Professional Documents
Culture Documents
H I G H L I G H T S G R A P H I C A L A B S T R A C T
A R T I C L E I N F O A B S T R A C T
Article history: In municipal solid waste management system, decision makers have to develop an insight into the processes
Received 7 December 2016 namely, waste generation, collection, transportation, processing, and disposal methods. Many parameters
Received in revised form 25 February 2017 (e.g., waste generation rate, functioning costs of facilities, transportation cost, and revenues) in this system
Accepted 25 February 2017 are associated with uncertainties. Often, these uncertainties of parameters need to be modeled under a situ-
Available online 8 April 2017
ation of data scarcity for generating probability distribution function or membership function for stochastic
mathematical programming or fuzzy mathematical programming respectively, with only information of
Editor: Simon Pollard
extreme variations. Moreover, if uncertainties are ignored, then the problems like insufficient capacities of
waste management facilities or improper utilization of available funds may be raised. To tackle uncertainties
Keywords:
of these parameters in a more efficient manner an algorithm, based on interval analysis, has been devel-
Facility location problem
Interval analysis oped. This algorithm is applied to find optimal solutions for a facility location model, which is formulated
Municipal solid waste management to select economically best locations of transfer stations in a hypothetical urban center. Transfer stations
Optimization are an integral part of contemporary municipal solid waste management systems, and economic siting of
Uncertainty transfer stations ensures financial sustainability of this system. The model is written in a mathematical pro-
gramming language AMPL with KNITRO as a solver. The developed model selects five economically best
locations out of ten potential locations with an optimum overall cost of [394,836, 757,440] Rs.1 /day ([5906,
11,331] USD/day) approximately. Further, the requirement of uncertainty modeling is explained based on
the results of sensitivity analysis.
© 2017 Elsevier B.V. All rights reserved.
* Corresponding author at: Centre for Environmental Science and Engineering, Indian Institute of Technology Bombay, Mumbai 400076, India.
E-mail address: skarmakar@iitb.ac.in (S. Karmakar).
URL: http://www.cese.iitb.ac.in/people/facinfo.php?id=skarmakar (S. Karmakar).
1
Rs. is Indian currency and 1 Rs. = 0.015 USD as of 18th October 2016.
http://dx.doi.org/10.1016/j.scitotenv.2017.02.207
0048-9697/© 2017 Elsevier B.V. All rights reserved.
V. Yadav et al. / Science of the Total Environment 603-604 (2017) 760–771 761
1. Introduction (Huang et al., 1995; Wang et al., 2012) and ignorance of uncertainty
may lead to problems like the insufficient capacity of facilities or
Municipal solid waste (MSW) management is an important envi- incomplete collection of MSW generated. The next subsection con-
ronmental concern due to (i) rapid population growth; (ii) increment tinues the discussion on uncertainties in MSW management systems
in MSW generation rate; (iii) environmental protection; (iv) human through a comprehensive literature review.
health risk; and (v) shrinking of disposal site options because of
urbanization (Huang and Chang, 2003; Xu et al., 2009a; Eiselt and 1.2. Accounting for uncertainty
Marianov, 2015). Impacts on human health, ensuring environmental
protection and shrinking of disposal site options or land availability As already discussed, many system parameters in MSW manage-
for disposal sites have ruled out all the possibilities of having small ment system e.g., waste-generation rate, functioning cost of facilities,
landfills and dumping grounds in the vicinity of human settlements. transportation cost, and revenues are associated with uncertainties;
Consequently, large sanitary landfilling has been found to be the only therefore, it is necessary to account for this system with uncertain-
viable option for disposal of MSW, for instance, the number of land- ties of variables and parameters. In system analysis, there are several
fills in USA reduced from 8000 (in 1988) to 1908 (in 2010) (Stevens, approaches to deal with uncertainty such as stochastic program-
2002; Eiselt and Marianov, 2015). The reduced number of disposal ming, fuzzy programming, and interval programming (Shmoys and
sites results longer MSW transportation routes. Larger collection vehi- Swamy, 2006; Chang et al., 2008; Fan et al., 2012).
