Available online at www.sciencedirect.

com

Systems Engineering Procedia 5 (2012) 137 – 143

International Symposium on Engineering Emergency Management 2011

Sample Average Approximation Method for Chance Constrained

Stochastic Programming in Transportation Model of Emergency
Management
Deng Chunlin* Yang Liu
School of Public Administration, Xiangtan University, Xiangtan 411105, P.R.China
Department of Mathematics and Computational Science, Xiangtan University, Xiangtan 411105, P. R. China

Abstract

In this paper, an optimal model for the transportation of emergency resource is established on chance
constrained stochastic programming. We use Conditional Value at Risk (CVaR) to approximate the chance
constraint, and we get the approximation problem of the chance constrained stochastic programming by using the
sample average approximation (SAA) method. For a given sample, the SAA problem is a deterministic nonlinear
programming (NLP) and any appropriate NLP code can be applied to solve the problem. The model and method
provide a new way for the emergency logistics management engineering.

©
© 2012 PublishedbybyElsevier
2011 Published Elsevier
Ltd.Ltd. Selection
Selection and peer-review
and peer-review under responsibility
under responsibility of Desheng
of Desheng Dash Wu Dash Wu.

Keywords: emergency managemen engineering; chance constraint; Conditional Value-at-Risk; sample average approximation; transportation
model

1. Introduction

In recent years, frequent emergencies have brought immeasurable losses to our country. Public health
emergency events such as the SARS outbreak in 2003, the bird flu outbreak in 2005 and the swine flu outbreak in
2009; major traffic accidents such as Jiaoji Railway trains collision in 2008, Wenzhou Railway High-speed rear-
end events in 2011; major geological disasters such as Wenchuan earthquake in 2008 and so on, these emergencies
have caused significant casualties and major property losses, damaged the ecological environment and caused
serious harm to the society. Emergencies occur along with a large number of emergency material needs. How to
use reasonable transportation routes and establish the optimal programming of emergency material scheduling in
delivering the emergency material resources to the demand points in time directly affects the effectiveness of
public incident rescue operation. The Research on emergency resources transportation can ensure the effective

2211-3819 © 2012 Published by Elsevier Ltd. Selection and peer-review under responsibility of Desheng Dash Wu.
doi:10.1016/j.sepro.2012.04.022
138 Deng Chunlin and Yang Liu / Systems Engineering Procedia 5 (2012) 137 – 143

allocation and adequate supply of emergency resources transportation and can minimize the losses from sudden
events.
Experts at home and abroad have studied the emergency resources transportation problems from different
perspective, different objective functions and different constraint conditions. Knott (1988) established a linear
programming model for a transportation problem of rescue food [1]. The aim of the rescue material transportation
model was to provide easily understood decision tools, help the operation personnel to make decision, and provide
the optimal combination on food, trucks and camps. George (1998) studied the transportation optimization model
for radioactive dangerous articles and introduced the emergency problem in the research, but he only considered
the nearest emergency locations to participate the emergency rescue, and he did not consider that the single saving
point would not meet the demand of emergency [2]. Linet (2004) established an emergency relief goods
transportation model based on integer programming [3]. He also proposed an algorithm to solve the model and
used examples to verify the validity of the algorithm. Shi Qin and Huang Zhipeng (2006) considered a multi-
objective transportation optimization model in [4] and they mainly considered minimizing the total transportation
expense, i.e. minimizing the largest one-way expense from the supply stations to the demand points. They also
provided a method for solving the optimization model with non-dominated solutions.
In this paper, we consider the problem of the rescue material transportation model in the emergency
management engineering. Because there are so many uncertain parameters such as the time, the area, the number of
the emergency resources, etc. which we should consider in the material transport model of emergency management,
we will use the related theory of the chance constrained stochastic programming in the description of the model.
Conditional Value at Risk (CVaR) will be used to approximate the chance constraint and a sample average
approximation (SAA) method is applied to a CVaR approximated chance constrained stochastic minimization
problem. Specifically, the optimal solutions and stationary points obtained from solving sample average
approximated problems converge with probability one (w.p.1) to their true counterparts. For a fixed sample, the
SAA problem is a deterministic NLP and there are many existed NLP code can be applied to solve the problem.
The rest of this paper is organized as follows. In Section 2, we discuss the chance constrained programming
and the handling of the chance constraints. In Section 3, we establish the mathematical model of the emergency
resources transportation problems. In Section 4, we give the approximation problem of the chance constrained
stochastic programming by using the sample average approximation (SAA) method. Conclusion is given in
Section 5.

