You are on page 1of 14

Available online at www.sciencedirect.

com

Computers & Industrial Engineering 56 (2009) 1081–1094


www.elsevier.com/locate/caie

Path selection model and algorithm for emergency


logistics management
Yuan Yuan a,*, Dingwei Wang b
a
Department of Management Science and Engineering, School of Business Administration, Northeastern University, Shenyang, China
b
Institute of Systems Engineering, School of Information Science and Engineering, Northeastern University, Shenyang, China

Available online 26 September 2008

Abstract

Path selection is one of the fundamental problems in emergency logistics management. Two mathematical models for
path selection in emergency logistics management are presented considering more actual factors in time of disaster. First a
single-objective path selection model is presented taking into account that the travel speed on each arc will be affected by
disaster extension. The objective of the model is to minimize total travel time along a path. The travel speed on each arc is
modeled as a continuous decrease function with respect to time. A modified Dijkstra algorithm is designed to solve the
model. Based on the first model, we further consider the chaos, panic and congestions in time of disaster. A multi-objective
path selection model is presented to minimize the total travel time along a path and to minimize the path complexity. The
complexity of the path is modeled as the total number of arcs included in the path. An ant colony optimization algorithm is
proposed to solve the model. Simulation results show the effectiveness and feasibility of the models and algorithms pre-
sented in this paper.
 2008 Elsevier Ltd. All rights reserved.

Keywords: Emergency logistics management; Path selection; Disaster extension; Path complexity; Modified Dijkstra algorithm; Ant
colony optimization algorithm

1. Introduction

In recent years, frequent natural disasters and man-made catastrophic events have brought great loss to
human beings (Ergonul, 2005; Chandre, Baskett, & Gallagher, 2007). As an emerging research area, emer-
gency management is attracting more and more attention of researchers (Kuwata & Takada, 2004; Mathieu,
2006; Nezih & Walter, 2006; Tufekci & Wallace, 1998). Logistics support is one of the major activities in disas-
ter response (Özdamar, Ekinci, & KÜÇÜKYAZICI, 2004). Commodities such as food, shelter and medicine
must be sent from the supply center to the affected area as quick as possible to support rescue operation and
help wounded people. Furthermore, important or hazardous materials must be transferred from the affected
areas to safety areas.

*
Corresponding author. Tel.: +86 24 83671468.
E-mail address: neuyy@126.com (Y. Yuan).

0360-8352/$ - see front matter  2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.cie.2008.09.033
1082 Y. Yuan, D. Wang / Computers & Industrial Engineering 56 (2009) 1081–1094

Path selection is one of the fundamental problems in logistics management. Among the existing researches
on emergency logistics management (Dai, Wang, Yang, & Lv, 1995; Chang, Tseng, & Chen, 2007; Sheu, 2007;
Yi & Özdamar, 2007), several complicated models have been built considering the disaster conditions. But
most of them consider the travel time on each arc of the logistics network as a constant. In fact, the travel
conditions on the arcs will be greatly affected by disaster extension especially under some disasters like hur-
ricane and flood that will extend gradually in time and space (Farahmand, 1997; Tufekci, 1995). The travel
speed on each arc will decrease continuously under disaster extension, and the decrease extent will differ with
the positions of the arcs and the severity of the disaster.
On the other hand, most of the existing research works on emergency logistics management took time
as the most important parameter to be considered. The objective of the model was to minimize the time
needed to complete the logistics transmission process. But disasters may cause great chaos and conges-
tions. People will be in great panic under dangerous conditions and the sense of panic will affect their
ability to follow a suitable path. Therefore the logistics transmission path should be as simple as possible
so that it can be followed more easily under disaster conditions. The complexity of the logistics transmis-
sion path should also be taken into account as an objective of the path selection model for emergency
logistics management.
In this paper, we focus on the path selection problem in emergency logistics management and build
mathematical models to select the best path. The motivation of this research is to consider more actual
factors in time of disaster when building models. The factors we take into account include the real-
time effect of disaster extension on the travel conditions of the arcs and the path complexity of the
path. Two mathematical models are built in our study and algorithms are developed to solve the
model. Our method for studying path selection problem in emergency logistics management is pre-
sented in Fig. 1.
As described in Fig. 1, first a mathematical model is built for path selection in emergency logistics manage-
ment considering the real-time effect of disaster extension. The objective of the model is to minimize total tra-
vel time along the path. The travel speed on each arc is modeled as a continuous decrease function with respect
to time. The proposed problem is different from classical shortest path problem since the travel speed on each
arc will decrease with time continuously. A modified Dijkstra algorithm is designed to solve the time-varied
shortest path problem presented in the model. Based on this single-objective model, we take path complexity