cles are found to be cost effective for longer transportation of MSW In stochastic programming, uncertainty is usually characterized
as compared to smaller collection vehicles (Eiselt, 2007). Transfer by a probability distribution of the parameters. In several MSW man-
stations (TSs) are junctions of smaller and larger collection vehicles agement studies, stochastic programming models have been devel-
known as primary collection vehicles (PCVs) and secondary collec- oped by Gorelick (1990), Huang et al. (2001), Li and Huang (2006)
tion vehicles (SCVs) respectively. Being a junction of PCVs and SCVs, and Wang et al. (2012). However, the problem in this methodology
locations of TSs are very crucial as far as economical aspect is con- lying in difficulty to generate probability distributions with inade-
cerned. Therefore, locations of TSs should be chosen such that the quate data e.g., daily MSW generation. It may be difficult to state a
overall cost of MSW management system is minimized similar to any reliable probability distribution for MSW generation as it fluctuates
other facility location problems (e.g., fire stations, and warehouses) within a certain interval (Marti, 1990). It has been further observed
(Owen and Daskin, 1998). TSs play an important role in collection that it is extremely hard to solve a large stochastic programming
system such as (i) volume reduction due to compaction process; (ii) with all uncertain data with given probability distributions (Birge
transportation cost reduction; (iii) increment in collection frequency; and Louveaux, 2011).
and (iv) flexibility in siting final disposal/processing facilities (Cui et Fuzziness is another type of uncertainty characteristics and can
al., 2011; USEPA, 2002; Yadav et al., 2016b). This problem statement not be described by probability distributions (Möller et al., 2003).
is graphically depicted in Fig. 1. Additionally, modern TSs are fully In fuzzy mathematics, parameters are considered as fuzzy sets and
closed to prevent the entries of animals and flies, hence maintain a are described by possibility distributions. Fuzzy sets allow its mem-
hygienic condition in the vicinity. Therefore, facility location model bers to have different grades of membership (membership function)
for TSs siting will help to improve the efficiency of collection system, within the interval [0, 1]. These membership functions of fuzzy
while decreasing the costs and health hazards. Yadav et al. (2016a) objective and constraints are determined subjectively by the decision
proposed a facility location model for choosing economically best makers. In past, several studies have proposed MSW management
locations for TSs among the given potential locations with determin- system with the inclusion of fuzzy programming e.g., Fan et al.
istic parameters. This study is an extension of Yadav et al.’s (2016a) (2014), Karadimas et al. (2006), Ojo and Anyata (2009) and Srivas-
model by including the inherent uncertainty of MSW management tava and Nema (2011). However, fuzzy models can not effectively
system’s parameters with a demonstration on a hypothetical case incorporate inherent uncertainties with imprecise coefficients of the
study. Also, an algorithm is developed to account the uncertainties of objective function and constraints (Huang et al., 1993). Also, it is dif-
variables and parameters of a facility location problem using interval ficult to create membership functions with very limited information
analysis approach. The next subsection reviews past efforts made to such as the extremes of waste generation rate (Almeida et al., 2011).
implement facility location models in MSW management systems. This issue leads to difficulties in generating sound decision schemes.
Huang et al. (1992) introduced the applications of algorithms
1.1. Facility location models for MSW management based on grey/interval programming in the field of MSW manage-
ment. In grey or interval programs, the values of parameters vary
Different approaches have been followed by the researchers to within two known extreme values. These algorithms have the advan-
establish a financially sustainable MSW management system for any tage in type of input data requirement over fuzzy and stochastic
city. Fig. 2 describes these different approaches along with their programming such as insufficiency of data for generating probability
specific objectives and representative literatures. These approaches distribution or fuzzy membership function (Sun et al., 2014). Also, an
include routing of vehicles under given circumstances (Beltrami and effective communication of uncertain information can be achieved
Bodin, 1974; Chang et al., 1997; Karadimas et al., 2007; Benjamin using these algorithms (Huang et al., 1997). In last two decades, a
and Beasley, 2010); selection of appropriate locations from a number number of algorithms based on grey programming have been devel-
of given potential locations (facility location problems) and expan- oped and proposed for MSW management systems, for example, grey
sion of size or capacity of certain facility over a defined period of linear programming (Huang et al., 1993), grey dynamic programming
time (Huang et al., 1994b; Nie et al., 2004; Li et al., 2008; Xu et (Huang et al., 1994a), and grey mixed integer linear programming
al., 2009b). Evidently, facility location problems are one of the most (Huang et al., 1995). Further, Karmakar and Mujumdar (2006, 2007)
prevalent approaches to make MSW management system financially used interval grey numbers to address the uncertainty of fuzzy mem-
sustainable (Marks and Liebman, 1970; Gottinger, 1988; Kirca and bership functions and expressed upper and lower bounds of fuzzy
Erkip, 1988; Or and Curi, 1993; Costi et al., 2004; Badran and El- membership functions as interval grey numbers.