2. Chance constrained stochastic programming and the handling of the joint chance constraints

Although many ways have been proposed to model uncertain quantities, stochastic models have proved their
flexibility and usefulness in diverse areas of science and engineering. This is mainly due to solid mathematical
foundations and theoretical richness of the theory of probability and stochastic processes, and to sound statistical
techniques of using real data. Optimization problems involving stochastic models occur in almost all areas of
science and engineering, from telecommunication and medicine to finance. This stimulates interest in rigorous ways
of formulating, analyzing, and solving such problems. Due to the presence of random parameters in the model, the
theory combines concepts of the optimization theory, the theory of probability and statistics, and functional analysis.
Moreover, in recent years the theory and methods of stochastic programming have undergone major advances[5-12].
In this paper, we will consider the problem of the rescue material transport model with uncertainties in the
emergency management. The main difference in our mode is that we use the chance constraints in the description of
the stochastic demand for the rescue material transport model.
2.1 Chance Constraint
Consider the following chance constrained minimization problem[12]:
min f ( x) s.t. P( x) d D , ˄1˅
x X

where X  R n is a closed set, f : R n o R is a continuous function, D  (0,1) is a given significance level, and
 Deng Chunlin and Yang Liu / Systems Engineering Procedia 5 (2012) 137 – 143 139

P ( x) : Pr{c( x, [ ) ! 0} ˄2˅
is the probability that the constraint is violated at point x  X . We assume that [ is a random vector, whose
probability distribution P is supported on set ;  R d , and the function c : R n u ; o R
is a Carathéodory function.
The chance constraint P(x) d D can be written equivalently in the form [7,12]




























































Pr{c( x, [ ) d 0} t 1  D .


















































˄3˅

Let us also remark that several chance constraints

Pr{ci ( x, [ ) d 0, i 1, K , q} t 1  D








































































˄4˅

can be reduced to one chance constraint (3) by employing the max-function

c( x, [ ) : max ci ( x, [ ). ˄5˅
1d i d q

2.2 Approximation of the chance constraint

It is well-known that the stochastic programming with chance constraints is difficult to solve as the chance
constraint function (5) is generally non-convex, has no closed form and is difficult to evaluate. Various approaches
have been proposed in the literature to address these difficulties. For example, a number of convex conservative
approximation schemes have been proposed, including the quadratic approximation [5], the CVaR approximation [6]
and the Bernstein approximation [7]. These approximation schemes typically find a feasible suboptimal solution to
the stochastic programming with chance constraints. CVaR approximation is one of the most widely used
approximation schemes since it is numerically tractable and enjoys nice features such as convexity and monotonicity.
Our focus here is on CVaR approximation. Recall that Value-at-Risk (VaR) of a random function c( x, [ ) is defined
as
VaR1D (c( x, [ )) : min{K : Pr ob{c( x, [ ) d K} t 1  D }
KR
and CVaR is defined as the conditional expected value of c( x, [ ) exceeding VaR, that is,
1
CVaR1D (c( x, [ )) : ³c ( x ,[ ) tVaR1D (c ( x,[ )) c( x, [ ) P (d[ ).
D
The latter can be reformulated as
1
min (K  E[(c( x, [ )  K )  ]),
KR D
see [6]. In the case when c( x, [ ) is convex at x for almost every [ , CVaR1D (c( x, [ )) is a convex function, see [6,
Theorem 2]. Observe that the chance constraint (4) can be written as VaR1D (c( x, [ )) d 0 while CVaR is often
regarded as an approximation of VaR. Therefore we may consider the following approximation scheme for (4) by
replacing the change constraint with a CVaR constraint
CVaR1D (c( x, [ )) d 0. (6)

