Capacited VRP with AMBA. Tarantilis and Kiranoudis (2007)
applied a flexible AMBA for real-life transportation
operations. Derigs and Reuter (2009) presented results on an
implementation of the attribute-based hill climber heuristic
to the open VRP which is a parameter-free variant of the
tabu search principle. Zachariadis et al., (2010) introduced
an Adaptive Memory for the algorithmic framework for the VRP
with Simultaneous Delivery and Pickup (VRPSDP) which collects
and combines promising solution features to generate high-
quality solutions.
2.4.2.3 Genetic Algorithm (GA)
GAs are population-based algorithms that simulate the
evolutionary process of species that reproduce. GA causes
the evolution of a population of individuals encoded as
chromosomes by creating new generations of offspring through
an iterative process that continues until some convergence
criteria are met. At the end of this process, it is expected
that an initial population of randomly generated chromosomes
will improve and be replaced by better off-springs. The best
chromosome obtained by this process is then decoded to
obtain the solution (Holland, 1975).
Literature on the application of GA to VRP is limited.
Baker and Ayechew (2003) presented a GA for solving the VRP.
The representation of this algorithm was influenced by the
non-binary representation of Chu and Beasley (1997) for the
Generalised Assignment Problem. Xiong et al, (1998) modeled
rolling batch planning as VRPTW and used GA and heuristics
to solve the problem. Prins (2004) developed an effective GA
in which a VRP solution is represented as a sequence in
which a vehicle must visit all clients, assuming the same
vehicle performs all the trips by turn. Hanshar and Ombuki-
Berman (2007) applied GA for the Dynamic VRP. Salhi and
Petch (2007) proposed a hybrid GA, which used a new non-
binary chromosome representation and which is enhanced by a
domain specific data structure, appropriate genetic
operators and a scheme for chromosome evaluation for the VRP
with multiple trips. Prins (2008) proposed a simple and
effective GA for the VRP. Ho et al, (2008) proposed two hybrid
GAs for the multi depot VRP. In the first one the initial
solutions are generated randomly and in the second one the
Clarke and Wright saving method and the nearest neighbour
heuristic are incorporated into hybrid GA for the
initialisation procedure and found that the performance of
the second one is superior in terms of the total delivery
time.
Literature on the application of GA to TSP is in a good
number. GA was explored by Ulder et al, (1991) and Potvin
(1996) for TSP; Cotta et al, (1995) Hybridising with Branch
and Bound Techniques for TSP; Kureichik et al, (1997) used GA
with new features to TSP against Premature Convergence;
Nagata and Kobayashi (1997) used a high-power GA for TSP.
Schmitt and Amini (1998) developed GA and evaluated by
solving 5000 TSPs; Ochi et al, (1998) presented a new hybrid
metaheuristic which uses Parallel GA and Scatter Search
coupled with a decomposition-into-petals procedure for
solving a class of Vehicle Routing and Scheduling Problems.
While addressing TSP Tesfaldet and Hermosilla (1999)
used a Lamarckian GA; Larranaga et al, (1999) and Andal
Jayalakshmi et al, (2001) used a Hybrid GA. Hwang (2005)
proposed GA for VRP and TSP as well; Pankratz (2005) proposed
a Grouping GA for solving the pickup and delivery problem
with time windows which features a group-oriented genetic
encoding in which each gene represents a group of requests
instead of a single request. Snyder and Daskin (2006)
presented an effective heuristic which combines GA with a
local tour improvement heuristic applied to generalised TSP
and termed as random-key GA. Marinakis et al, (2007)
formulated a new bi-level formulation for VRP and tested
with a solution method using GA and also tested it with TSP.
Snyder and Daskin (2006) proposed a random-key GA for
the generalised TSP. Samanlioglu et al, (2007) arrived at
approximate and sometimes optimal solutions to the symmetric
TSP using a hybrid approach that combines a Random-Key GA
with a local search procedure. Bae et al, (2007) developed an
integrated VRP model using heuristic method and the improved
GA of which operators and initial population are improved.
This was tested for TSP Problems too. Marinakis et al, (2007)
proposed a bi-level GA for both VRP and TSP. Ganesh and
Narendran (2007) proposed a multi-phase constructive
heuristic that clusters nodes based on proximity, orients
them along a route using shrink-wrap algorithm and allots
vehicles using generalised assignment procedure and employed
GA for an intensive final search.
2.4.2.4 Ant Colony Optimisation (ACO)
ACO is one more memory-based meta-heuristic, which simulates
the ant’s ability in determining the shortest path between
food and the nest (Tarantilis et al. 2005).
Bullnheimer et al. (1998) presented an application of Ant
System (AS) for solving the VRP. Reimann et al. (2002)
developed an AS algorithm based on the transformation of the
simultaneous route construction mechanism proposed by Clarke
and Wright (1964) into a rank-based AS. Reimann et al. (2004)
proposed an AS approach, called D-Ants, built on the
algorithm developed by them earlier (Reimann et al. 2002).
Bell and McMullen (2004) applied ACO to a set of VRPs.
ACO was employed by Mazzeo and Loiseau (2004) for the
capacitated VRP, Manfrin (2004) and Reimann et al, (2004) for
VRP and Lin and Cai (2006) for VRP and mail delivery
problems. Montemanni et al, (2005) proposed a dynamic VRP,
based on the Ant Colony System paradigm. Li and Tian (2006)
presented an ant colony system hybridised with local search
for solving the Open VRP. Reimann and Ulrich (2006) compared
the backhauling strategies in VRP using ACO. Rizzoli et al,
(2007) proposed ACO for realworld VRP. Lee et al, (2008)
proposed an enhanced ACO for capacitated VRP.
Fuellerer et al (2009) approached VRP with an ACO
algorithm which combines two different heuristic measures
(with respect to loading and routing) within one pheromone
matrix. Yu et al. (2009) proposed an improved ACO, which
possesses a new strategy to update the increased pheromone,
called ant-weight strategy, and a mutation operation, to
solve VRP. Gajpal and Abad (2009) used a multi-ant colony
system to solve VRPB.
Dorigo and Gambardella (1997) proposed an ACO for the
TSP. Binachi et al. (2002) proposed an ACO approach to the
Probabilistic TSP. Stüzle and Dorigo (1999) and Hung et al,
(2007) proposed the ACO for TSP. Li and Gong (2003) proposed
a Dynamic ACO for TSP. Branke and Guntsch (2004) showed that
ACO works well even when only an approximate evaluation
function is used, which speeds up the algorithm, leaving
more time for the actual construction and applied it for the
Probabilistic TSP. Qingbao and Lingling (2007) analysed of
the convergence of ant colony with the TSP. Liu (2005)
applied Rank-based ACO applied to dynamic TSP. Yang et al
(2008) focused on the generalised TSP with ACO and to avoid
locking into local minima, a mutation process and a 2-opt
local searching technique are also introduced. Donati et al
(2008) proposed a multi ant colony for VRP. Çatay (2010)
proposed an ACO employing a new saving-based visibility
function and pheromone updating procedure. Yu et al., (2011)
presented an improved ACO with coarse-grain parallel
strategy, ant-weight strategy and mutation operation for the
multi-depot vehicle routing problem.
2.4.2.5 Particle Swarm Optimisation (PSO)
PSO is also one of the population based stochastic
optimization techniques which is inspired by the social
behavior of bird flocking or fish schooling. This technique
searches a space by adjusting the trajectories of individual
vectors, called “particles”, conceptualized as moving points
in multidimensional space. The individual particles are
drawn stochastically on the basis of the positions of their
own previous best performance and the best previous
performance of their neighbours. The application of this
technique to routing problems emerged recently.
