You are on page 1of 34

Department of Industrial & Management Engineering

Indian Institute of Technology, Kanpur

Cost Minimization For Warehouse allocation Problem


(Developing Strong Constraints for the Multi
-Commodity SSCWLP)

Thesis Supervisor:
Prof. R.R.K. Sharma

by
Parag Tyagi
(13114016)
1

Outline
Introduction
Literature Survey
Previous Basic Formulations
New Formulation for Multicommodity case
Theorem/Proposition
Summery and Conclusions
References
2

Introduction

Introduction Continues

Warehouse allocation Problem is


encountered in areas like FMCG Supply
Chain, Fertilizers Distribution System, FCI
(Food Corporation of India) more
pronouncedly
For e.g. FCI has the following Structure

Introduction Continues
Function of FCI is to distribute the Food Grains
throughout the country for PDS and other
Government Schemes
FCI procures the Food Grains from the Mandies
and Stores in respective Central warehouses of
each Zone
From these Central warehouses Food Grains are
transported to District warehouses and further to
the Depots and PDS at the end
Problems encountered here are maintaining a
satisfactory level of operational and buffer stocks
of food grains to ensure national food security
and some unforeseen problems like low food
production, drought, flood, war, crop failure etc.
5

Literature Survey
Facility allocation, being a important decision
in supply chain planning has been studied by
many researchers and there have been
massive developments from last four decades
Warehouse allocation problem has been
undertaken in its variants forms,
For e.g. 1. SPLP/CPLP
2. Single commodity/ Multicommodity
3. Single Stage/Multi-Stage

Literature Survey continues

An overview of the work done in this field can


be looked from the review work done by
ReVelle and Eiselt(2005)
Our problem is a SSCWLP, in which facility
location is considered at a single stage
Single Stage facility location problem has
been attempted by many authors such as
Geoffrion and Graves(1974) and
Sharma(1991) and very interestingly they
have given different formulations
Geoffrion and Graves(1974), Sharma (1991),
Sharma(1996) and Kouvelis (2004) have used
weak formulations of the problem
7

Literature Survey continues

Geoffrion and Graves (1974) formulations


In this paper, formulations have been
developed for a Multi-Commodity
capacitated single period real life problem
Problem taken here is of a major food firm
with 17 commodities classes, 14 plants, 45
possible distributions center (DC)sites and
121 customers zones
MODEL
i :
index for commodities,
j :
index for plants,
k :
index for possible distribution center
(DC) sites,
l :
index for customer demand zones, 8

Literature Survey continues

Sij : supply (production capacity) for commodity i at


plant j,
Dil : demand for commodity i in customer zone l,
Vk ,Vk : minimum, maximum allowed total annual
throughput for a
DC at site k,
fk : fixed portion of the annual possession and
operating costs for a DC at site k,
vk : variable unit cost of throughput for a DC at site
k,
cijkl : average unit cost of producing and shipping
commodity i from plant j through DC k to customer
zone l,
xijkl : Variable denoting the amount of commodity i
shipped from plant j through DC k to customer zone9 l

Literature Survey continues


ykl : a 0-1 variable that will be 1 if DC k serves customer
zone l, and 0 otherwise.
zk : a 0 -1 variable that will be 1 if a DC is acquired at site k,
and 0 otherwise.
#The problem can be written as the following mixed integer
linear program (MILP):
(1) Minimizex0; y,z=0,1 ijklcijklxijkl + k[fkzk + vkilDilykl]
Subject to:
(2) kl xijkl < Sij ,
all i,j
(3) j xijkl = Dilykl ,
all i,k,l
(4) k ykl = 1,
all l
(5) Vkzk ilDilykl Vkzk,
all k
(6) Linear configuration constraints on y
and/or z
10

Literature Survey continues

Model clearly states that each customer zone can


get the commodities from one DC only (constraint
(4))
Quantity ilDilykl the total annual throughput of the
kth DC which will always be within the permissible
capacity limits of a particular site to be economical
A solution technique based on Benders
decomposition is developed, implemented and
successfully applied to solve the model
Benders well known partitioning procedure is
applied in such a way that Multi-Commodity LP
sub-problem decomposes into as many
independent classical transportation problems as
there are commodities
11