Haggar, 2006; Yeomans, 2007; Yadav et al., 2016a). Facility location To solve grey/interval programming problems, the two-step
problems are the broad classes of optimization problems that have method is used. In this method, the parent problem is transformed
been widely studied in operations research fraternity (Hamacher and into two submodels, one for the most favorable case and another for
Drezner, 2002; Melo et al., 2009; Laporte et al., 2015). Often, parame- the least favorable case. It is expected that interval solutions of these
ters of MSW management systems are inherently uncertain in nature submodels include all possible optimal solutions of each and every
762 V. Yadav et al. / Science of the Total Environment 603-604 (2017) 760–771
subproblem. However, Zou et al. (2009), Rosenberg (2009) and Chang Therefore, a novel algorithm is developed to solve a facility location
(2015) questioned the rationalities of the solutions obtained using problem efficiently; which is a mixed integer bilinear problem with
two-step method. Zou et al. (2009) found the problems of infeasi- uncertain decision variables and parameters. The developed method-
bility and nonoptimality in two step method. Also, submodels for ology has been discussed in the forthcoming section.
the most favorable case gives most optimistic solution, whereas sub-
models for the least favorable case gives most pessimistic solution. 2. Methodology
Nevertheless, in practice, decision makers prefer a balanced deci-
sion and not a decision based on extreme conditions (Zou et al., For the development of proposed methodology, the concept of
2009). Further, Rosenberg (2009) found that in solution using two- Bhurjee and Panda (2012) has been broadened to encompass interval
step method, the range of objective function is wider than objective variables with bilinear constraints to solve ‘Facility Location Prob-
function range for best/worst scenarios. Subsequently, Huang and lems’. This approach considers parametric representation of interval
Cao (2011) proposed a three-step method to address the infeasibility programming problems. The objective function of parametric repre-
issue of two-step method. Notwithstanding, the three-step method sentation is aggregated to a deterministic equivalent of parent model
made the problem of nonoptimality more severe (Chang, 2015). To by multiplying with preference functions and then integrating the
overcome all these issues, Bhurjee and Panda (2012) proposed alto- resultant function. Deterministic equivalents are explicit programs
gether a new algorithm for finding optimal solutions based on an which have exactly same set of optimal solutions as parent model
interval programming with uncertain parameters. In this method- (Ziemba and Vickson, 2014). It is also shown that for any preference
ology, parametric form of interval valued functions was considered. functions, obtained solutions will always be optimal.
However, this methodology had a limitation in terms of address- Brief introduction to interval operations is given in this section.
ing integers, bilinear functions and uncertain decision variables. Let [AL , AU ] with AL ≤ AU be an interval on the set of real numbers
R, where AL and AU are lower and upper bounds. If AL = AU = A, where, i is an index for sources i = {1, . . . , N} and j is an index for
then A = [A, A] is called degenerate interval. The set of all closed facilities j = {1, . . . , O}, O
is any natural number between 1 and O.
intervals on R is denoted by I. The product space I × I × . . . × I is [Z L (Xnv , D), Z U (Xnv , D)] is the objective function; n = N ∗ O; Eq. (2),
Eq. (3) and Eq. (5) are inequality, equality and bilinear constraints
Tn times
respectively. Fj , GLi , OCjL , OCjU , [TCijL , TCijU ], ALij , AUij , [BLj , BUj ] and CL are
denoted by In and Cnv ∈ In , Cnv = C1L , C1U , . . . , CnL , CnU . Parametric
representation of any point in [AL , AU ] is defined as at = AL + t(AU − system parameters. XijL , XijU , Dj and Eij are continuous and binary
AL ), t ∈ [0, 1]. Let ∗ ∈ {+, −, • , /} be a binary operation on the set of variables respectively. The feasible set for GFLP is denoted by S, which
real numbers. The binary operation between two intervals [AL , AU ] is the set of (Xnv , D) and satisfies Eqs. (2)–(6). The solution for GFLP is
and BL , BU ∈ I can be defined as follows: defined with respect to partial ordering as follows:
∗
AL , AU BL , BU = at1 ∗ bt2 |t1 , t2 ∈ [0, 1] Definition 2.2. An interval (Xn∗ v , D ) ∈ S is said to be an optimal
solution of GFLP if there is no (Xnv , D) ∈ S with
≡ min at1 ∗ bt2 , max at1 ∗ bt2 .