Wang et al. (2006) proposed a novel real number encoding method
of PSO for an Open VRP. In addition they also applied
several heurist methods into the post-optimisation
procedure, such as Nearest Insertion algorithm, GENI
algorithm, and 2-Opt, after decoding. Teodorović (2008)
presented a classification and analysis of the Swarm
intelligence systems for the Transportation problems. The
techniques include ACO, PSO, Bee colony optimisation and
stochastic diffusion search. Belmecheri et al., (2010)
proposed a PSO to solve the VRP with Heterogeneous fleet,
Mixed Backhauls, and time windows.
Ai and Kachitvichyanukul (2009) presented a PSO
algorithm for solving a VRPSDP. The formulation is a
generalisation of three existing VRPSDP formulations. Li et al,
(2008) proposed PSO for the Electronic design automation
modeled as TSP. Shi et al. (2007) presented a PSO based
algorithm for the TSP. An uncertain searching strategy and a
crossover elimination technique were used to accelerate the
convergence speed. They also proposed another PSO-based
algorithm applied to solve the generalised TSP by employing
the generalised chromosome. Two local search techniques were
also used to speed up the convergence. Onwubolu and Clerc
(2004) solved the optimal path problem for automated
drilling operations by a new heuristic approach using PSO.
Wang et al, (2003) also proposed PSO for the TSP. Contribution
of Sofge et al. (2002), Secrest (2001) and Secrest and Lamont
(2001) in solving TSP with PSO are also significant.
2.4.2.6 Memetic Algorithm
Memetic Algorithm was introduced in the late 80s to denote a
family of meta-heuristics that has central theme as
hybridisation of different algorithmic approaches for a
given problem. Special emphasis was given to the use of a
population-based approach in which a set of cooperating and
competing agents were engaged in periods of individual
improvement of the solutions while they sporadically
interact (Moscato and Cotta 2003).
Prins and Bouchenoua (2005) tried Memetic Algorithms
for solving the VRP, Capacitated VRP and General Routing
Problems with Nodes, Edges and Arcs. Contribution of Lacomme
et al. (2004) and Belenguer et al. (2006) is also notable in
investigating vehicle routing with memetic algorithms. Lima
et al. (2004) described a Memetic Algorithm for the
Heterogeneous Fleet VRP.
Tavakkoli-Moghaddam et al. (2006) proposed a memetic
algorithm which uses different local search algorithms. To
make use of the power of memetic algorithm, inter and intra-
route node exchanges were also used as a part of their
evolutionary algorithm. Fallahi et al. (2008) proposed a
memetic algorithm with a post-optimisation phase based on
path relinking and TS method. Créput and Koukam (2009)
studied the hybridisation of the self-organising map in an
evolutionary algorithm to solve the Euclidean TSP which is
considered to be memetic neural network algorithm.
Moscato and Norman (1992), Moscato and Tinetti (1992),
Buriol et al, (2004) and Gutin et al, (2008) approached TSP with
a memetic approach. Krasnogor and Smith (2000) introduced a
Memetic Algorithm with Self-Adaptive Local Search. Merz and
Freisleben (2001) proposed Memetic Algorithms for the TSP.
Merz (2002) compared several memetic algorithms,
incorporating local search methods.
Aarts and Verhoeven (1997) provided some test results
using Memetic Algorithms for the TSP with 2-opt (Lin, 1965)
and variable depth neighborhoods (Lin and Kernighan, 1973)
as local search techniques. Larranaga et al, (1999) studied
various representations and operators used in GA for solving
TSP. They presented crossover and mutation operators to
tackle the TSP with GA having different representations such
as binary representation, path representation, adjacency
representation, ordinal representation and matrix
representation. Liu et al. (2006) proposed an effective PSO
based Memetic Algorithm for the TSP. In this, a novel
encoding scheme was developed and an effective local search
based on SA with adaptive meta-Lamarckian learning strategy
was proposed and incorporated into PSO. Buriol et al. (2004)
proposed the algorithm for the Asymmetric TSP. Duan and Yu
(2007) addressed TSP with Hybrid ACO using Memetic
Algorithms. Créput and Koukam (2008) presented an extension
of the self- organising map by embedding it into the memetic
algorithm for VRP.
2.5 INTERACTIVE APPROACHES
These are simple approaches that can be tailored to suit a
particular application. It can be based on intuition,
simulation, preference or some type of graphics to aid the
decision maker in a ‘what if’ mode. This can be called as
quick and dirty procedure (Doll, 1980; Cullen et al. 1981; Hill,
1988; Potvin and Rousseau, 1994; Nussbaum et al, 1997; Hwang,
1999; Du, et al. 2007).
Hurrion (1980) described a visual interactive method of
improving solutions for the TSP. Wu and Liou (1993) devised a
mechanism to include the elastic ring in the Potts neural
encoding approach to solve the TSP. Baker and Carreto (2003)
described a graphical-user-interface and a heuristic based
on a greedy randomised adaptive search procedure which was
developed to work in combination to tackle the basic VRP.
2.6 HYBRID APPROACHES
Analysts have also attempted hybrid approaches, combining
two or more of those suggested in the preceding paragraphs.
Some of these are reported to have a high potential to
provide good solutions at low computational time (Laporte et
al. 2000).
Given the nature of the combinatorial optimisation
problem, a lot of approaches have been made for variants of
both VRP and TSP and have been discussed all the way through
the review of the variants. Nevertheless, to be precise a
few of the recent works are due to Liu et al. (2006), who
combined PSO with Memetic Algorithms and proposed it for TSP.
They also developed a novel encoding scheme and an effective
local search based on SA with adaptive meta-Lamarckian
learning strategy and incorporated into PSO. Tam and Ma
(2004) combined Guided local search, TS and SA and applied to
solve VRPTW. Liu et al. (2006) proposed an effective PSO based
Memetic Algorithm for the TSP. Fang et al, (2007) proposed a
hybrid algorithm which integrates PSO with simulating SA to
solve the TSP. Duan and Yu (2007) proposed a hybrid ACO
Using Memetic Algorithms for TSP. Tiejun et al, (2008)
proposed a hybrid new method named multi-agent approach
based on GA and ACO to solve the TSP. Perboli et al, (2008)
presented a hybrid algorithm based on GA and TS for solving
the Capacitated VRP.
Repoussis (2010) proposed a hybrid evolution strategy
for the open VRP which manipulates a population of ‘μ’
individuals using a (μ+λ)-ES. At each generation, a new
intermediate population of λ offspring is produced via
mutation, using arcs extracted from parent individuals. The
selection and combination of arcs is dictated by a vector of
strategy parameters. A multi-parent recombination operator
enables the self-adaptation of the mutation rates based on
the frequency of appearance of each arc and the diversity of
the population. Each new offspring is further improved via a
memory-based trajectory local search algorithm, while an
elitist scheme guides the selection of survivors.
2.7 APPLICATIONS OF TSP and VRP
The following are some examples of the multitude of TSP and
VRP applications in manufacturing and service sectors:
Routing of automated guided vehicles which are considered
as one of the most appropriate modes for material handling
in contemporary flexibly automated production environments
(Reveliotis, 2000).
Minimisation of the distribution costs in a multi-facility
production system (Dhaenens-Flipo, 2000).
Determination of vehicle routes for material delivery
within the premises of a plant operating under a Just-In-
Time philosophy (Vaidyanathan et al, 1999).
Sequencing of the operations in single or multi-feeder
printed circuit board manufacturing unit (Altinkemer et al.
2000).
Rolling batch planning (Xiong et al.1998).
A few real-life examples (Ganesh et al. 2007a), which are
variants of the classical VRP, are listed below:
Upper limit on the time of delivery - e.g., Milk and
Newspaper
Barred time windows - e.g., Urban solid waste removal
Conflicting/competing time windows - e.g., Mobile catering
Combined routing and scheduling - e.g., Mobile hospital,
Mobile court
Independent multiple depots
Interdependent multiple depots
A diverse application of TSP and VRP with different set of
variants is listed in Table 2.1 and some of the applications
are explained in detail.
