International Journal of Trend in Scientific

Research and Development (IJTSRD)

International Open Access Journal
ISSN No: 2456 - 6470 | www.ijtsrd.com | Volume - 1 | Issue – 6

An Enhanced Agglomerative Clustering Algorithm for Solving

Vehicle Routing Problem
V. Praveen V. Hemalatha M. Poovizhi
Assistant Professor, CSE, Assistant Professor, CSE, Assistant Professor, CSE,
N.S.N College of Engineering & N.S.N College of Engineering & N.S.N College of Engineering &
Technology, Karur, Tamilnadu Technology, Karur, Tamilnadu Technology,
hnology, Karur, Tamilnadu

ABSTRACT

An aggrandized solution is designed for the vehicles known locations and known demand for any
to reduce the total cost of distribution by which it can commodity, and that required goods can be delivered
supply the goods to the customers with its known to the customer from a single depot by some
capacity can be named as a vehicle routing problem. calculated amount of delivery vehicles with some
In variable Neighborhood search method, mainl mainly an basic conditions and constraints are specified [1]:
[1
efficient vehicle routing can be achieved by
(i) The demands of all customers are met
calculating the distance matrix value based on the
(ii) Each customer is served by exactly one vehicle
customer’s location or the path where the customer’s
(iii)
iii) For each route the total demands must not exceed
resides. The main objective of the paper is to reduce
the capacity of the vehicle which is already defined.
the total distance travelled to deliver the good
goods to the
customers. The proposed algorithm is a hierarchy From a depot different products must be distributed to
based enhanced agglomerative clustering algorithm several retailers. An efficient collection (or)
technique which is used in the data mining scenario distribution of goods keeps transport inventories
invento low,
effectively. The proposed algorithm decreases the it saves resources and energy. Therefore, vehicle
total distance assigning to each route and the routing is one of the important topics for this kind of
important thing need to consider is that, this enhanced problems.
clustering algorithm can reduce the total distance
when compared to the previously proposed variable
Neighborhood search method.

Keywords: High Performance Concrete, Marble Dust,

Steel fibers, Super Plasticizer.

1. Introduction
In general, there are many practical applications
which can provide efficient distribution of goods to Fig.1 Example for Vehicle Routing Problem
the customers. Goods in the sense it can be any home
appliance products which are used daily. The vehicle The vehicle routing problem is a common name given
scheduling problem was first formulated in the year to a whole class of problems involving the visiting of
1959 [3]] and in that set of customers with each of its customers by using vehicles. These problems derive

@ IJTSRD | Available Online @ www.ijtsrd.com | Volume – 1 | Issue – 6 | Sep - Oct 2017 Page: 919
 International Journal of Trend in Scientific Research and Development (IJTSRD) ISSN: 2456-6470
their name from the basic practical problem of II. Literature Survey
supplying geographically dispersed customers with
goods using a number of vehicles operating from a The first article for the ‘Truck dispatching problem’
common goods depot (or) warehouse. was published by Dantzig and Ramser presented a
larger truck dispatching problem, that is referred to as
An example for a single repository based vehicle D&R problem and many more enhanced solutions are
routing problem is shown in the Figure 1. For a found which is similar to this article after it was
classical vehicle routing problem, the best solution is published.
to serve the goods to the customers exactly once by
starting from and ending to the depot. The Clark and Wright algorithm[4] is one of the most
popular heuristic algorithms in the vehicle routing
The main objective is to reduce the overall problem area. Cordeau et al. described an analysis that
transportation cost by satisfying all the constraints. parallel version is much better because merge yielding
The cost for transporting the goods can be reduced by the largest saving is always implemented, but the
minimizing the total distance travelled and as well as sequential version keeps expanding the same route
the total number of vehicles. While comparable to the until there is no longer feasible route.
classical vehicle routing problem, the majority of the
real world problems are much more complex to solve. Chopra and Meindl [2] provide a solution for vehicle
In general, the classical vehicle routing problem is route planning, in that they present a routing and
based on some constraints like the total vehicle scheduling, transportation problem for a company in
capacity or some time interval to reach the customers. which they use a method called savings matrix
method. The algorithm can be classified into four
A single depot vehicle routing problem uses a single
steps which are: (1) Identify the distance matrix for
depot (or) warehouse for delivering the goods to the
the given location, (2) calculating the saving matrix
customers, several algorithms and saving methods are
using the distance matrix values, (3) assigns,
proposed for solving the single depot based real time
customers to vehicles or routes, and (4) sequence the
problems. In general, the VRP is a Combinatorial
customers within the routes. The first two steps are
Optimization Problem and it consists of two main
explained clearly. The third step is that assign the
things are depot and destinations. A formal example
customers to vehicles and routes by, initially each
for this kind of problem is Soft Drink Company. In
customer are assigned to a separate route. If the two
that they are travelling from the company to all the
routes can provide a feasible solution by which it
retail stores to distribute the products and again came
doesn’t cross the limited capacity means it can be
back to the company. The main constraint followed is
combined. The procedure is continued until no more
to visit the customers only once.
feasible combinations are possible.
Vehicle routing problem is also known as delivery
For a transportation problem, Lumsden and Jonsson
problem. For example, sewage must be collected from
described a similar explanation but it is not clear.
households and industries to a distracting place, so for
Rand[6] made an analysis and presents an article
that transportation everyone needs an efficient route
about the different saving methods for the vehicle
to travel from one place to another. It is very much
routing problems. In that he argues about the parallel
useful for the day to day transportation because it
version, because it is not always better than the
reduces the cost of forming the routes based on the
sequential version. Parallel version is a heuristic and
capacity of the vehicle.
there is no guarantee from the obtained results that it
The rest of the paper is organized as follows. Section produces the optimal solution or near optimal
II presents the literature review. Section III presents a solution.
statistical example, using variable Neighborhood
The Tabu Search heuristic algorithm [7] for the
search method. Section IV describes the proposed
vehicle routing problem having both capacity and
methodology based on Agglomerative clustering
route length restrictions. The algorithm considers a
algorithm method. Section V describes the
sequence of adjacent solutions obtained by repeatedly
comparison results. The final conclusion is presented
removing a vertex from its current route and
in Section VI. reinserting it into another route. This is done by
means of a generalized insertion procedure previously
developed by the authors. During the course of the