Literature Survey continues

Sharma (1991) formulations

Mixed zero-one integer linear programming


problem is formulated for a real life problem
of Fertilizers production-distribution system
Model has been formulated by considering
the two seasons each of six months keeping
in mind the cropping seasons(usually of six
months)
95 % Fertilizers are moved from plants to
rake points and from there moved to
secondary points and then finally reach the
market by LCV
In the Model plants, rake points, secondary
12
points and markets, will be referred to as

Literature Survey continues


MODEL:

i : index for point type


j : index for number of particular point type
k : index for period
h : index for product type
u : index for nutrient type
H : total number of products
U : total number of nutrients
K : total number of periods in six months duration (assumed
even for convenience)
N(i) : total number of points of category i
Bijkh, Eijkh : beginning and ending inventory at point number j of
point type i in period k of product type h is represented by
respectively
Lij : location variable which is 1 if it is decided to locate a
warehouse at point number j of point type i (where i varies
from 2 to 3) and 0 otherwise
D4jk : quantity of nutrient demanded of type u at market j
(point of type 4) in period k

13

Literature Survey continues


P1jkh : quantity produced at plant number j (point of type 1) in
period k of product type h
QTR4jkh : quantity received at market number j (point of type 4) in
period k of product type h
S1ij, S2ij : Space booked at point number j of point type i in the
first and last three months of a six monthly season respectively
Ti1,j1,i2,j2,k,h : quantity transported from point number j1 of point type
i1 to point number j2 of point type i2 in period k of product type
h
TTi,j,k : total quantity transported from point number j of point
type i in period k
fhu : fraction of nutrient of type u obtainable from product of
type h
SN(i, j ) : set of points of type i + l to which a point of type i
having a number j can supply products
RN(i, j ) : set of points of type i - 1 from which a point of type i
having a number j can receive products
LPljkh, UPljkh and CPljkh : respectively the lower limit, upper limit and
unit cost of variables Pljkh for all j, k and h

14

Literature Survey continues

Distribution cost is composed of the


Production cost: CPljkh * P1jkh ;
Cost of space: CSlij *S1ij + CS2ij*S2ij ;
cost of carrying inventory: 0.5*CIijkh*(Eijkh + Bijkh )
fixed cost of locating a warehouse: CLij * Lij
Objective Function (Minimizing Total Cost):
Minimize jkhCPljkh * P1jkh + ijkpmnCTijkpmn * Tijkpmn
+
ijkhCIijkh *0.5* ( Eijkh + Bijkh) + ijCSlij
*Slij +
ijCS2ij *S2ij + ij CLij * Li j

Constraints:

15

Literature Survey continues

(I) Sub-problem Constraints:


P1jkh + B1jkh = E1jkh + mSN(1,j) T1j2mkh
for all j=1.N(1), k=1,K, h=1.H
mRN(i,j)T(i-1)mijkh+Bijkh = Eijkh + mSN(i,j)Tij(i+1)mkh
for all i=2,3 , j=1.N(1), k=1.K and for h=1..H

....(1)
...(2)

QTR4jkh = mRN(4,j)T3m4jkh
for all j,k,h
..(3)
Lower and Upper limits on variables defined
.(4)
(II) Linking Constraints
for all u=1U, j=1M and k=1.K
..(5)
hu *QTR4jkh D4jku
Bij(k+1)h = Eijkh for i=1,2,3 , j=1N(i), k=1.K-1 and h=1H
..(6)
TTijk = xSN(i,j) Tij(i+1)xkh for i=1,2,3 j=1.N(1), k=1..K
..(7)
for all i=1,2,3, j= 1.N(1), k= 1..0.5K
.
ijkh S1ij
(8a)
for all i=1,2,3, j= 1.N(1), k= (0.5K+1)..K
.
ijkh S2ij
(8b)
16

Literature Survey continues


(III) Location Constraints:
Linear constraints representing the condition that if a
warehouse is located at a particular point then the
space booked must be within the permissible lower
and upper limits
...(9)
TTijk , Ti1,j1,i2,j2,k,h , Eijkh , Bijkh , QTR4jkh , P1jkh, S1ij, S2ij 0
..(10)
Lij = (0,1) for all i , j
..(11)
Benders 'Decomposition is used as a solution
procedure
Two Lagrangian relaxation procedures are
developed to solve the primal problem , one of
them converging at faster rate
17

New Formulations developed for


SSCWLP Multi-Commodity case
Multi-Commodity SSCWLP is a family of real life
problems like FMCG, FCI etc.