2.7.1 Multiple Interdependent Depot VRP (MIDVRP)
In organisations with more than one depot, it is often the
case that each depot is sovereign, with its own fleet of
vehicles and its own geographical customer area to serve. In
such cases, the organisation would simply face a number of
similar single-depot VRPs. In other cases depot operations
are interdependent and vehicles leaving one depot may, after
delivering to customers, end up at another depot (Marinakis
and Migdalas 2007). These problems are called multiple
interdependent-depot-VRP (MIDVRP). Some of the examples are:
Garbage collection in a crowded city, special buses arranged
by the government during functions, etc.
2.7.2 Multiple Commodities VRP (MCVRP)
In some cases, the vehicles are partitioned so that
different commodities are stored in segregated compartments.
Each customer may specify the required quantities of various
commodities. This characterises a multi-commodity-VRP
(MCVRP).
2.7.3 Vehicle Scheduling Problem (VSP)
Vehicle scheduling problems can be thought of as routing
problems with additional constraints imposed by time periods
during which various activities may be carried out (Bodin et
al. 1983).
Some of the constraints which make VSP are:
1. The length of the time that a vehicle may be in
operation before it must return to the depot for
service or refueling.
2. The fact that certain tasks can only be carried out by
certain vehicle types.
3. The presence of a number of depots where vehicle may be
housed.
Example: Grocery store distribution.
2.7.4 School Bus Routing and Scheduling Problem (SBRSP)
In the SBRSP there are a number of schools, wherein each is
assigned a set of bus stops, with a given number of students
assigned to each stop, and time windows for the delivery and
pick-up of the students (Bodin et al. 1983). The problem is to
minimise the number of buses used and total transportation
costs while serving all the students and satisfying all the
time windows.
2.7.5 Routing and Scheduling with Full Loads and Time Window
(RSFLTW)
In the problem of RSFLTW, a set of demands is specified for
a set of origin-destination pairs like transportation
problem. Each demand is a full load that must be loaded onto
a vehicle at an origin and unloaded at a destination. These
stops must satisfy pre specified time window constraints and
the aim is to design routes and schedules for the fleet of
vehicle. (Bodin and Golden 1981). The objective can be
minimising the number of vehicles and the total distance
travelled.
2.7.6 Newspaper Distribution VRP
Newspaper distribution problem of a newspaper company is a
complex one. A newspaper company will have printing centres
where newspapers are printed and distributed to local
distribution centres everyday. Newspapers are delivered to
subscribers from the local distribution centres. Newspapers
should be delivered sufficiently early before the readers
leave for their work places. If there are frequent late
deliveries, readers would consider switching their
subscriptions to other newspapers. Timeliness is one of the
most important requirements in newspaper delivery. However,
since newspapers need to be printed as late as possible in
order to contain the most up-to-date news, there is usually
only a very short time available between printing and
delivery. Although the newspaper company tries to minimise
the printing time and to increase the printing capacity
nowadays, there are still chances of being late without an
efficient delivery scheme. Delivery plan is needed to
deliver newspapers to as many local distribution centres as
required using minimum number of vehicles in the stipulated
time (Ree and Yoon, 1996).
2.7.7 Recyclable Material Collection VRP
Large university campuses may face the problem of collecting
the waste / refuse produced by offices, classrooms,
laboratories, etc. This problem becomes especially complex
when there are many buildings spread out over a large area
(Bommisetty et al. 1998). In a given five-day work week, the
vehicles are needed to visit various buildings in order to
collect the waste in a prescribed pattern. The problem is to
minimise the distance travelled by the vehicle, the vehicle
and the collection time. The constraints could be required
collection frequency, number of vehicles, and volume in
terms of bins, vehicle capacity and time constraints.
2.7.8 Earthquake - Food Distribution VRP
Distribution of food to earthquake affected area is a
complicated distribution problem. This is because in general
there is neither an inventory system in place nor are their
plans for transporting goods to these regions. There is a
need to determine optimal patterns of food supply and
inventory allocation for earthquake affected area. It is
also essential to formulate the VRP incorporating inventory
allocation and the optimal distribution based on minimising
the amount of pains and suffering of the affected people
instead of travel distance. Time is also the main factor to
cover almost many affected areas. A longer route with fewer
vehicles and a high efficiency of supply is more attractive
proposition to this kind of problem (Hwang 1999).
2.7.9 Perishable Food Distribution VRP
Managing perishability of foods is a difficult problem in
distribution management. In food industries perishability
arises for products like milk, vegetables and meat. The time
period between preparation date and delivery of these
products is of a major interest for both producers and
retailers. Inefficient distribution process will cause a
serious decline in quality, loss of sale efficiency,
increase in storage cost, and decrease in order volume. The
problem is to find a set of optimal routes for heterogeneous
capacity of vehicles, with the constraints on number of
vehicles, capacity of vehicles, waiting and servicing time
for each customer, and the distance traveled (Tarantilis and
Kiranoudis, 2001b).
2.7.10 Bank Cheque Collection VRP
In the case of bank cheque collection, the problem is to
determine the routes and departure date / day and times for
vehicles from the head office to distribute the cheque and
other forms and to retrieve the cheques that accumulate at
the retail branch locations. The vehicle will return back to
head office after all collections and distributions are
over. The complication of the problem is to predict the
retrieval volumes. When the schedule result in delayed
retrieval, some cheques miss the deadlines and resulting in
handling fees. If the retrieval is too early, then there
will be large accumulations at the cheque processing
facility centre. (Anbuudayasankar et al 2008).
2.7.11 Vehicle Routing for Transporting Hazardous Material
The transportation of hazardous materials evolve greater
amount of risk. The problem becomes more acute because of
the increase in the quantity of such materials to be
transported these days. One way to overcome this risk is to
take appropriate routing decisions which can lead to the
determination of alternative routes, incase needed with
respect to truck driver’s decisions. The problem is to find
a risk minimised routes with the constraints of distance
travelled and time. The precedence can be given to truck
driver’s preference. (Tarantilis and Kiranoudis, 2001a;
Zhang et al 2005)
2.7.12 Share Auto VRP
The tendency of the people to engage share autos to reach
their destinations rather than using public transport system
increases the load on the environment and raises issues
about the quality of life. The use of share autos in the
main road of the cities creates high level of air pollution,
parking problems, noise pollution and traffic congestion.
So, designing alternative routes for the share autos with
less travel distance is the objective. The preferences of
the public and norms of the government should be taken care
of while designing the routes.
Table 2.1 Diverse Applications of TSP and VRPSl.No Application Variant Literature
1 School Bus Routing
Multi objective VRP and TSP Bowerman et al. (1995)VRP Braca et al. (1997)Multi objective VRP and TSP Li and Fu (2002)Open VRP with Time Dead Lines Ozyurt et al. (2005)VRP with Coupled Time Windows Fugenschuh (2006)VRP Park and Kim (2010)
2 Food Distribution
VRP with Time Windows (VRPTW) Hsu and Feng (2003)Ioannou et al. (2001)
VRPEvans and Norback (1985)Prindezis et al. (2003)Johansson (2006)
TSP with Time Windows Bräysy et al. (2009a)
3 Waste Collection Problem
VRP with Crew Constraints Angelelli and Speranza (2002)
VRP Tung and Pinnoi (2000)Nuortio et al. (2006)
VRPTW Kim et al. (2006)
4 Milk Collection and Distribution
VRP Boldon et al. (1996)Two-Period TSP Butler et al. (1997)
VRP Tarantilis and Kiranoudis (2000)
VRP with Heterogeneous Fixed Fleet Tarantilis et al. (2004)
VRPTW Marshall et al. (2006)
Periodic VRP Claassen and Hendriks (2007)
5 Container Transport
VRP
Cattrysee et al. (1996)Kim and Kim (1999)Koo et al. (2004)Imai et al. (2007)Stahlbock and Voβ (2008)Bandeira et al. (2009)
Period VRP Baptista et al. (2002)VRPTW Lee et al. (2003)VRP and TSP Steenken et al. (2004) Capacitated VRP Karlaftis et al. (2009)Multi-traveling salesman problem with time windows (m-TSPTW)
Zhang et al. (2009)
6 City Logistics Dynamic VRP Taniguchi et al. (2000)Montemanni et al. (2005)
Multi Depot VRPTW (MDVRPTW) Chiu et al. (2006)
VRPTarantilis and Kiranoudis(2002)Bräysy et al. (2009b)
7Company Employee Distribution Relations
VRP with Load Balancing Lee and Ueng (1999)
Table 2.1 Diverse Applications of TSP and VRP (Contd...)