Consider real-time effect of


disaster extension

The travel speed on the arc: a


continuous decrease function
with respect to time

Single-objective time-varied
Modified Dijkstra algorithm
shortest path model

Consider the complexity of


the path

The complexity of the path:


total number of arcs included
in a selected path

Multi-objective time-varied Ant colony optimization


shortest path model algorithm

Fig. 1. Method for studying path selection problem in emergency logistics management.
Y. Yuan, D. Wang / Computers & Industrial Engineering 56 (2009) 1081–1094 1083

into consideration and a multi-objective path selection model for emergency logistics management is built. The
two objectives of the model are to minimize total travel time along the path and to minimize the complexity of
the path. The complexity of the path is modeled as the total number of arcs included in a selected path. This
multi-objective time-varied shortest path problem can’t be solved through classic algorithms and the modified
Dijkstra algorithm presented above. An ant colony optimization algorithm is proposed to solve the multi-
objective optimization model. Simulation results show the effectiveness and feasibility of the model and
algorithm.
This paper is organized as follows. A single-objective model to minimize the travel time along a path and a
multi-objective model to minimize the travel time and path complexity are presented in Section 2. In Section 3,
a modified Dijkstra algorithm is designed to solve the single-objective path selection model. An Ant Colony
Optimization algorithm is designed to solve the multi-objective path selection model in Section 4. In Section 5,
numerical experiments are carried out to test the models and algorithms presented in this paper. Finally, we
present our conclusions in Section 6.

2. Path selection model for emergency logistics management

2.1. Definition of variables and parameters

(1) An emergency logistics network is defined by a directed graph G(V, A), where V = {t1, t2,   , tn} is the
set of nodes and A # V  V is the set of arcs. Let t1, t2,   , tn denote the nodes in the network, where t1
is the source node and tn is the destination node.
(2) lij denotes the length of the arcs between nodes ti and tj, where (ti, tj) e A .
(3) s0ij is the travel speed on arc (ti, tj) under normal conditions. Define sij(t) as the travel speed on arc (ti, tj)
at time t under disaster conditions.

Observing the extension processes of some disasters such as flood and hurricane, we can find that the travel
speed on each arc of the network will decrease with the extension of disasters in time and space. The decrease
extent of the travel speed is affected by the position of the arc and the type of the disaster, etc. Hence, without
loss of generality, we assume the decrease function of the travel speed as follows:
sij ðtÞ ¼ s0ij  aij  ebij t
Here, aij, bij are the decrease parameters that determine the decrease extent of the travel speed function sij(t).
aij, bij can be estimated according to the distance from arc (ti, tj) to the disaster center, the vulnerability of the
arc and the type of the disaster etc.

(4) Let tij be the time needed to travel through arc (ti, tj). ti denotes the time when the logistics reach node ti,
tj denotes the time when the logistics reach node tj along arc (ti, tj). It is obvious that tij = tjti .
(5) xij is the decision variable in the model. xij = 1 when arc (ti, tj) is included in the fixed path and xij = 0
when arc (ti, tj) is not included in the fixed path.
(6) P denotes the transmission path of logistics which is a sequence of nodes in the network. Let pk be the
sequence number of node vpk in the network, then path P can be represented as ðvp1 ; vp2 ;    ; vpk ;    ; vpK Þ,
where 1 6 pk 6 n, k is the travel sequence of node vpk along path P. Path P starts at the source node and
terminates at the destination node, that is, p1 = 1 and pk = n. Path P should not have circles considering
the time pressures in emergency logistics management.
(7) Let ET ðP ; vpk Þ denote the travel time from node vp1 to node vpk along path P ¼ ðvp1 ; vp2 ;    ; vpk Þ, where
1 6 pk 6 n. We can easily obtain:

X
k1
ET ðP ; vpk Þ ¼ tpm pmþ1 ¼ ðtp2  tp1 Þ þ ðtp3  tp2 Þ þ    þ ðtpk  tpk1 Þ ¼ tpk
m¼1
1084 Y. Yuan, D. Wang / Computers & Industrial Engineering 56 (2009) 1081–1094