An obvious advantage of (6) is that the constraint function is convex as long as c( x, [ ) is convex. This kind of
convex approximation scheme was firstly proposed by Nemirovski and Shapiro [7] and has now been widely
applied in stochastic programming. It is easy to verify that (6) can be reformulated as
1
H ( x,K ) : K  E[(c( x, [ )  K )  ] d 0, K  R, (7)
D
see for instance [8, Section 1.4].
The main challenge here is to handle the expected value E[(c( x, [ )  K )  ] . If we are able to obtain a closed
form of the function, then problem becomes an ordinary nonlinear constraint. However, in practice, it is often
difficult to do so not only because there is a max operation on the integrand but also it requires complete information
140 Deng Chunlin and Yang Liu / Systems Engineering Procedia 5 (2012) 137 – 143

on the distribution of [ or multidimensional integration. A well known method to tackle this problem is sample
average approximation (SAA) method which is also known wider various names as Monte Carlo method, sample
path optimization (SPO) method [ 9] and stochastic counterpart, see [10] for a comprehensive review. The basic idea
of SAA can be described as follows. Let [ 1 , K , [ N be an independent and identically distributed (i.i.d.) sampling of
[ . We can get the following sample average approximation problem for (7) just like the method in [10]:
1 N j
H N ( x, K ) : K 
¦ (c( x, [ )  K )  d 0. (8)
DN j 1
We call (8) is the SAA approximation of (7). Obviously, (8) is easy to calculate when the sample is given.

3. Model establishment

The main objective of this paper is to establish an appropriate mathematical model which contains uncertainties
for the emergency rescue materials transportation and distribution problems, and provide the corresponding
solutions. We hope that the new model and method will provide decision support tools for emergency management
personnel in sudden incident emergency relief supplies and help them establish emergency logistics plans. So that
the vehicle will select the sequence and a specific path from the emergency logistics center transport relief supplies
to the demand points in the shortest possible time and as little as possible cost to meet the needs of the local disaster
rescue. This will also minimize the losses from the sudden events in the end.
In this paper, we will establish the mathematical model for the emergency relief materials transportation and
distribution problem which is a comprehensive transportation network problem consisting of professional enterprise
or disaster relief organizations, warehouses of special equipment or material reserve specifically for disaster relief,
warehouse of production of special materials and equipment for relief of enterprise, the disaster site, the existing
land transportation network and so on.
3.1 Assumptions in the model:
1. Assume that the emergency relief materials transportation network includes three main parts: production and
supply enterprises for emergency material and equipment, professional warehouse for disaster relief and
disaster materials reserve, distributing center for scene of the accident of material.
2. Assume that the shortest paths to every destination are known.
3. Assume that all uncertainties in the model will affect the total demand for the emergency material and
equipment on the site of distributing center, so the total demand in the model is a random vector.
3.2 The related symbols in the model :
m : the number of production enterprises or warehouse for relief of disaster supplies and equipment or
location;
n : the number of special warehouse for disaster relief and disaster reserve or location;
k : the number of the scene of the accident material distribution center;
i : relief of disaster supplies and equipment types;
i
xmn : transportation volume from the enterprise to the special warehouse of supplies equipment, a decision
variable;
lmn : the shortest path from the enterprise transport to the special warehouse of supplies equipment, a known
quantity;
i
ymk : transportation volume from the enterprise to the scene of the accident distributing center of goods and
materials, a decision variable;
pmk : the shortest path from the enterprise to the scene of the accident distributing center of goods and materials,
a known quantity;
 Deng Chunlin and Yang Liu / Systems Engineering Procedia 5 (2012) 137 – 143 141

i
z nk : transportation volume from the special warehouse to the scene of the accident distributing center of goods
and materials, a decision variable;
qnk : the shortest path from the special warehouse to the scene of the accident distributing center of goods and
materials, a known quantity;
i
pm : production or storage quantity of relief of disaster supplies and equipment production enterprises;
Qni : general reserve of special warehouse supplies or equipment for disaster relief disaster reserve, a known
quantity;
Rki : total demand for the scene of the accident distributing center of goods and materials supplies or equipment,
a random variable.
In the total system of the emergency relief materials transportation and distribution, in addition to the accident
brings uncertainties, there are many other uncertain factors we should consider in the mathematical model. For
example, the material and equipment reserve in special equipment warehouse may not be fully able to meet all
kinds of requirement of the accidents, one should spend a large amount of cash on reserving the equipment and
regular maintenance, but there is no enterprise that are willing to take this responsibility. So the actual decision-
making is an optimization problem of uncertainties and allows violating the tolerance constraints to a certain extent.
Therefore, we establish the following stochastic programming model combined chance constraints:
i i i
min ¦ (¦ ¦ x mn l mn  ¦ ¦ y mk p mk  ¦ ¦ z nk q nk )
i m n m k n k