8 Furniture Transport
VRP with Heterogeneous Fixed Fleet Prins (2002)
9 Public Transport VRP
Bodin (1990)Sateesh and Ray (1992)Nurcahyo et al. (2006)Lam et al. (2009)
10 Soft Drink Industry
VRP Golden and Wasil (1987)Zeng et al. (2008)
TSP with Pick up and Delivery Gendreau et al. (1999)11 Brewing Industry VRP Eibl et al. (1994)
12
Land Transportation of air cargo forwarder
VRP with Backhauls and Time Windows with Heterogeneous Fleet of Vehicles
Cheung and Hang (2003)
VRP with Time Windows (VRPTW) Kritikos and Ioannou (2010)
13 Sugar Cane Transport MDVRPTW Abel et al. (1981)
14 RetailDistribution VRP Soehodho and
werdinigngsih (2003)
15 Fresh MeatDistribution Multi Depot VRP Tarantilis and Kiranoudis
(2002a)
16 Mail Carrier (or)Mail Delivery
VRP Tarantilis et al. (2002c)Period TSP Paletta 2002
17 Ship Routing
Multi-trip VRP Fagerholt (1999)
TSP with Time Windows Fagerholt andChristiansen (2000)
TSP with Time Windows Fagerholt (2001)
18 Emergency Planning VRP with Mixed delivery andpickup Ozdaman et al. (2004)
19 Gas distribution VRP Day et al. (2009)
20 AgriculturalTransport VRP
Osvald and Stirn (2008)Bochtis and Sørensen(2009a)Bochtis and Sørensen
(2009b)
21Live Stock(Animal)Distribution
VRP Grlbkovskaia et al. (2006)
22 Post BoxCollection TSP Laporte et al. (1996)
23 Rural PostmanProblem TSP Eiselt et al. (1995)
24 PerishableProducts
VRP Federgruen et al. (1986)Hsu et al. (2007)
VRP with Time Windows (VRPTW) Chen et al. (2009)
25Analysis of thestructure ofcrystals
TSP Bland and Shallcross(1987)
26 Overhauling of gasturbine engines TSP Plante et al. (1987)
27Drilling ofprinted circuitboards
TSP Ancău (2008)
28 Material handlingin a warehouse TSP Ratliff and Rosenthal,
(1983)
29 Clustering of dataarrays TSP Lenstra and Rinooy Kan,
(1975)
30Sequencing of jobson a singlemachine
TSP Gilmore and Gomory,(1964)
2.7.13 Vendor Managed Distribution Systems
In vendor-managed systems, distribution companies estimate
customer inventory level in such a way to replenish them
before they run out of stock. Hence, demands are known
before and all customers are static. Yet, because demands
are uncertain, some customers may run out of stock and have
to be served urgently. This is a real time VRP with varying
load and constant time. (Archetti et al 2007).
2.7.14 Taxi Cab services
In taxi cab services, almost every customer is dynamic and
demand rate is usually high. Sometimes vehicles become idle
from time to time. Using the idle vehicle to meet unexpected
demand can be thought of. Directing these idle vehicles to
customers, when the scheduled vehicle is unable to reach the
customers, due to traffic jam is an important issue to be
considered. (Holly, 2007).
2.8 OBSERVATIONS AND RESEARCH GAP
A careful analysis of literature on the variants,
methodologies and applications of TSP and VRP published
hitherto reveals the following: It is evident that there is
a need to address a variety of complex variants of the
classical TSP and VRP pertaining to Balanced Logistics,
Reverse Logistics, Distribution Logistics and Emergency
Logistics. TSPs and VRPs in the real world often need to
include additional factors such as multi depot, time window,
route length, heterogeneous capacity, sequential and
simultaneous loads etc. The underlying principle of the work
is to consider additional real-time factors and constraints
in the complex routing variants and to develop solutions for
the same. The impetus for the development of the work is an
urge to provide better algorithms for the logistics decision
makers for the complex routing variants. This research
rationale is a "living document" that will continue to
change as the research evolves in terms of different
variants and solution methodology. The contribution lies in
the development of effective and competitive new / derived
(devised based on leveraging the concepts from the existing
approach or solution methodology) / combination of new and
derived, unified solution methodologies (heuristics / meta-
heuristics) for the new / less researched complex routing
variants.
The aim of the study is of two folds:
To explore and identify the new / less researched
challenging, practical, complex variants of routing
problems in the current global logistics trend based on
academic and business literature and from the interview
conducted with chief officers of global supply chain.
To propose and develop new / derived and unified
methodologies to solve the above routing problems in order
to support the decision making ability of logistics
managers.
2.8.1 Problem in Balanced Logistics
According to the council of Logistics Management, logistics
is “the process of planning, implementing and controlling
the storage of goods, services and related information from
point of origin to point of consumption for the purpose of
conforming to customer requirements”. This reveals that the
importance is not only given to strategic level (or)
tactical level but also to the operational level in which
the actual implementation of the transfer of goods from one
point to various destinations comes in to picture. The
customer satisfaction can be improved with the increase in
the level of logistics service which in turn has a major
impact on revenues, especially in markets with homogeneous
low price products where competition is based on the
promptness in delivery of the products.
The integration of inventory and transportation is also
a key aspect in reducing the total supply chain cost.
Particularly to reduce the operating cost the management
must determine the proper balance between inventory and
transportation costs. Frequent trips between depot and the
customer reduces the inventory cost but at the same time
transportation cost is high whereas infrequent trips leads
to high inventory cost and low transportation cost. In the
distribution of products from a depot to multiple customers,
assigning loads to vehicles is as important as deciding the
routes for the vehicles. In routing, the order in which the
customers are visited will determine how long the delivery
will take place and the time of returning to the depot.
There are times when inhuman weights are lifted, high
risks are taken and even acrobatics performed! But, at later
stages, the automation has invaded the in-bound logistics of
shop-floor. Set-up times have fallen drastically, precision
has improved, and quality is being talked of in the parts
per million ranges. Companies may not be interested so much
in human issues in out-bound logistics and distribution. A
burning issue today is the workload balancing between
drivers in transshipments of goods. Rarely do business
interests align with this kind of human issues. Nevertheless
it leads to the price of poor service and loss of goodwill
that may be the result of workers' fatigue. These factors
are bound to affect end-product pricing and constitute an
indirect cost that customer pays.
The original version of this problem is referred to as
multiple TSP (mTSP) in literature. The problem of finding
the minimum route length, in either time (or) distance, from
a depot through a set of customers to be served is
imperative for the competitive advantage which is known as
the classical TSP. Instead of single vehicle, if there are
multiple vehicles serving the customers, then it is the
mTSP. A generalisation of the standard mTSP is the well-
known TSP. The problem can be defined simply as the
determination of a set of routes for m salesmen, who all
start from and return to a single home city (depot).