2.2. Recursive method for computing the travel time of a path

Based on the definitions above, for an already known path P ¼ ðvp1 ; vp2 ;    ; vpk Þ, we can obtain:
t p1 ¼ t 1 ¼ 0 ð1Þ
Z tp
k
spk1 pk ðtÞdt ¼ lpk1 pk ; 2 6 k 6 K ð2Þ
tpk1

In Eq. (2), if we have already known the lower limit of the integral tpk1 , the integrand spk1 pk ðtÞ and the integral
result lpk1 pk , the upper limit of the integral tpk can be obtained through solving Eq. (2).
The recurrence relations presented by Eqs. (1) and (2) show that the total travel time of path P can be
obtained through calculating the arriving time tpk at node vpk ð1 6 pk 6 nÞ recursively.

2.3. Path selection model for emergency logistics management under real-time effect of disaster extension

The formulation of path selection model under real-time effect of disaster extension is described as follows:
Model I:
n X
X n
min tij xij ð3Þ
i¼1 j¼1

s.t.
Z tj
sij ðtÞdt ¼ lij ð4Þ
ti

tij ¼ tj  ti ð5Þ
t1 ¼ 0 ð6Þ
bij t
sij ðtÞ ¼ s0ij  aij  e ð7Þ
8
<1
> i¼1
Xn Xn
xij  xji ¼ 1 i¼n ð8Þ
>
:
j¼1 j¼1 0 otherwise
j–i j–i
Xn 
6 1 i–n
xij ð9Þ
¼0 i¼n
j¼1
j–i
xij ¼ 0; 1; i ¼ 1; 2;    ; n; j ¼ 1; 2;    ; n ð10Þ
The objective of model I is to minimize the total travel time along a path. Eqs. (4), (5) and (6) are the recursion
formula of total travel time along a path, which means that arc (ti, tj) is traveled through with the speed sij(t)
during time period tij. Eq. (7) is the decrease function of the travel speed on arc (ti, tj) under real-time effect
disaster extension. Constraint (8) restricts the value of xij to constitute a feasible path from the source node t1
to the destination node tn. Constraint (9) ensures that there are no circles in the path. Constraint (10) is the 0–1
integer constraint of the decision variable xij.

2.4. Multi-objective path selection model for emergency logistics management

Related researches have shown that most congestion and panic happen at the intersections of two arcs in
the emergency network. For example, most traffic delays in regional evacuations occur at intersections (Tho-
mas & Justin, 2003). When traveling along a path, the fewer arcs included in the path, the more easily the path
can be followed. The complexity of a path can be simply represented as the total number of arcs included in a
Y. Yuan, D. Wang / Computers & Industrial Engineering 56 (2009) 1081–1094 1085

fixed path. Based on the single-objective path selection model presented in 2.3, a multi-objective path selection
model can be built taking into account both time factor and path complexity factor. The objectives of the
model are to minimize travel time along a path and to minimize the path complexity respectively. The model
can be formulated as follows:
Model II:
X
n X
n
min f1 ¼ tij xij ð3Þ
i¼1 j¼1

X
n X
n
min f2 ¼ xij ð11Þ
i¼1 j¼1

s.t.
(4)–(10)
The constraints in model II are the same with those of model I.

3. Modified Dijkstra algorithm to solve model I

Dijkstra algorithm is one of the classical algorithms to solve shortest path problem effectively (Ahuja Rav-
indra, Magnanti Thomas, & Orlin James, 1993). The basic idea of the algorithm is to find shortest path from
the source node t1 step by step. Dijkstra algorithm maintains a label T with each node ti, which is an upper
bound of the total weight on the shortest path from node t1 to node ti. The algorithm also sets a label P for
each node that denotes the total weight of the shortest path from source node t1 to each node ti. At any inter-
mediate step, the algorithm modifies the T labels of nodes and set P label for a certain node, then it adds the
node to the set of nodes with P labels. Thus the number of nodes with P labels will increase by one after each
step and the shortest paths from the source node t1 to all the other nodes in the network will be found after at
most (n1) steps.
As described in Section 2.2, in path selection problem for emergency logistics management under real-
time effect of disaster extension, the travel time tij on arc (ti, tj) is determined not only by the length of
the arc lij and the travel speed function sij(t), but it is also determined by the time when the logistics reach
the origin node ti of arc (ti, tj) since the travel speed on each arc is a decrease function with respect to time.
A modified Dijkstra algorithm is designed to solve model I as the classical Dijkstra algorithm can’t be
applied to solve the time-varied shortest path problem presented in this paper directly. The method to
achieve the travel time of a path in classical Dijkstra algorithm is modified in terms of the recursive method
described in Section 2.2, and the correctness of the algorithm is proved according to the characteristics of
the model and the algorithm.