@ IJTSRD | Available Online @ www.ijtsrd.com | Volume – 1 | Issue – 6 | Sep - Oct 2017 Page: 920
 International Journal of Trend in Scientific Research and Development (IJTSRD) ISSN: 2456-6470
algorithm, infeasible solutions are allowed. Numerical Table I Distance And Demands For Customers
tests on a set of benchmark problems indicate that
Customer Location Demand
Tabu Search outperforms the best existing heuristics,
and Tabu route often produces the best known
solutions. So many new methods can produce best 1 (40,40) 12
results compared to Tabu Search.
2 (36,26) 21
The Adaptive Memory Procedure (AMP) [8] was first
proposed by Rochat and Taillard. It is an 3 (21,45) 25
enhancement of Tabu Search (TS) to solve the VRP.
It was motivated by the work of Glover regarding 4 (45,35) 15
surrogate constraints. An important principle behind
AMP is that good solutions may be constructed by 5 (10,10) 16
combining different components of other good
solutions. A memory con- 5 training components of 6 (55,45) 24
visited solutions is kept. Periodically, a new solution
is constructed using the data in the memory and 7 (26,59) 12
improved by a local search procedure.
For the VRP problem, a Variable Neighborhood 8 (55,15) 17
Search algorithm[9] in which four Neighborhood
structures are designed to find the planning of trips 9 (40,30) 20
within the four different Neighborhood structures. By
which two of them is used to minimize the total length 10 (20,14) 25
and the other two is used to minimize the overall time.
It includes a shaking step where three of the
Neighborhood structures are probably used based on If Cab≠Cba means then the problem is asymmetric.
the need. The algorithm was tested over a set of Starting from the central warehouse, goods are
benchmark problems and the obtained solutions were delivered to the customers: 0-10. Initially the distance
compared with five previously proposed algorithms. for each customer is given. In Table 1, locations and
the demands for each customer are given.
III. Statistical Example
According to the existing scenario, Variable
The ultimate base for the vehicle routing problem is Neighborhood Search method uses the symmetric cost
travelling salesman problem because the impulsion is for returning to the depot, i.e. the distance from 1 to 5
“visit all the customers only once” used in the vehicle is the same as the distance from 5 to 1.
routing problem is as same as that of travelling
salesman problem. Depict that a delivery man must The location for the depot (or warehouse) is (40, 40)
visit some ‘n’ number of customers and returned to (x-axis and y-axis values) and obviously the demand
the starting point after visiting all the customers only is zero. Based on the customer location, initially
once and the total cost for visiting all the customers is distance matrix is calculated using the Eq.(1) and its
the dominant problem. cost are symmetric. Eq.(1) represents the distance
The solution is to obtain a minimum [5] cost line to between the customer ci and the depot k.
visit all the customers only once. Infer when the cost
Eq.(1), Distance matrix formula
for travel from city a to city b is equals to the cost of
city b to city a, then the problem is considered as
symmetric. D(ci , k )  ( xci  xk ) 2  ( yci  yk ) 2