In practice it is observed in many instances that more
than one salesman (or) vehicle starts from a single place to
serve the customers at different locations and returns to
the same place. This exactly resembles mTSP and hence it is
appropriate to model a real life situation with mTSP. Some
reported applications are Print press scheduling
(Gorenstein, 1970., Carter and Ragsdale, 2002), School bus
routing problem (Angel et al. 1972), Crew scheduling (Svestka
and Huckfeldt, 1973., Lenstra and Rinnooy Kan, 1975., Zhang
et al. 1999), Interview scheduling (Gilbert and Hofstra, 1992),
Mission planning (Brummit and Stentz, 1996, 1998, Yu et al.
2002., Ryan et al. 1998), Hot rolling scheduling (Tang et al.
2000), Design of global navigation satellite system
surveying networks (Saleh and Chelouah, 2004) etc.,
Now that it is observed that the balancing of the
workload along with the route optimization is essential. An
extension of mTSP with workload balancing (mTWB) is
developed. The objective of this model is equalising the
workload among the available drivers / vehicles with the
intention of finding out the optimal number of such entities
required to cover a set of nodes. The problem is addressed
under the conditions that all the drivers must return to the
node of origin (depot) and that every node will be visited
exactly once. This variant is applicable for both
manufacturing and service industries.
Many drivers may not object to drive more than 8 hours
a day, but they may not be happy visiting more vendors since
loading and unloading is a tiresome task. The logistics
manager is to find set of vendors for each driver for a
given day. Many drivers approach logistics manager
personally and request to allot them for fewer vendors. Now
the challenge of logistics manager is to allot a set of
vendors for each driver to balance the workload. It is also
assumed that the workload is almost equal at every vendor
location.
This variant mTWB is not a new variant. This is
already addressed by Chandran et al. (2006). The objective
is to develop new / derived / combination of new & derived
and unified solution methodologies to solve the variants
multiple traveling salesman problem mTSP and mTWB. Three
solution methodologies are proposed for the variants mTSP
and mTWB. The first heuristic is the combination of new and
derived approach. A new clustering approach is proposed and
it is combined with the derived approach of shrink-wrap
algorithm. The second heuristic is the combination of new
and derived approach, where as a new clustering approach is
proposed and it is combined with the derived approach of 2-
Opt algorithm. The third heuristic is the combination of new
and derived approach, where as a new min-max function with
prufer number concept is proposed and it is coupled with the
derived meta-heuristic process of Genetic Algorithm. The
contribution lies in the development of two heuristics
(combination of new and derived) and one meta-heuristic
(combination of new and derived) as a unified solution
methodology to solve both mTSP and mTWB.
2.8.2 Problem in Reverse Logistics
Sustainable supply chain is the management of goods from
suppliers to manufacturer / service provider and then to
customer and back. Here there is a need to reduce the carbon
footprint in the logistics. Reverse logistics is the
process of transshipment of goods / the containers from
their typical final destination to pre-destination or
origin, mainly for the reason of capturing value or for the
proper further use, re-use or disposal of the goods. Reverse
logistics is also applicable to the containers such as cans,
bottles etc. which carry the goods from the manufacturer to
the customers and supposed to bring back for re-supply of
the goods in the same container after the cleaning process.
Economic and environmental impacts are the main inspirations
for planning the reverse logistics of the channel for the
containers (Alshamrani et al. 2007).
This problem aims at returning containers to the depot
that are associated with goods / products delivered
previously on a route. The pick-up load needs to be replaced
in the place of delivery load. This is termed as constrained
capacity. The research problem is to deliver the goods to
customers and simultaneously pick-up the used containers
such as bottles, cans etc in the same vehicle in the place
of the delivered loads. So, the loads which picked from the
customer’s needs to be adjusted in the place of the load
delivered. The requirement is to route the vehicle with due
consideration to the loads involved in delivery as well as
pick up with the constrained capacity. The pick-up and
delivery should be performed simultaneously so that each
node is visited only once by the vehicle. This new variant
is termed as simultaneous delivery and pick-up problem with
constrained capacity (SDPC).
The base variant of this new variant is Traveling
Salesman Problem with Simultaneous Delivery and Pick-up
(TSPSDP) addressed by Ganesh et al. (2007b). The objective is
to develop new and unified solution methodologies to solve
the variants TSP and SDPC. Three solution methodologies are
proposed for the variants TSP and SDPC. The first, second and
third heuristic is the combination of new and derived
approach where as a new construction approach based on
branch and bound concept is proposed, the first heuristic is
coupled with the derived meta-heuristic process of Genetic
Algorithm, second heuristic is coupled with the derived
meta-heuristic process of Simulated Annealing and third
heuristic is coupled with the derived meta-heuristic process
of Hybrid Genetic Algorithm and Simulated Annealing. The
contribution lies in the introduction of new variant to the
domain of logistics and the development of three
construction heuristic based meta-heuristics (combination of
new and derived) as a unified solution methodology to solve
both TSP and SDPC.
2.8.3 Problem in Distribution Logistics
Transportation is generally considered as being a
significant factor of economic activities in any company.
The problem of appropriate usage of the vehicle fleet
appears as a matter of restricted resources of the company
and expectations of customers. The classical VRP with
backhauls (VRPB) is an extension of the VRP where two types
of customers are served from a single depot by a fleet of
vehicles. The first type of customers is known as “linehaul”
customers who require delivery of goods to their specified
location and the second type is known as “backhaul”
customers who require pickups from their specified
locations. In recent years, it became more obvious that in
real-world applications allowing vehicles, which are
returning from linehaul customers, to visit backhaul
customers leads to significant saving in the distribution
cost. Therefore the classical VRPB and its variants have
attracted the attention of researchers. The feature “each
vehicle has to serve backhaul customers, if any, after all
linehaul customers are served” defines the fact that in the
classical VRPB deliveries after pickups are not allowed. In
theory this restriction reduces the complexity of the
problem and in practice it avoids the problems that may rise
because of rearranging goods on the vehicle and supports the
fact that linehaul customers have priority over backhaul
customers. However, it can be easily proposed that ignoring
this restriction may reduce the total travelling cost.
Therefore, the mixed VRPB is defined as an extension of the
classical VRPB where the constraints and the objective are
the same as in the classical VRPB but deliveries after
pickups are allowed. This difference makes the mixed VRPB
(MVRPB) more difficult to solve than the classical VRPB. The
main reason behind this difficulty is the need to check the
capacity constraints for possible violation for every arc of
each route before inserting a customer into a new position
on any route. In the classical VRPB it is enough to check
capacity constraints violations in the corresponding part of
the route (backhaul or linehaul parts) while in the MVRPB
these capacity constraints have to be checked for every link
between the customers. Therefore, although the capacity
constraints seem to be similar for both problem types they
become more restrictive in the MVRPB.
This research addresses the variant MVRPB for the
application of third party logistics (3PL) service provider
organisation. The 3PL service providers are playing an
important role in the management of supply chains. The
global and competitive business environment of 3PLs has
recognised the significance of a speedy and proficient
service towards the customers in the past few decades.
Particularly in warehousing, distribution and transportation
services, customers anticipate improved lead times, fill
rates, inventory levels, etc.
This variant MVRPB is a not a new variant. This is
already addressed by Goetschalckx and Jacobs-Blecha (1989).
The objective is to develop new solution methodologies to
solve the variant MVRPB. Three solution methodologies are
proposed for the variant MVRPB. The first heuristic is the
derived approach. A derived clustering approach is combined
with the derived approach of Or-opt algorithm. The second
heuristic is the combination of new and derived approach;
where as a concept of changeover is coupled with the derived
meta-heuristic process of Simulated Annealing algorithm. The
third heuristic is the combination of new and derived
approach, where as a new composite concept with local search
is proposed and it is coupled with the derived meta-
heuristic process of Genetic Algorithm. The contribution
lies in the development of one heuristic (derived) and two
meta-heuristics (derived and combination of new and derived)
to solve MVRPB.