3.1. The modified Dijkstra algorithm

Let P(tj) be the P label of node tj and T(tj) the T label of node tj. Si is the set of nodes with P labels after ith
step of the algorithm. Let k(tj) be the predecessor of node tj along a path P from the source node t1 to node tj.
M is a large enough positive number. Based upon the above discussion, the proposed modified Dijkstra algo-
rithm can be summarized as follows:

Step1: Initialization (iterative step i = 0). Let S0 = {t1}, P(t1) = 0, for "tk–t1, let T(tk) = +1, k(tk) = M
and m = 1 .
Step2: If tn e Si, we can obtain that P(tn) is the shortest travel time needed from the source node t1 to node
tn, then the corresponding path is the best path selected, the algorithm terminates. Otherwise Rt go to step3.
Step3: For each node tj where (tm, tj) e A and tj R Si, let tm = P(tm), solve the equation tmj smj ðtÞdt ¼ lmj to
achieve tj. If tj<T(tj), let T(tj) = tj and k(tj) = m, go to step4, else go to setp4 directly;
Step4: Let T ðvji Þ ¼ min fT ðvj Þg; P ðvji Þ ¼ T ðvji Þ; S iþ1 ¼ S i [ fvji g, m = ji, i i + 1, go to step2.
vj RS i
1086 Y. Yuan, D. Wang / Computers & Industrial Engineering 56 (2009) 1081–1094

3.2. Proof of the modified Dijkstra algorithm

To prove the correctness of the proposed algorithm, first we have Lemma 1 as follows.
Rt
Lemma 1. tj is a monotone increasing function of ti in the equation tij sij ðtÞdt ¼ lij with respect to tj if sij(t) is a
monotone decreasing and integrable function with respect to t when te[0,+1).
Let tHij be the travel time needed on arc (ti, tj) e A along path H. Define tHj;...;k as the time needed to travel
from node tj to node tk along path H. Let Pbestvk be the shortest travel time needed from node t1 to node tk.
Based upon above definitions and the conclusion of Lemma 1, the correctness of the algorithm can be proved
through mathematical induction.
Proof. To prove the correctness of the algorithm, we only need to prove that in each step of the algorithm, for
every tk e Si, the P label of node tk i.e. P(tk) is the shortest travel time needed from source node t1 to node tk .

(1) When i = 0, the conclusion is right obviously.


(2) Suppose the conclusion is right when i = n, i.e. for "tj e Sn, ET ðPbestvj ; vj Þ ¼ P ðvj Þ .

When i = n + 1, from step3 and step4 of the algorithm, we can obtain that T ðvjn Þ ¼ min fT ðvj Þg;
vj RS n
S nþ1 ¼ S n [ fvjn g. Suppose H is an arbitrary feasible path from node t1 to node vjn . Since t1eSn and
vjn R S n , along path H there must exist an arc whose origin node is in set Sn while destination node is not
in set Sn. Suppose (tr,tl) is the first one among those arcs along path H, i.e. tr e Sn ,vl R S n , then
ET ðH ; vjn Þ ¼ ET ðH ; vr Þ þ tHrl þ tHl;...;jn . From the inductive assumption, P(tr) is the shortest travel time needed
from node t1 to node tr, i.e. ET ðH ; vr Þ P ET ðPbestvr ; vr Þ. Then from Lemma 1 and Eqs. (1) and (2) which stand
for the recursive method for computing the travel time along a path, we can get:
Pbest vr
ET ðH ; vr Þ þ tHrl ¼ ET ðH ; vl Þ P ET ðPbestvr ; vr Þ þ trl ¼ ET ðPbestvr ; vl Þ:
According to the rule for modifying the T labels of nodes in the algorithm, since tr e Sn vl R S n , we can get
ET ðPbestvr ; vl Þ P T n ðvl Þ. Furthermore, from the rule for selecting node vjn to add in set S, we can get
T n ðvl Þ P T n ðvjn Þ. Then
ET ðH ; vjn Þ ¼ ET ðH ; vl Þ þ tHl;...;jn P T n ðvl Þ þ tHl;...;jn P T n ðvjn Þ ¼ P ðvjn Þ;
i.e. ET ðP bestvjn ; vjn Þ ¼ P ðvjn Þ. The conclusion is obtained. h