t1 ,t2 t1 ,t2
/ [BL , BU ]. For k ∈ R,
In the case of division, it is assumed that 0 ∈
∗ ∗
Z L (Xnv , D), Z U (Xnv , D) Z L (Xn∗ U n∗
v , D ), Z (Xv , D )
k AL , AU = {kat |t ∈ [0, 1]} ≡ min(kat ), max(kat ) . ∗ ∗
and Z L (Xnv , D), Z U (Xnv , D) = Z L (Xn∗ U n∗
v , D ), Z (Xv , D ) . (7)
t t
The set of closed intervals I is a partially ordered set (Ishibuchi Based on Definition 2.1 and parametric concepts for intervals, the
and Tanaka, 1990; Moore, 1966). In the partial ordering, two distinct parametric form of GFLP, named as GFLPt , is defined as follows:
intervals are compared with the corresponding element of the inter-
vals. Partial ordering considered in this paper is similar to Bhurjee
O
N
O
N
O
and Panda (2012), which is given as follows: (GFLPt ) min zt (xu , D) Fj ∗Dj + octj ∗xuij + tct
∗xuij
ij
j=1 i=1 j=1 i=1 j=1
(8)
Definition 2.1. For [AL , AU ], [BL , BU ] ∈ I,
N
[AL , AU ] [BL , BU ] if and only if at ≤ bt ∀t ∈ [0, 1], subject to atij ∗ xuij ≤ bt
, j = {1, . . . , O
}, (9)
j
i=1
Theorem 2.1 that for any positive preference functions p1 and p2 , 3. Application on a hypothetical case study
obtained solutions will always be optimal for GFLP.
Since each tj , tij
and uij are mutually independent and lie between This section demonstrates an application of the developed
0 to 1, so the integration of objection function is a function of X , X , L U algorithm to solve a facility location problem for choosing best
and Dj only, that is h(XL , XU , D) (say). Hence GFLPI can be written as locations of TSs on a hypothetical case study in an economically
optimal manner. The hypothetical city covers an area of approxi-
mately 192 km2 . The population of this city is estimated to be 1.8
min h(X L , X U , D).
(X L ,X U ,D)∈S million by the end of year 2035, as time horizon of this study is
20 years. It is assumed that the population is uniformly distributed
across the city. MSW generation rate of the city will be [0.4, 0.5]
Theorem 2.1. If (XL* , XU* , D∗ ) ∈ S is an optimum solution of GFLPI , then
∗ kg/capita/day. MSW of city is composed of [40, 45]% compostable
v , D ) ∈ S is an optimal solution of GFLP for any preference functions
(Xn∗
materials and [15, 22]% recyclables. The city already has a land-
p1 and p2 .
fill site and a composting plant. These facilities are 20 km away
from the city boundary as shown in Fig. 3. Now, the municipal cor-
Proof. Let (XL* , XU* , D∗ ) ∈ S be an optimum solution of GFLPI . Assume poration wants to have TSs in the city. The municipality has also
that (Xn∗ ∗
v , D ) is not an optimal solution of GFLP, then there is some identified ten potential locations for TSs (see red stars in Fig. 3).
(Xv , D) ∈ S with satisfying the relation (7). From the definition of
n
However, as far as MSW generated in the city is concerned, ten
partial ordering in Definition 2.1, relation (7) becomes for some TSs are not needed for this city. Therefore, it is required to choose
(XL , XU , D) ∈ S best locations and number for TSs among given ten potential loca-
tions. The proposed facility location model will minimize overall
zt (xu , D) ≤ zt (x∗u , D∗ ), ∀t, u (14) cost of MSW management system by choosing the best location
for TSs among the potential locations. The next subsection pro-
and for at least one t̂ = t and û = u, vides the details of the proposed MSW management system with
TSs.
zt̂ (xû , D) = zt̂ (x∗û , D∗ ). (15)
For real valued functions p1 : [0, 1]2n → R+ and p2 : [0, 1]n → R+ , 3.1. Proposed MSW management scheme
the above relations (14) and (15) imply that for some (XL , XU , D) ∈ S
Fig. 4 describes the proposed scheme for MSW flow in the study
such that
area. The proposed scheme for MSW collection in the hypothetical
city includes (i) transportation of segregated MSW from sources (i.e.,
p1 (t, t
)p2 (u)zt (xu , D) dtdt
du < p1 (t, t
)p2 (u)zt (xu , D) dtdt
du. residential, commercial and institutional) to collection points in the
3n 3n
door to door collection manner using pre-primary collection vehicles
That is, for some (XL , XU , D) ∈ S, (PPCVs); (ii) transportation of accumulated MSW to TSs using PCVs;
(iii) transportation of MSW to composting plant or landfill using
SCVs; and (iv) transportation of composting plant residuals to landfill
h X L , X U , D < h X L∗ , X U∗ , D∗ .
using special collection vehicles (SPCVs). Mathematical representa-
tion of cost associated with these four stages, fixed and variable
This is the contradiction that (XL* , XU* , D∗ ) ∈ S is an optimum costs of existing and proposed facilities are given in subsequent
∗
v , D ) is an optimal solution of GFLP.
solution of GFLPI . Hence (Xn∗ paragraphs.