2.8.4 Problem in Urgency Logistics
The majority of existing supply chain research focuses on
managing and/or optimising the commercial supply of goods
and services. The supply process that deals during emergency
situations is an important domain for supply chain
management that has so far received little attention. The
unpredictability of the nature of the emergency and the
stake of adequate and timely delivery dominate this unique
and challenging material flow problem. Planning is a
critical process in those emergency and urgency situation.
Route planning systems are driven by choices of
objectives and the system picks the suitable algorithm based
on the need to provide the route. However, some experienced
drivers may choose the own route and the solution by the
system is much useful for inexperienced driver (Husdal,
1999).
At the same time, there are some special urgency /
forced requests and pressure from some of the customers to
take-back the used empty containers in order to free their
inventory space. Many customers are in high demand of space
and the demand of orders is also highly volatile. They would
like to move the empty containers out of the company as
early as possible in order to meet their emergency orders.
In the delivery and pick-up problem, the route planning
is normally made with respect to optimal cost. But, if there
is a specific urgency / forced requests, then the optimality
should be compromised with this aspect of service. In this
situation, some pick-up sequence is forced in the vehicle
routing and the route planning should be designed to satisfy
the urgency requests.
The original version of this problem is referred to as
VRP with Backhauls (VRPB) (Süral and Bookbinder, 2003; Wade
and Salhi, 2002). But, in the proposed variant the sequence
of forced pick-ups is considered. A set of pick-up nodes are
forced to visit and is known prior. The pick-up nodes should
be visited in sequential manner. The sequence of visiting
those pick-up nodes is not an input and it is considered for
optimization. The variant includes the constraints to
restrict the inclusion of delivery customers until a given
set of pick-up customers are served. So, this new variant is
an extension of MVRPRB with forced backhauls with the
objective of serving the urgency / forced requests in the
middle of delivery routes. This variant is termed as VRP
with Forced Backhauls (VRPFB). But, looking at the reality,
two objectives in VRPFB are considered and it is called as
Bi-Objective Vehicle Routing Problem with Forced Backhauls
(BVFB).
This variant BVFB is a new variant. The base variant of
this new variant is Vehicle Routing Problem with Backhauls
(VRPB) addressed by Goetschalckx and Jacobs-Blecha (1989)
and the bi-objective was coined by Thammapimookkul and
Charnsethikul (2001). The objective is to develop new /
derived / combination of new & derived and unified solution
methodologies to solve the variants VRPB and BVFB. Three
solution methodologies are proposed for the variants VRPB and
BVFB. The first heuristic is the derived approach. The
concept of arc removal is coupled with the derived savings
algorithm. The second heuristic is the derived approach;
where as a concept of node swap is coupled with the derived
savings algorithm. The third heuristic is the combination of
new and derived approach, where as a new fitness concept is
proposed and it is coupled with the derived meta-heuristic
process of Genetic Algorithm. The contribution lies in the
introduction of new variant to the domain of logistics and
the development of two heuristics (derived) and one meta-
heuristic (combination of new and derived) as a unified
solution methodology to solve both VRPB and BVFB.
2.8.5 Common Observations
Many of the authors (Potvin et al. 1996; Duhamel et al. 1997;
Fisher and Jaikumar,1981; Toth and Vigo, 1999) have
suggested the use of a constructive heuristic to obtain good
initial solutions for a meta-heuristic so that its
convergence can be accelerated. Only a few authors have
considered the use of hybrid approaches to solve different
variants of VRP. Glover et al. (1995) and Osman and Kelly
(1996) have pointed out that hybrid approaches focus on
enhancing the strengths and compensating for the weaknesses
of two or more complementary approaches. The aim is the
generation of better solutions by combining the key elements
of competing methodologies.
The quality of solutions obtained by many of the
proposed heuristic methods has not been established through
comparative evaluation with optimal solutions. While meta-
heuristics can yield better solutions, the computational
effort required by them often inhibits their use. There is a
scope for the application of multi-phase heuristics that use
a combination of intuitive and classical methods to
construct good initial solutions which, in turn, serve as
inputs for an intensive search using meta-heuristics. This
could yield quality solutions at reasonable computation time
(Johnson et al. 1991).
2.9 REVIEW OF THE VARIANTS OF TSP ADDRESSED IN THIS BOOK
Two variants of TSP are addressed namely mTSP and Bi-
objective TSP with Simultaneous Delivery and Pick-up.
Literature on these variants addressed in this study is
presented in the next few paragraphs.
2.9.1 Multiple Travelling Salesmen Problem (mTSP)
The mTSP is an extension of TSP with more than one salesman
all of them starting from one city (depot), visiting
different cities, and coming back to the starting city. mTSP
consists of finding tours for all m salesmen, who all start
and end at the depot, such that each intermediate city is
visited exactly once and the total cost of visiting all
cities is minimised. It can also be measured in terms of
distance, time, etc.
Hong and Padberg (1977) transformed an mTSP with fixed
charges for the assignment of salesmen and with a symmetric
cost matrix to a standard symmetric TSP for the ease of
handling. Lenstra and Rinnooy Kan (1979) have showed that
only linear admissible transformations are obtained by
adding constants to the rows and columns of a scalar
multiple of the distance matrix. Berenguer (1979) dealt the
mTSP with the transformation of the distance matrix and
analysed it in the linear context. Mole et al. (1983)
established the route first-cluster second heuristic to the
mTSP given that each salesman can visit any number of
customers in a stated range.
Kalantari et al. (1985) extended the branch and bound
algorithm of Little et al. (1963) to the TSP with pickup and
delivery customers which included single and multiple
vehicle cases as well as infinite and finite capacity cases.
Gavish and Srikanth (1986) developed a branch-and-bound
method for solving large scale mTSP and developed lower
bounds through a Lagrangean relaxation. Desrosiers et al.
(1988) proposed Lagrangian Relaxation Methods for solving
the Minimum Fleet Size mTSP with Time Windows to find the
minimum number of vehicles required to visit once, a set of
nodes.
Ferland and Michelon (1988) formulated a vehicle
scheduling problem and developed a heuristic and exact
methods for a single type of vehicle and shown that the
methods can be extended in a straightforward fashion to the
multiple-vehicle-types problem. Jonker and Volgenant (1988)
improve the standard transformation of the symmetric,
single-depot, mTSP to one on a sparser edge configuration.
Okonjo-Adigwe (1988) proposed an effective method of
balancing the workload amongst salesmen which addressed
large size problems and the addressed VRP is an extension of
the mTSP.
Wacholder et al. (1989) developed an efficient neural
network algorithm for solving the mTSP. They have introduced
a new transformation of the N-city M-salesmen mTSP to the
standard TSP. This algorithm was tested by them on many
problems with up to 30 cities and five salesmen. Okonjo-
Adigwe (1989) addressed the adult training centre problem
which resembles the mTSP. He proposed both heuristic and
exact algorithms to derive the best routing for four
vehicles which provide a daily service from a depot to 38
locations.
Fogel (1990) proposed a parallel processing approach to
solve the mTSP using evolutionary programming which
considers two salesmen with an objective of minimising the
difference between the lengths of the routes of each
salesman, in which he practiced the inversion mutation as
the genetic operator. Exceptionally good near-optimal
solutions were obtained for the problems with 25 and 50
cities with his evolutionary approach. Self-organising
approaches have also been successfully applied to the mTSP
where in Goldstein (1990) developed an extended elastic net
approach.
Hsu et al. (1991) proposed a neural network approach based
on the self-organised feature map model to solve the mTSP.