@ IJTSRD | Available Online @ www.ijtsrd.com | Volume – 1 | Issue – 6 | Sep - Oct 2017 Page: 921
 International Journal of Trend in Scientific Research and Development (IJTSRD) ISSN: 2456-6470
Table II Distance Matrix Calculation Each vehicle is having same number of capacity limit.
The maximum vehicle capacity defined for each
vehicle in the above example is 70. Table 3 shows
Cij 0 1 2 3 4 5 6 7 8 9 that the result with the total distance of 221 and three
vehicle is needed for that transportation. This method
0 - 1 2 7 4 1 2 2 1 3 reduces the distance at a certain minimum compared
5 0 2 6 4 9 0 3 to the earlier methods.

1 - 2 1 3 2 1 2 6 2 This decrease is not enough when the vehicle is used

4 3 1 7 7 2 0 daily for delivering the goods to the customers, so for
these kind of methods are proposed to solve all kinds
of vehicle routing problems.
2 - 2 3 3 1 4 2 3
6 7 4 5 5 4 1 IV. Proposed Method
The total distance is calculated while travelling to
3 - 4 1 3 2 7 3 deliver the products to different destination points.
3 4 1 2 3 The total distance is directly proportional to the total
cost and total time. In vehicle routing problem, the
4 - 1 5 4 3 1 main objective is need to find a route with minimum
1 2 5 6 1 total distance. In the proposed work total distance
travelled is considered as the first parameter.
5 - 3 3 2 4
2 0 1 7 Large instance set of problems which are mainly
related to real world applications require more
6 - 5 3 2 number of vehicles to cover several numbers of
3 2 5 distribution points. In the proposed work planning to
cover the several numbers of destination points with
7 - 2 3 minimum number of vehicles. So the number of
1 5 vehicles is considered as a second parameter.
In existing Variable Neighborhood Search method,
8 - 2 initially the distance matrix values are calculated
1 using the Eq.(1) and based on the calculated distance
value Neighborhood for all the customers are found.
9 - A. Exaggerated Agglomerative Clustering Algorithm
(EACA)
Using the variable Neighborhood search method, the
nearest neighbor for each customer is found. Based on The proposed approach is based on Agglomerative
that, the vehicles are routed for different destination clustering algorithm for solving the vehicle routing
points. Shaking step is applied at the last to minimize problem with multiple repositories. By following the
the overall total distance. Exaggerated Agglomerative Clustering Algorithm,
different clusters are found which is mainly used for
Table III Solution For VNS Algorithm With Three
routing the vehicles efficiently from the repository to
Routes
all the destination points.
Trip Total Total A.1 Exaggerated Agglomerative Clustering Algorithm
Distance Demands
The algorithm steps as follows:
Route1 0-3-8-4-0 50 51
1 Choose the number of clusters, k.
Route2 0-5-9-1-0 83 70 2 Randomly generate k clusters and determine the
cluster centers, or directly generate k random
Route3 0-2-6-7-0 88 54 points as cluster centers.
3 Assign each point to the nearest cluster center.
4 Recompute the new cluster centers.

@ IJTSRD | Available Online @ www.ijtsrd.com | Volume – 1 | Issue – 6 | Sep - Oct 2017 Page: 922
 International Journal of Trend in Scientific Research and Development (IJTSRD) ISSN: 2456-6470
2456
5 Repeat the two previous steps until some Table IV Customers With Location And Demands
convergence criterion is met or the assignment has
not changed. Customers A(x-axis) B (y-axis)
(y Demands
6 Apply the search procedure forr finding the final 1 40 40 12
routes 2 36 26 21
3 21 45 25
4 45 35 15
A.2 Flowchart for Exaggerated Agglomerative 5 10 10 16
Clustering Algorithm 6 55 45 24
7 26 59 12
8 55 15 17
9 40 30 20
10 20 14 25