MSW generated at each source will be segregated as compostable
The proposed algorithm is summarized using pseudo-code for- waste, recyclables and waste to landfill. Recyclables will be col-
mat in Algorithm 1. The application of proposed algorithm on MSW lected periodically by private firms from all the generated sources,
management system of a hypothetical urban center is provided in and the municipal corporation is not responsible for this collection.
the subsequent section. MSW generators are encouraged to segregate recyclables at source as
recyclables are associated with direct revenues. MSW collector will
collect rest of the waste from doorstep using PPCVs on a daily basis.
PPCVs are small vehicles which can move inside narrow streets and
Algorithm 1. Steps of proposed algorithm in pseudo-code format. may be manual or automated. To avoid the commingling of MSW,
these PPCVs have inbuilt partitions. However, the distance traveled
by PPCVs will vary with the urbanization of the city. Also, it is dif-
ficult to generate geo-spatial data for narrow streets of the city. To
address this issue, the whole city is assumed to consist of Thiessen
polygons with respect to each and every collection point. Thiessen
polygons provide an individual area of influence around each and
every point of consideration (Okabe et al., 2009). In this case, the area
of influence of an individual polygon implies area of MSW collection
using a specific PPCV. PPCVs collect MSW from each polygon and
bring MSW to their respective collection points. In order to ensure
p
an unsophisticated analysis, the average of all the perimeters pi
of Thiessen polygons is further equated to the perimeter of a square
N
p
grid (ps ) i.e., pi ≈ N ∗ ps , where N is the total number of collection
i=1
points or Thiessen polygons. These tiny square grids have collection
points at the intersection of diagonals (see Fig. 5). Then, maximum
distance traveled by any PPCV will be equivalent to the perimeter
V. Yadav et al. / Science of the Total Environment 603-604 (2017) 760–771 765
Fig. 3. Map of study area with potential locations for transfer stations and MSW collection points.
¯ of these squares. Therefore, transportation cost using PPCVs
CP SCVs. SCVs are large trucks with the capacity of 15–20 tons. Com-
will be postable fraction of MSW will be transferred to SCVs without com-
paction as compostable waste is highly dense and do not require any
compacting. Also, compacted compostable waste creates an anaero-
N bic environment and requires a longer time period for final product
¯ TCP L , TCP U
CP GLi , GUi (16) (Tuomela et al., 2000). Transportation cost from TSs to landfill or
i=1
composting plant using SCVs will be
Fig. 4. Proposed MSW management scheme with transfer stations in the hypothetical case study.
facilities (landfill and composting plant) and establishment cost of where, Fj is the establishment cost for the TS j, in Rs./day; OCjL , OCjU ,
TSs:
OCkL , OCkU and OClL , OClU are operational cost for the TS j, com-
O
N
O
posting plant k and landfill l respectively in Rs./ton; Dj is the binary
Fj ∗ D j ⊕ [OCjL , OCjU ] [XijL , XijU ]
j=1 i=1 j=1
variable (=0 or 1), which is to take one if a TS is to be set up at
candidate location j (= 1, 2, . . . , O).
P
O O
P
Q
Therefore, the objective function ([ZL , ZU ]) i.e., the total MSW
⊕ OCkL , OCkU Xjk
L U
, Xjk ⊕
management cost (in Rs./day) for this model will be
j=1 k=1 j=1 k=1 l=1
OClL , OClU XjlL , XjlU ⊕ Xkl
L U
, Xkl (20) min[Z L , Z U ] = Eq.(16) ⊕ Eq.(17) ⊕ Eq.(18) ⊕ Eq.(19) ⊕ Eq.(20) (21)
V. Yadav et al. / Science of the Total Environment 603-604 (2017) 760–771 767
p
Fig. 5. Transportation of MSW using pre-primary collection vehicles (PPCVs); where pi is the perimeter of ith Thiessen polygon, ps is the perimeter of any square grid, and N is
the number of collection points.
The developed model includes following constraints: where, CTSj is the capacity of TS j in ton/day. These con-
straints also confined that TS j will set up iff it is selected by
1. Demand constraint i.e., all the MSW collected at collection model.
point i is equal to the MSW transported to TS j:
O
U
O
Xjk ≤ CCPk ∀k = 1, 2, . . . , P (28)
XijL , XijU = GLi , GUi ∀i = 1, 2, . . . , N (22) j=1
j=1
where, CCPk is the capacity of composting plant k in ton/day.
4. If collection point i chooses j as its TS, then all the MSW col-
where, GLi , GUi
is the amount of daily MSW collected at
lected at collection point i has to be transferred to TS j only.
collection point i.