Gilbert and Hofstra (1992) introduced a new multi-period
mTSP with a polynomial heuristic method to the scheduling of
tour brokers and vendors at conventions of the tourism and
travel industry. The heuristic method is capable of
producing non-conflicting set of salesmen's tours. Brummit
and Stentz (1996) explored a dynamic environment that
involves multiple mobile robots in determining the optimal
path for each robot and to achieve the goals of the mission.
Chan and Merrill (1997) addressed a probabilistic multiple-
travelling-salesman-facility-location problem with an
asymptotic analysis using space-filling curve heuristic
which responds to stochastic demands. They suggested that
this procedure can be executed in the field with minimal
computational requirements. Torki et al. (1997) were motivated
by the outstanding performance of adaptive Neural Network
approach in the TSP and hence devised an algorithm to extend
the domain of applicability of this approach to the mTSP.
Modares et al. (1999) approached mTSP with several algorithms
based on self-organising neural network. A comprehensive
empirical study was provided by the authors in order to
investigate the performance of the algorithms. They stated
that the proposed algorithm exemplify significant advances
in the quality of the solution as well as the computational
efforts for most of the experimented data. Somhom et al. (1999)
introduced a new algorithm in competition-based network to
solve the minmax mTSP with an objective to minimise the
maximum distance among all salesmen travelled. They applied
the revised 2-opt exchange heuristic algorithms and the
elastic net algorithm to the minmax mTSP problem. They also
tried the combination of the adaptive algorithm with a
simple improvement heuristic and compared it with the
recently adaptive TS.
Tang et al. (2000) formulated the model for hot rolling
production scheduling with a case study at Shanghai Baoshan
Iron & Steel Complex and solved it using a new modified GA.
Chan et al. (2001) formulated A multiple-depot, multiple-
vehicle, location-routing problem with stochastically
processed demands and suggested a solution method for these
class of problems. Sofge et al. (2002) compared a variety of
evolutionary computation algorithms and paradigms for
solving the mTSP. They used a neighborhood attractor schema
(a variation on K-means clustering), the “shrink-wrap”
algorithm for local neighborhood optimisation, PSO, Monte-
Carlo optimisation, and a range of GAs and evolutionary
strategies for solving the same. Wang and Regan (2002)
described a solution method for the mTSP with time window
constraints. Their model described an iterative solution
technique in which explicit time constraints were replaced
by binary flow variables. Calvo and Cordone (2003)
introduced the overnight security service problem which was
modeled as a single-objective mixed integer programming
problem. They decomposed the problem as two subproblems
wherein one is a capacitated clustering problem and the
other is an mTSP with time windows and solved the problems
by employing the heuristic approach. Chan and Baker (2005)
addressed a mixed integer multiple depot, mTSP facility-
location problem formulation, expanded to include vehicle
range and multiple service-frequency requirements. This was
used to validate a heuristic solution for location and
routing in which a combination of the minimum spanning tree
and a modified Clarke-Wright procedure was employed.
Bektas (2006) reviewed the mTSP and its practical
applications and highlighted some formulations and described
exact and heuristic solution procedures proposed for mTSP.
Jacobson et al. (2006) demonstrated the use of generalised hill
climbing algorithms to determine optimal search strategies
over multiple search platforms for the mTSP. Computational
results with this algorithms applied to the mTSP
demonstrated that near optimal search strategies over
multiple search platforms can be obtained more competently
when limited computing budgets are available. Their interest
was mainly on the military applications. Junjie and Dingwei
(2006) suggested that classical optimisation procedures were
not adequate for the mTSP problems and thus made an attempt
to show how the ACO can be applied to the mTSP with ability
constraint.
Chandran et al. (2006) proposed a clustering approach to
solve the mTSP. The proposed method found to achieve a good
balance of workloads among the clusters, each of which is
visited by a salesperson, when tested over a range of data-
sets. Carter and Ragsdale (2006) proposed a new approach to
solving the multiple travelling salesperson problem using
GA. In that a new GA chromosome and related operators were
proposed for the mTSP. Fügenschuh and Martin (2006)
addressed the mTSP with a multi-criteria approach for
optimising bus schedules and school starting times. They
discussed the legal framework for the school and trip
starting times and presented a multi-objective mixed-integer
linear programming formulation for the same. They also
developed a two-stage decomposition heuristic and applied it
to practical data sets from three different rural German
counties. Kara and Bektas (2006) extended the classical mTSP
by imposing a minimal number of nodes that a traveller must
visit as a side condition. They considered single and multi-
depot cases and proposed integer linear programming
formulations for both, with new bounding and sub tour
elimination constraints.
Estévez-Fernández et al. (2006) introduced multiple
longest TSP. Mitrović-Minić and Krishnamurti (2006) found the
vehicle bounds for the mTSP with time windows by covering
two precedence graphs with the minimum number of paths. Qu et
al. (2007) solved the mTSP with a columnar competitive model
of neural networks incorporating with a winner-take-all
learning rule. Malik et al. (2007) presented an algorithm with
an approximation factor of 2 for the mTSP with Multiple
Depot. Chung et al. (2007) developed mathematical models for
the container road transportation in Korean trucking
industries by utilising the standard formulations of well
known TSP and VRP. They also presented Heuristic algorithms
to solve the models.
Vallivaara (2008) proposed a Team ACO for the mTSP with
MinMax Objective in which a team of ants is allowed to
construct solutions to the problem. Muralidharan et al. (2008)
developed an efficient mechanism for multi-robot
coordination. Their work was on the robot team that works
together on tasks that should be made to share the workload
efficiently in a cost effective manner. They formulated an
mTSP model and presented a novel approach for multi-robot
coordination using combinatorial auctions in order to
maximise robot utilisation and at the same time minimising
incurred costs. They also used Visual Basic System to
simulate the system and they validated the robustness of the
mechanism. Contributions by Li and Lu (2010) with a self-
organizing algorithm, Yamamoto et al., (2010) with an adaptive
routing method and Ghafurian and Javadian (2011) with an ant
system are considered to be recent additions in the
literature to solve the mTSP.
2.9.2 TSP with Simultaneous Delivery and Pick-up
TSP with Simultaneous Delivery and Pick-up, also known as
TSPSDP, remains a less researched problem. The Vehicles such
as mail vans and buses used for public transport need to be
routed both for delivery and pick up. Here as the service
node is visited only once by the vehicle, both delivery and
pick-up should take place at the same time, i.e.
simultaneously. In the limited literature on this problem,
Tang and Galvao (2002) used four different heuristics, viz.,
initial node heuristic, cheapest feasible insertion
heuristic (Mosheiov, 1994), minimal spanning tree heuristic
(Anily and Mosheiov, 1994) and cycle heuristic (Gendreau et al.
1999). Ganesh et al. (2007b) developed a heuristic called CAGE
(Construction of initial solution by Agglomeration and GA
for Effective search) to solve TSPSDP.
2.10 REVIEW OF THE VARIANTS OF VRP ADDRESSED IN THIS BOOK
Two variants of VRP namely Mixed Vehicle Routing Problem
with Backhauls (MVRPB) and Bi-Objective Vehicle Routing
Problem with Forced Backhauls (BVFB) are addressed here.
Literature on these variants addressed in this study is
presented in next few paragraphs.
2.10.1 Vehicle Routing Problem with Backhauls (VRPB)
Heuristics are commonly used because of the complexity to
deal with the classical VRPB and it has been studied by
several researchers for more than two decades. The first
constructive method for the classical VRPB was proposed by
Deif and Bodin (1984) which is the extension of Clarke and
Wright’s (1964) savings algorithm. Goetschalckx and Jacobs-
Blecha (1989) formulated the first mathematical problem
explicitly dealing with the VRP with clustered Backhauls.