Strong and weak formulations have been


developed for the general case of the problem
Empirically , with the help of GAMS it is shown
that Strong Constraints forms the superior
bound than the weak Constraints
Higher Bounds leads to greater PRUNING of the
search tree when used in B&B (hence
reduction in savings of CPU time)

18

MODE
L:model we have used the formulation style of
In the
sharma and sharma (2000)
Indexes:
i: Plants
j: Warehouses
k: Markets
m: Commodity
XPWijm : Number of units shipped from plant i to
warehouse j of the commodity m
XWMjkm : Number of units shipped from warehouse j to
market k of commodity m
CPWijm : Cost of shipping of one unit from plant i to
warehouse j of the commodity m
CWMjkm : Cost of shipping of one unit from warehouse j to
market k of the commodity m
Yj : Binary Location variable which is 1 if it is decided to
locate a warehouse at location j and 0 otherwise
19

Model continues

Fj : Fixed cost and maintenance cost for


establishing a warehouse at location j
Sim : Supply capacity of plant i for
commodity m
CAPj : Capacity of the warehouse at
location j , it is assumed that all the
commodity are of same density and
confined the same space
Dkm : Demand for commodity m at market
k
M : A very large number, here taken as
20

Model continues

Objective
Function:

Minimize Z = i,j,m) *(CPWi,j,m)
j,k,m
j

)*(CWMj,k,m) +

*Fj

(O)
Subjected to :
; i,m
(Supply
i,j,m Si,m
Constraints)
i,j,m = j,k,m ; j,m (flow balance constraint)
= Dk,m
; k,m
Constraints)
; j
i,j,m CAPj
j,k,m

(1
)

(Demand
(Capacity

21

Model continues


Linking
Constraints

M*Yj ; j (it will be a very weak


constraint) ..(2)
(3) And (4) are likely to prove
XPWi,j,m Si,m*Yj
; i,m,j
...(3) Constraints
to be STRONG
i,j,m

XWMj,k,m Dk,m*Yj

; k,m,j

(4)

Positive Constraints and relaxations:


XPWi,j,m 0 ; i,j,m (5)
XWMj,k,m 0 ; j,k,m
Yj (0,1)

; j

Formulation I : (O), (1), (2) and (5)

22

Theorem/Proposition
It is proposed that LP relaxation of (II) LP
relaxation of (I) i.e. LP relaxations of formulation
(II) forms a superior bound than LP relaxation of
formulation (I) and therefore reach to optimal
solution in lesser CPU time
Theoretical Proof: from linking constraints [ (3),
(4) and (5)]
Since D's << M and Ss << M;
Therefore feasible region of formulation II <
feasible region of formulation I.
Then for the minimization problem objective
23
function value of formulation II will be higher

Proposition continues
Feasible region for formulation
Feasible region for
(II)
formulation (I)

Suppose A is
the point of
Min for
formulation (I)

Clearly point
A is not in the
fesible region
of formulation
(II)
Optimal value of formulation (II) will be at a
point other than the point A and hence it will
be a greater value than the formulation (I)
optimal value
24

Proposition continues

# Empirical Verification using GAMS :


To prove that formulation (II) is significantly better
than formulation(I) statistically
Samples problems are created of the size 50x50x50
and 100x100x100 for the following four categories as
per input data:
Average Demand for each commodity in each market is
taken Category
as 125 units
Average Warehouse
Average Supply
Capacity
a.

30 % more than the


Average demand

30% more than the


average warehouse
capacity

b.

30 % more than the


Average demand

125% more than the


average warehouse
capacity

c.