2. Mass balance constraints i.e., the amount of daily MSW in
GLi ∗ Eij ≥ XijL ∀ i = 1, 2, . . . , N, j = 1, 2, . . . , O (29)
and out of TSs and composting plants should be equal:
N
P
Q GUi ∗ Eij ≥ XijU ∀ i = 1, 2, . . . , N, j = 1, 2, . . . , O (30)
XijL , XijU = L
Xjk U
, Xjk ⊕ XjlL , XjlU ∀j = 1, 2, . . . , O
i=1 k=1 l=1
O
(23) Eij = 1 ∀ i = 1, 2, . . . , N (31)
j=1
P
N
L U
Xjk , Xjk = aL , aU XijL , XijU ∀j = 1, 2, . . . , O (24) where, Eij is the binary variable which takes one if TS
k=1 i=1 (j = 1, 2, . . . , O) has been chosen for collection point (i =
Q
O 1, 2, . . . , N).
L U
Xkl , Xkl = bL , bU L
Xjk U
, Xjk ∀k = 1, 2, . . . , P (25) 5. Constraint on maximum number of TSs i.e., the number of
l=1 j=1 TSs should not exceed the number of potential locations for
TSs CL:
where, [a L , a U ] is the compostable fraction of MSW gener-
ated in the city and [bL , bU ] is the fraction of compostable
O
Dj ≤ CL (32)
waste to the residual of composting plant k.
j=1
3. Capacity constraints i.e., the quantity of MSW transported to
TSs, composting plant does not exceed their capacities:
6. Non-negativity constraint for amount of MSW need to be
collected:
N
XijU ≤ CTSj ∗ Dj ∀j = 1, 2, . . . , O (26)
i=1 XijL , Xjk
L
, XjlL , Xkl
L
≥ 0; ∀i, j, k, l (33)
P
Q
U
Xjk + XjlU ≤ CTSj ∗ Dj ∀j = 1, 2, . . . , O (27) The proposed model is clearly a facility location model with
k=1 l=1 uncertain parameters and variables. The uncertain attributes of MSW
768 V. Yadav et al. / Science of the Total Environment 603-604 (2017) 760–771
subject to constraints
O
O
XijL = GLi ; XijU = GUi , ∀i = 1, 2, . . . , N (34)
j=1 j=1
N
P
Q
XijL = L
Xjk + XjlL , ∀j = 1, 2, . . . , O (35)
i=1 k=1 l=1
N
P
Q
XijU = U
Xjk + XjlU , ∀j = 1, 2, . . . , O (36)
i=1 k=1 l=1
P
N
L
Xjk = aL ∗ XijL ∀ j = 1, 2, . . . , O (37)
k=1 i=1
P
N
U
Xjk = aU ∗ XijU ∀ j = 1, 2, . . . , O (38)
Fig. 6. Mosaic plot describing all the uncertain parameters with their respective k=1 i=1
sources of uncertainty (this figure is not to scale).
Q
O
L
Xkl = bL ∗ L
Xjk ∀ k = 1, 2, . . . , P (39)
l=1 j=1
the capacity of TSs should be neither very small (increases the over- Dj ≤ CL, (46)
j=1
all cost), nor very large (makes TSs crowded with vehicles, and hence
reduces the collection efficiency). Consequently, the capacity for a XijL , Xjk
L
, XjlL , Xkl
L
≥ 0; ∀ i, j, k, l (47)
TS has been chosen as 250 TPD based on the methodology given in
Yadav et al. (2016a,b). Fixed and operational costs of TSs and opera-
where ℵ is total number of parameters in zt (xu ),
tional costs of existing facilities are taken as the typical cost values of
Indian urban centers. The MSW transportation costs along with the
N
N
O
layout of the study region are determined using geographic infor- ¯ ∗ TCP(t1 )
zt (xu ) =CP Gi (ti ) + TCij (tij )Xij (uij )
mation system (GIS). GIS is a digital computer application, which i=1 i=1 j=1
is designed to capture, storage, analysis, and display of geospatial
Eq. (16) Eq. (17)
data. GIS further provides an ability to map, measure distances and
accumulation of all different associated information (Fischer, 2006).
O
P
O
Q
+ TCjk (tjk )Xjk (ujk ) + TCjl (tjl )Xjl (ujl )
The data compiled is used as parameter values of the mathemati-
j=1 k=1 j=1 l=1
cal model (see Eqs. (16)–(33)). This model is solved using developed
algorithm (see Section 2). Results and analysis of results generated Eq. (18)
subsequently.