Table V Distance
ce Matrix Calculation

Cij 0 1 2 3 4 5 6 7 8 9

0 - 1 1 1 3 2 2 3 9 2
0 6 2 4 5 5 0 5

1 - 2 1 3 2 3 2 6 2
5 3 1 7 5 2 0

2 - 2 3 3 1 4 2 3
6 7 4 5 6 5 2

Fig.2 Exaggerated Agglomerative Clustering 3 - 4 1 3 2 8 3

Algorithm 4 5 1 3 3

The flowchart for the Exaggerated Agglomerative 4 - 5 5 4 3 1

Clustering Algorithm (EACA) is shown in the Figure 8 2 6 7 1
2. Using the Neighborhood search procedure, efficient
routes are found for the vehicles which are going to 5 - 3 3 2 4
deliver the goods to the customers. Simultaneously 3 0 2 7
the total load for each vehicle doesn’t exceed the
maximum capacity of that vehicle. 6 - 5 3 4
A.3 Statistical Example 3 3 6
A set of 10 customers with their locations and 7 - 2 3
demands are given in the Table 5. When the clustering 2 6
is performed for a set of values, there may be two or
more values can be formed. Based on the algorithm of
8 - 2
Agglomerative clustering, two centroid values are
6
found with two sets of different customers.
9 -

Each customer belongs to one centroid point,

otherwise called as depot. Using the Euclidean
distance formula, the distance
istance between the customers

@ IJTSRD | Available Online @ www.ijtsrd.com | Volume – 1 | Issue – 6 | Sep - Oct 2017 Page: 923
 International Journal of Trend in Scientific Research and Development (IJTSRD) ISSN: 2456-6470
2456
to particular depots is calculated and finally the total The proposed method shows a better solution against
distance and the total number of vehicles needed to the previously proposed methods and exaggerated
perform efficient transportation is found. agglomerative clustering methods which are mainly
used in the data mining concepts. Among all the
The final route points and their values which satisfy
existing methods, exaggerated
exagg agglomerative
the termination condition
ition is shown in the Table 6.
clustering method can reduce the total distance of
This method provides a better solution compared to
about 7% while using multi depots for delivering the
the previous proposals by different authors.
products to customers.
Table VI EACA Result With Three Routes
References
Trip Total Total
Distance Demands [1] Behrouz Afshar-Nadjafi,
Nadjafi, Alireza Afshar-Nadjafi,
Afshar
2014. “A constructive heuristic for time dependent
Route1 0-1-8-3-0 24 56
multidepot vehicle routing problem with time- time
Route2 0-4-9-5-0 92 65 windows and heterogeneous fleet”, Journal of king
saud university-Engineering
Engineering sciences.
Route3 0-6-2-7-0 86 49
[2] Chopra, S., Meindl, P., ., 2004. Supply Chain
Strategy, Planning and Operation, 2nd
Management –Strategy,
V. Comparison Results ed. Pearson Prentice hall, Upper Saddle River.
The sample problems were solved by NetBea
NetBeans IDE [3] G.B.Dantzig, J.H.Ramser, 1959. “The truck
and the proposed Exaggerated Agglomerative dispatching problem”, Management Science, Vol. 6,
Clustering Algorithm
lgorithm shows the best result while No. 1, pp.80-91.
compared to Variable Neighborhood Search
algorithm. [4] G.Clarke, J.Wright, 1964. “Scheduling of vehicles
from a central depot to a number of delivery points”,
The result is analyzed based on the parameter called Operations Research, Vol. 12, No. 4, pp. 568-581.
568
total distance. The comparison graph is shown in the
Figure 3. [5] Gaurav Sharma, Preeti Bansal, 2012. “Min-Min
“Min
approach for scheduling in grid environment”,
International
onal Journal of Latest Trends in Engineering
VI. Conclusion and Technology(IJLTET), Vol 1, Issue 1.
While calculating the savings matrix approach for the
[6] Graham K Rand, 2009. “The life and times of
Variable Neighborhood method, it provides a good
savings method for vehicle routing problems”,
solution for the small instance set, but for a large
http://www.orssa.org.za, Volume 25(2),pp.125-145.
25(2),pp.125
instance set it doesn’t yield a better result.
[7] Gendreau M., A. Hertz and G. Laporte, A tabu
search heuristic for the vehicle routing problem,
Management Science 40 (1994), pp. 1276–1290.
1276
[8] Olvera A. and O. Viera, Adaptive Memory
Programming for the Vehicle Routing Problem with
Multiple Trips, Computers & Operations Research
Resear 34
(2007), pp. 28–47.

[9] Mohamed Cheikh, Mustapha Ratli, Omar

Mkaouar, Baseem Jarboui, A Variable Neighborhood
Search Algorithm for the vehicle routing problem
with Multiple trips, Electronic notes in Discrete
Mathematics 47(2015), Elsevier, pg 277-284
277
Fig.3 Total Distance

@ IJTSRD | Available Online @ www.ijtsrd.com | Volume – 1 | Issue – 6 | Sep - Oct 2017 Page: 924