125% more than the


Average demand

30% more than the


average warehouse
capacity

25

Proposition continues
Random values are generated for supply, warehouse
capacity, demand , costs of transportation and fixed
cost of warehouse location within the GAMS code
50 Problems are solved for each of the four categories
(total number of problem instances solved = 400, 200
of each of the size 50x50x50 and 100x100x100)
Each of the problem (both formulations) is solved on a
Intel(R) Core(TM) i7-4770S CPU @ 3.10GHz
Objective function values and CPU time for both the
formulation in each of the probelm instance is recorded
Statistical analysis (t-test) is done to check the
superiority of formulation (II) over formulation (I) in
terms of Objective function value and CPU time with a
confidence level of 99.5%
26

Proposition continues
#Statistical Analysis
Hypothesis tests is conducted as follows:
For objective function values
: Percentage improvement in Objective function value of
formulation
(II) over formulation (I)
Null hypothesis,
H0 : = 0
Alternate hypothesis, Ha : > 0
Similarly For CPU Time
t : Percentage increase in CPU time for formulation (II)
over formulation (I)
Null hypothesis,
H0 : t = 0
Alternate hypothesis, Ha : t > 0
27

Proposition continues
#Statistical Results t-values calculated for the
hypothesis tests are tabulated as follows:
Problem
Category

a.
b.
c.
d.

Problem Size
(50x50x50)

Problem Size
(100x100x100)

t-values for
bound
improvemen
t

t-values
increase in
CPU time

t-values for
bound
improvemen
t

t-values
increase in
CPU time

376.21

23.56

492.08

52.38

374.52

313.76

27.16

26.98

307.33

26.23

509.38

43.56

288.51

21.17

407.35

38.28

28

Proposition continues

Critical value for t-stats at = 0.005 is 2.68 for


d.o.f. = 49 and 2.6264 for d.o.f. = 2.6264
Comparing the calculated t-values with critical
t-values we can easily reject null hypothesis for
both (Bound value and CPU time)

29

Summery and Conclusions


Relaxed formulation(II) i.e. Strong Constraints
forms the superior bounds than the Relaxed
formulation(I)i.e. weak constraints
CPU time taken for strong formulations is
significantly more than the CPU time for weak
formulations
Benefit of better bounds using stronger
formulations is nullified by their higher
computational time
We therefore now proceed to make hybrid
formulation for multi-commodity SSCWLP to
overcome the drawback of more computational
time : to obtain the better bounds with lesser
CPU time
30

Hybrid
Formulation
s

31

References
[1] Geoffrion AM and Graves GW (1974). Multicommodity
distribution system design by Benders decomposition.
Management Science 20(5): 822844.
[2] Drezner T Drezner Z and Salhi S (2002). Solving the multiple
competitive facilities location problem. European Journal of
Operational Research 142(1):138151
[3] FCI (Food Corporation of India ) official website., http
://fciweb.nic.in/
[4] Sharma RRK (1996). Food grains distribution in the Indian
context: An operational study, in Tripathi, A. and Rosenhead, J.
(ed.) Operations Research for Development, Ahmedabad: New
Age International Publishers, New Delhi.
[5] Sharma RRK and Berry V (2007). Developing new formulations
and relaxations of single stage capacitated warehouse location
problem (SSCWLP): Empirical investigation for assessing relative
strengths and computational effort. European Journal of
Operational Research
[6] Sharma RRK and Saxena A (2002). Dual based procedure for
the special case of transshipment problem.OPSEARCH (INDIA)
32
[7] Verma P and Sharma RRK (2007). Vertical Decomposition

References
[8] Sharma RRK (1991). Modeling a fertilizer distribution

system. European Journal of Operational Research 51: 2434.


[9] ReVelle CS and Eiselt HA (2005). Location analysis: A
synthesis and survey. European Journal of Operational
Research.
[10] ReVelle CS Eiselt HA and Daskin MS (2008). A
bibliography for some fundamental problem categories in
discrete location science. European Journal of Operational
Research
[11] Kouvelis P Rosenblatt MJ and Munson CL (2004). A
mathematical programming model for global plant location
problem: Analysis and insights. IIE Transactions
[12] Sharma RRK and Sharma KD (2000). A new dual based
procedure for the transportation problem. European Journal of
Operational Research 122(3): 611-624.

33

Thank You
34

You might also like