O
P
Q
+ OCl (tl )(Xjl (ujl ) + Xkl (ukl ))
j=1 k=1 l=1
Minimize Z(X) = p1 (t)p2 (u)zt (xu )dtdu
ℵ Eq. (20)
V. Yadav et al. / Science of the Total Environment 603-604 (2017) 760–771 769
Fig. 7. Economically best locations for transfer stations in the hypothetical case study.
This deterministic equivalent for the hypothetical city is written For this analysis, parametric representation of uncertain parameters
in a mathematical programming language (AMPL) (Gay, 2015) and is considered i.e., for any uncertain parameter AL , AU ∈ I, [AL , AU ] =
L U
solutions are found using KNITRO 5.2 as solver (Byrd et al., 2006). {at = (1 − t) ∗ A + t ∗ A | t ∈ [0, 1]} (see Section 2). The sensi-
This model is solved on Passpoli server of Industrial Engineering and tivity index of any uncertain parameter is equal to the slope of their
Operation Research, IIT Bombay with Debian 8.1 operating system. respective linear function of sensitivity parameter t. The normal-
The deterministic equivalent has 39,990 variables and 44,015 con- ized sensitivity index of each uncertain parameter is given in Fig. 8.
straints. The solutions of the proposed model provides economically Higher value of sensitivity index implies higher sensitivity of param-
best locations for TSs, details of the amount of MSW has to be trans- eter. As can be seen in Fig. 8, MSW collected at collection point i is
ported using PCVs and SCVs in an economically optimal manner and the most sensitive parameter of the model. This is intuitive as higher
minimized overall daily cost for MSW management in the hypotheti- collection of MSW will lead to higher costs for transportation and
cal city. As discussed in Section 3, the municipality has identified ten processing. Compostable fraction of MSW is the second most sensi-
potential locations for TSs siting. The model chooses five best loca- tive parameter as slight fluctuation in this fraction requires a special
tions out of given ten potential locations. These economically best attention in terms of transportation and processing. The operating
five locations can be seen in Fig. 7. The optimum overall cost for the cost of landfill is the most sensitive among all of the operating costs
MSW management in the hypothetical city is found to be [394836, as the major fraction of the MSW is being sent to landfill. This is a
757440] Rs./day ([5906, 11331] USD/day) approximately. Further, to prevalent situation in many developing countries.
find the scope of having uncertainty analysis of proposed model a
comprehensive sensitivity analysis is performed.
Huang, G.H., Baetz, B.W., Patry, G.G., 1994a. Grey fuzzy dynamic programming: Okabe, A., Boots, B., Sugihara, K., Chiu, S.N., 2009. Spatial Tessellations: Concepts and
application to municipal solid waste management planning problems. Civil Engi- applications of Voronoi Diagrams. John Wiley & Sons.
neering Systems 11 (1), 43–73. Or, I., Curi, K., 1993. Improving the efficiency of the solid waste collection system
Huang, G.H., Baetz, B.W., Patry, G.G., 1994b. Waste ow allocation planning through a in Izmir, Turkey, through mathematical programming. Waste Manag. Res. 11,
grey fuzzy quadratic programming approach. Civ. Eng. Syst. 11, 209–243. 297–311.
Huang, G.H., Baetz, B.W., Patry, G.G., 1995. Grey integer programming: an application Owen, S.H., Daskin, M.S., 1998. Strategic facility location: a review. Eur. J. Oper. Res.
to waste management planning under uncertainty. Eur. J. Oper. Res. 83, 594–620. 111, 423–447.
Huang, G.H., Huaicheng, G., Guangming, Z., 1997. A mixed integer linear program- Rosenberg, D.E., 2009. Shades of grey: a critical review of grey-number optimization.
ming approach for municipal solid waste management. J. Environ. Sci. China 9, Eng. Optim. 41, 573–592.
431–445. Saltelli, A., Tarantola, S., Campolongo, F., Ratto, M., 2004. Sensitivity Analysis in
Ishibuchi, H., Tanaka, H., 1990. Multiobjective programming in optimization of the Practice: A Guide to Assessing Scientific Models. John Wiley & Sons.
interval objective function. Eur. J. Oper. Res. 48, 219–225. Shmoys, D.B., Swamy, C., 2006. An approximation scheme for stochastic linear
Karadimas, N., Loumos, V., Orsoni, A., 2006. Municipal solid waste generation mod- programming and its application to stochastic integer programs. J. ACM 53,
elling based on fuzzy logic. 20th European Conference on Modelling and Simu- 978–1012.
lation: Modelling Methodologies and Simulation Key Technologies in Academia Srivastava, A., Nema, A., 2011. Fuzzy parametric programming model for integrated
and Industry. pp. 309–314. solid waste management under uncertainty. J. Environ. Eng. 137, 69–83.