i i
s.t. ¦ (¦ x mn  ¦ y mk ) d ¦ p mi ,
i n k i
i i
¦ ¦ z nk d ¦ Q n , ˄9˅
i k i
i i
Pr{¦ (¦ y mk  ¦ z nk ) t ¦ R ki  H ki } t 1  D , k ,
i m n i
i i i
t 0,
x mn t 0, y mk t 0, i, m, n, k ,
z nk
where the indivisible resources such as equipment, personnel are integers, the other resources which can be
classified in terms of resources are the non-negative real numbers, H ki t 0 is the random disturbance, R ki is a
random vector, D is the confidence level.
The main difference between our model and the traditional transportation model is two-fold: one is that we use
the total transportation freight tonne kilometer as the goal in the objective function and the other is that we use the
chance constraint to explicit the uncertainties. In addition, the first and the second constraints in our model expresses
that the out amount does not exceed the available inventory, the third constraint expresses that we should meet the
need of decision-making in a given confidence level.

4. Sample average approximation method for the joint chance constrained stochastic programming

Now we consider the solution of model (9). According to Section 2, we can give the approximation problem of
(9) by using the sample average approximation (SAA) method :
i i i
min ¦ (¦ ¦ x mn l mn  ¦ ¦ y mk p mk  ¦ ¦ z nk q nk )
i m n m k n k

i i
s.t. ¦ (¦ x mn  ¦ y mk ) d ¦ p mi ,
i n k i
i i
¦¦ z nk d ¦ Qn , ˄10˅
i k i
142 Deng Chunlin and Yang Liu / Systems Engineering Procedia 5 (2012) 137 – 143

1 N i i i i
K ¦ (¦ R k  H k ¦ (¦ y mk  ¦ z nk )  K )  d 0, k ,
DN j 1 i i m n
i i i
x mn t 0, y mk t 0, z nk t 0, i, m, n, k ,
We refer to (9) as the true problem and (10) as its sample average approximation (SAA) problem. From [13],
we know that the optimal solutions and stationary points obtained from solving sample average approximated
problems converge with probability one (w.p.1) to their true counterparts. And for a fixed sample, (10) is a
deterministic NLP and therefore any appropriate NLP code can be applied to solve the problem. So we can solve the
approximation problem (10) easily.

5. Numerical Example

We have carried out the numerical experiments on the approximation scheme for (9) in Matlab by a simple
example. The mathematical model is
Min F ( x) 0.375 x1 x2  0.49 x3 x4  0.276 x5 x6  0.805 x7 x8  14.8 x1  14.4 x3  8.88 x5  8.88 x7
0.4 x1 x2  25 t 0,
 x3 x4  30 t 0 ,
15 x5 x6  120 t 0 ,
7.5 x7 x8  190 t 0 ,
s. t.
Pr ^2.5  0.4 x1  [1 t 0` t 0.8 5,
Pr ^ x3  8.8  [ 2 t 0` t 0.8 5
20 t x1 ,x2 ,x3 ,x4 ,x5 ,x6 ,x7 ,x8 t 0 ,
[1 ,[ 2 are independent and identically distributed random var iables with s tan d normal distribution.
To deal with the chance constraint, we apply CVaR to approximate it. In the tests, we apply SAA method to the
above problem and employ the random number generator rand in Matlab to generate the samples and solver fmincon
to solve the SAA problem of this model. To avoid the nonsmoothness of the constraint function in our Matlab
program, a smoothing technique is incorporated when we solve the SAA problem with fmincon.
The numerical results are displayed in Table 1. A few words about the notation. Appr.Sol denotes the
approximate optimal solution and Appl.Val denotes the optimal value of the objective function. For every fixed
sample size, we solve the SAA problem by setting the initial point to be (1, 2,1, 2,1, 2,1, 2) .