They developed a heuristic method for the multi vehicle case
in which the clustering as well as the routing part are
solved by means of a space filling curve heuristic. Toth and
Vigo (1996) proposed a cluster-first, route-second heuristic
for VRPB.
Anily (1996) developed a lower bound on the optimal
total cost and a heuristic solution for the VRPB. The routes
generated by the heuristic were formed such that the
backhaul customers are served only after the delivery
customers are served.
Goetschalckx and Jacobs- Blecha (1993) used a
clustering method in their second paper, which is based on
the generalised assignment approach proposed by Fisher and
Jaikumar (1981). In this approach line-haul customers and
the backhaul customers are sorted according to their
increasing distance from the depot and decreasing distance
to the depot respectively. By solving generalised assignment
heuristics, both customer sequences are divided into K
clusters. Then line-haul and backhaul routes are merged
according to the best combination of connections that has
the smallest distance at the same time not allowing any
backhaul customer before a line-haul customer is served. The
results obtained by this method are superior to their first
approach (Goetschalckx and Jacobs-Blecha, 1989). Thangiah et
al. (1996) described a route construction heuristic as well
as different local search heuristics to improve the initial
solutions.
Toth and Vigo (1999) proposed another two-phase method
for the classical VRPB. They solved both symmetric and
asymmetric VRPB problem using the cluster-first route-second
heuristic approach. In their approach, visiting backhaul
customers before line-haul customers and routes containing
only backhaul customers are not allowed. There are no
distance restrictions for the vehicles. In the first phase
of the heuristic, Lagrangian relaxation method is used to
cluster the line-haul and backhaul customers separately.
They tested and compared their approach on three different
sets of VRPB instances with respect to the optimal solutions
with the approaches proposed by Deif and Bodin (1984) and
Goetschalckx and Jacobs-Blecha (1989) and found to be
outperforming.
Mingozzi et al. (1999) and Toth and Vigo (1997) approached
VRPB with exact methods. Mingozzi et al. (1999) formulated the
VRPB as an integer programming problem and described a
procedure that computes a valid lower bound to the optimal
solution cost by combining different heuristic methods. The
proposed exact algorithm was claimed to solve problems up to
100 customers. Salhi and Nagy (1999) applied a cluster
insertion heuristic for single and multiple depot VRPB. Toth
and Vigo (1997) described a new (0-1) integer programming
formulation of the VRPB based upon a set-partitioning
approach. Wade and Salhi (2001) proposed an Ant System
Algorithm for the VRPB. Wade and Salhi (2002) investigated
the problem with insertion-type heuristic algorithms. Osman
and Wassan (2002) presented a reactive TS heuristic to solve
the VRPB. Ropke and Pisinger (2004) reviewed numerous ways
of modelling backhaul constraints and the various
restrictions on handling backhaul nodes.
Ropke (2005) addressed the VRP with pickup and delivery
and solved using Adaptive Large Neighborhood Search
heuristic, Branch-and-Cut algorithm and Branch-and-price
algorithms for the VRPB problems with time windows. Ropke
and Pisinger (2006) improved their own version of the large
neighborhood search heuristic (Ropke and Pisinger, 2004) to
solve VRPB. Brand˜ao (2006) presented a new TS algorithm
which was able to match almost all the best published
solutions and also found many new best solutions
particularly for a large set of benchmark problems. In a
nutshell Ropke and Pisinger (2006) and Brand˜ao (2006)
proposed competing results for the benchmark data sets
addressed so far. An extensive survey on VRPB and its sub
classes is available in Ropke (2005) and Parragh et al. (2008).
Currie and Salhi (2004) presented new TS algorithms for the
VRPB whereas Nagy and Salhi (2005) proposed modified
heuristic algorithms.
Crispim and Brandao (2001) applied the reactive TS and
variable neighbourhood descent to the VRPB. Tavakkoli-
Moghaddam et al. (2006) proposed a memetic algorithm which used
different local search algorithms (inter and intra-route
node exchanges) to solve the VRPB. Wassan (2007) proposed a
heuristic approach based on a hybrid operation of reactive
TS and adaptive memory programming to solve VRPB. Alshamrani
et al. (2007) developed a heuristic procedure for developing a
route design-pickup strategy planning, a reverse logistics
problem, motivated by blood distribution of the American Red
Cross. Imai et al. (2007) developed a sub-gradient heuristic
based on Lagrangian relaxation which consists of two sub-
problems: the classical assignment problem and the
generalised assignment problem. Gajpal and Abad (2009) used
multi-ant colony system to solve VRPB.
2.10.1.1 Mixed VRP with Backhauls (MVRPB)
The MVRPB is an extension of the classical VRPB where
deliveries after pickups are allowed and line-hauls and
backhauls are sequence-independent. There are a very few
papers addressing this problem. Golden et al. (1985) developed
an approach based on inserting backhaul nodes into the
routes formed by linehaul nodes. Casco et al. (1988) obtained
better results with a load-based insertion procedure which
considers the cost of inserting backhaul nodes. Mosheiov
(1994) investigated the TSP with delivery and pick-up (TSPDP)
and proposed a methodology to satisfy the load constraint.
Anily and Mosheiov (1994) presented a minimal spanning tree
approach for solving TSPDP. Salhi and Nagy (1999) extended
the method of Casco et al. (1988) by allowing backhauls to be
inserted in clusters.
Wade and Salhi (2003) introduced an enhanced ACO for
the MVRPB. Crispim and Brandão (2005) presented a hybrid
algorithm which comprises TS and variable neighbourhood
descent for the MVRPB. Wassan et al. (2008) implemented a
metaheuristic based on reactive TS for the problem. Tütüncü
et al. (2009) described a new visual interactive approach based
on Greedy Randomised Adaptive Memory Programming Search for
the mixed VRP with backhauls. Tütüncü (2010) also proposed
the same approach to solve the heterogeneous fixed fleet
vehicle routing problem with backhauls.
2.10.2 VRP with Forced Backhauls
In practice, in routing, it is indispensable to force the
vehicles to visit a set of nodes, for many reasons. This
important variant of VRPB can be thought of as forced VRP
with backhauls. Such type of problems has not been addressed
so far. Here the force is on the pick-up nodes in that the
vehicle is required to visit a set of pre-defined cluster.
The sequence of visiting the pick-up nodes are also
considered for optimisation.
Wade and Salhi (2002) coined the word restricted VRPB
(R-VRPB) in which mixed linehaul and backhaul customers are
permitted with a constraint of serving the backhaul
customers first. R-VRPB includes the constraints to restrict
the inclusion of backhaul customers until a given percentage
of the total linehaul load has been delivered.
They used a simple constructive heuristic, the R-INS
method, which used a greedy insertion heuristic (Salhi and
Nagy, 1999) to illustrate the practicality of this
restricted version of the mixed VRPB. In this algorithm, is
asked to set a restriction percentage on the insertion of
backhaul customers. In their procedure the linehaul
customers were routed first and then the backhaul customers
are inserted into the route and the RP was used to control
these insertions.
Tutuncu et al. (2009) proposed an inexpensive decision
support system based a new Greedy Randomised Adaptive Memory
Programming Search algorithm to solve the classical VRPB,
the mixed VRPB and the restricted VRPB in a visual
interactive environment. He claimed that the computational
results on VRPB benchmark test problems indicated that the
proposed visual interactive approach is effective towards
finding a compromise amoung the mixed, restricted and the
classical VRPB problems. The survey on literature reveals
that the forced backhauls is a new variant and it is not
addressed directly in the literature.
2.11 SUMMARY
This chapter presents a survey of literature on the selected
and complex variants of TSP and VRP that pertain to balanced,
reverse, distribution and urgency logistics. It also
highlights a report on variants, methodologies and
applications that call for further investigations. The gaps
identified from the literature provided the motivation for
the issues addressed in this book.