Karadimas, N.V., Papatzelou, K., Loumos, V.G., 2007. Genetic algorithms for munic- Stevens, E.S., 2002. Green Plastics: An Introduction to the New Science of Biodegrad-
ipal solid waste collection and routing optimization. Artificial Intelligence and able Plastics. Princeton University Press.
Innovations 2007: From Theory to Applications. Springer., pp. 223–231. Sun, W., An, C., Li, G., Lv, Y., 2014. Applications of inexact programming methods
Karmakar, S., Mujumdar, P., 2006. Grey fuzzy optimization model for water quality to waste management under uncertainty: current status and future directions.
management of a river system. Adv. Water Resour. 29, 1088–1105. Environ. Syst. Res. 3, 1–15.
Karmakar, S., Mujumdar, P., 2007. A two-phase grey fuzzy optimization approach for Tuomela, M., Vikman, M., Hatakka, A., Itävaara, M., 2000. Biodegradation of lignin in a
water quality management of a river system. Adv. Water Resour. 30, 1218–1235. compost environment: a review. Bioresour. Technol. 72 (2), 169–183.
Kirca, Ö., Erkip, N., 1988. Selecting transfer station locations for large solid waste USEPA, 2002. Waste Transfer Stations: A Manual for Decision-making. . United States
systems. Eur. J. Oper. Res. 35, 339–349. Environmental Protection Agency (EPA).,
Laporte, G., Nickel, S., da Gama, F.S., 2015. Location Science. Springer. Wang, S., Huang, G., Yang, B., 2012. An interval-valued fuzzy-stochastic programming
Li, Y., Huang, G., 2006. Minimax regret analysis for municipal solid waste manage- approach and its application to municipal solid waste management. Environ.
ment: an interval-stochastic programming approach. J. Air Waste Manage. Assoc. Model Softw. 29, 24–36.
56, 931–944. Xu, Y., Huang, G., Qin, X., Cao, M., 2009a. SRCCP: A stochastic robust chance-con-
Li, Y., Huang, G., Yang, Z., Nie, S., 2008. An integrated two-stage optimization model for strained programming model for municipal solid waste management under
the development of long-term waste-management strategies. Sci. Total Environ. uncertainty. Resour. Conserv. Recycl. 53, 352–363.
392, 175–186. Xu, Y., Huang, G., Qin, X., Huang, Y., 2009b. a stochastic robust fuzzy interval linear
Marks, D.H., Liebman, J., 1970. Mathematical analysis of solid waste collection. Pub- programming model for municipal solid waste management under uncertainty.
lic Health Service Publication. vol. 2104. Department of Health, Education and J. Environ. Inform. 14, 72–82.
Welfare, Washington, D.C. Yadav, V., Karmakar, S., Dikshit, A.K., Vanjari, S., 2016a. A feasibility study for the
Marti, K., 1990. Stochastic programming: numerical solution techniques by semi-s- locations of waste transfer stations in urban centers: a case study on the city of
tochastic approximation methods. Stochastic Versus Fuzzy Approaches to Multi- Nashik, India. J. Clean. Prod. 126, 191–205.
objective Mathematical Programming Under Uncertainty. Springer., pp. 23–43. Yadav, V., Karmakar, S., Dikshit, A.K., Vanjari, S., 2016b. Transfer stations siting in india:
Melo, M.T., Nickel, S., Saldanha-da Gama, F., 2009. Facility location and supply chain a feasibility demonstration. Waste Manag. 47, I–IV.
management-a review. Eur. J. Oper. Res. 196, 401–412. Yeomans, J.S., 2007. Solid waste planning under uncertainty using evolutionary simu-
Möller, B., Graf, W., Beer, M., 2003. Safety assessment of structures in view of fuzzy lation-optimization. Socio Econ. Plan. Sci. 41, 38–60.
randomness. Comput. Struct. 81, 1567–1582. Ziemba, W.T., Vickson, R.G., 2014. Stochastic Optimization Models in Finance. Aca-
Moore, R.E., 1966. Interval Analysis. Prentice-Hall Englewood Cliffs. demic Press.
Nie, Y., Li, T., Yan, G., Wang, Y., Ma, X., 2004. An optimal model and its application for Zou, R., Liu, Y., Liu, L., Guo, H., 2009. REILP approach for uncertainty-based decision
the management of municipal solid waste from regional small cities in China. J. making in civil engineering. J. Comput. Civ. Eng. 24 (4), 357–364.
Air Waste Manage. Assoc. 54, 191–199.
Ojo, O., Anyata, B., 2009. Fuzzy based solid waste management method. Adv. Mater.
Res. 62-64, 728–735.