Table 1: Numerical results for the mathematical model

Sample Size Appr.Sol Appr.val

N
N=500 x (2.4868, 0.0000, 0.0000, 1.5730, 2.9218, 2.7380, 11.2300, 7.2380) 2.2080
N=1000 xN ( 2.4778, 0.0000, 0.0000, 1.5637, 2.9308, 2.7296, 11.1996, 7.2076 ) 2.2080
N
N=1500 x (2.5243, 0.0000, 0.0000, 1.6545, 3.0299, 2.6403, 11.2593, 7.3376) 2.2080
N=2000 xN (2.4934, 0.0000, 0.0000, 1.6129, 2.6751, 2.9905, 10.9820, 7.1392) 2.2080

6. Conclusion

The transportation problem in emergency management is a special transportation problems. It is very important
in practical problems which should be solved urgently. In this paper, we consider the mathematical model for the
transportation and distribution problems of sudden public incident emergency rescue materials. Based on the theory
of the chance constrained stochastic programming, we have established a hybrid chance constrained stochastic
programming model. We use the Conditional Value-at-Rrisk (CVaR) function to approximate the stochastic chance
 Deng Chunlin and Yang Liu / Systems Engineering Procedia 5 (2012) 137 – 143 143

constraints, and we obtain an simple approximation model by using the sample average approximation (SAA)
method. Specifically, the optimal solutions and stationary points obtained from solving sample average
approximated problems converge with probability one (w.p.1) to their true counterparts. For a fixed sample, the
SAA problem is a deterministic NLP and there are many existed NLP code can be applied to solve the problem. The
model and the corresponding approximate method is universal in engineering and many practical problems. It is not
only an improvement the classical transportation model, but also has great application value.

Acknowledgements

This work is supported by the construct program of the key discipline in Hunan Province, National Science
Foundation of China (11171281), the Research Foundation of Education Bureau of Hunan Province (10C1284),
Social Science Fund Project of Hunan Province (11YBA300).

References

1. R. Knott., The Logistics of Bulk Relief Supplies[J]ˊDisastersˊ11 (1988) 113-115ˊ

2. F. George. List Routing and Emergency Response Team Siting for high-level radioactive waste shipments[J]ˊIEEE Transactions On
Engineering Managementˈ45(2) (1998) 141-152ˊ
3. O. Linet. Emergency Logistics Planning in Natural Disasters[J]ˊAnnual of Operation Research. 129 (2004) 218-219ˊ
4. Q. Shi, Z.P. Huang. A multi-objective optimization model of a transportation problem[J]. Journal of HeFei University of Technology
( Natural Science Edition ). 29(6)(2006) 772-774.
5. A. Ben-Tal and A. Nemirovski, Robust solutions of linear programming problems contaminated with uncertain data. Mathematical
Programming. 88 (2000) 411-424.
6. R. T. Rockafellar and S. Uryasev, Optimization of conditional value-at-risk. The Journal of Risk. 2 (2000) 21-41.
7. A. Nemirovski and A. Shapiro, Convex approximations of chance constrained programs, SIAM Journal on Optimization. 17 ˄2006˅
969-996.
8. W. Wang, Sample average approximation of risk-averse stochastic programs. A thesis presented to the academic faculty, Georgia Institute
of Technology, 2007.
9. S.M. Robinson, Analysis of sample-path optimization. Mathematics of Operations Research. 21 (1996) 513–528.
10. A. Rusczy´nski and A.Shapiro(eds.), Stochastic Programming, Handbooks in OR & MS, North-Holland Publishing Company, Amsterdam,
2003.
11. L. B. Miller, H. Wagner. Chance-constrained programming with joint constraints. Operations Research, 13(1965) 930-945.
12. A. Shapiro, D. Dentcheva, A. Ruszczynsk. Lectures on stochastic programming : modeling and theory. Society for Industrial and Applied
Mathematics, Philadelphia,2009.
13. L. J. Hong, Y. Yang, and L. Zhang. Sequential convex approximations to joint chance constrained programs: A Monte Carlo approach.
Operations Research. 59(3) (2011) 3 617-630.