4. Ant colony optimization algorithm to solve model II

4.1. Convert model II to a single-objective model

First we use the ideal point method to deal with the model and convert the multi-objective path selection
model into a single-objective model as follows:
f1  f1 f2  f2
MinF ¼ r1 
þ r2 ð12Þ
f1 f2
s.t.
(4)–(10)
* P
T * *
Here f ¼ ðf1 ; f2 Þ is the ideal point. r 2 R ¼ f r jri P 0; ri ¼ 1; i ¼ 1; 2g is the weight vector.
According to the theorem about ideal point method, the optimum solution of the model (12), (4)–(10) is a
noninferior solution of model II.
After converting to a single-objective model, model II can be solved through solving three single-objective
models. The method for solving model II can be described as follows:
Y. Yuan, D. Wang / Computers & Industrial Engineering 56 (2009) 1081–1094 1087

Step1: Use modified Dijkstra algorithm presented in Section 3.1 to get the optimal solution of the single-
objective model (3)–(10) and obtain the corresponding optimal value f1*.
Step2: Use classical Dijkstra algorithm to get the optimal solution of the single-objective model (11), (4)–
(10) and obtain the corresponding optimal value f2*.
Step3: Put the ideal point into (12) and solve the single-objective model (12), (4)–(10) to get the noninferior
solution of the original multi-objective model (model II).

4.2. Algorithm to solve the multi-objective path selection model

Model II presented in this paper is a multi-objective model and the traveling speed on each arc is not a
constant but a continuous decrease function with respect to time, so model II is a multi-objective time-varied
shortest path problem. Therefore although the ideal point f1* can be obtained through the modified Dijkstra
algorithm presented in Section 3 and the ideal point f2* can be obtained through classical Dijkstra algorithms,
the single-objective model (12), (4)–(10) can’t be solved through either of the above two algorithms.
As a novel metaheuristic algorithm, ant colony optimization (ACO) algorithm was proposed firstly to solve
TSP problem (Dorigo, Maniezzo, & Colomi, 1996). In recent years, ACO has been applied to solve many dif-
ferent combinational problems successfully. ACO should also be suitable to solve the path selection problem
presented in this paper since it was proposed based on the simulation of the way real ant find the shortest route
between a food source and their nest. Furthermore, every ant can construct a feasible path in every cycle of the
ACO algorithm. This theorem can effectively avoid the problem of unfeasible solutions caused by crossover
and mutation process in genetic algorithms. So we choose ACO to solve model II.
After converting the multi-objective path selection model into a single-objective model by ideal point
method, an ACO algorithm is designed to achieve the noninferior solution of the multi-objective path selec-
tion model for emergency logistics management.
The ACO algorithm to solve the multi-objective path selection model is as follows:

Step1: Initialize (cycles counter NC = 0). Set the value of fundamental parameters of the ACO algorithm
including the maximum number of cycles NCmax, the number of ants M. For every edge (ti, tj) set an initial
pheromone value sij(0) = c and set the initial pheromone updating value Dsij = 0.
Step2: Every ant constructs a path from the source node to the destination node.

First place all the M ants on the source node. At node i, the mth ant chooses node j to move to with the
probability:
8 a b
< Psnij gij ifðvi ; vj Þ 2 A
pmij ¼ sa gb ð13Þ
j¼1 ij ij
:
0 otherwise
Here gij is a domain dependent heuristic parameter, a, b are parameters that weight the relative importance of
pheromone trail and heuristic information.
In the traveling speed decrease function (7), bigger s0ij  aij  ebij means lower decrease extent of the speed on
arc (ti, tj). While in objective function (11), there is no difference among the weights of the arcs, so the heuristic
value used here is as follows:
 0
sij  aij  ebij if ðvi ; vj Þ 2 A
gij ¼ ð14Þ
0 otherwise
The process of the movement of the mth ant from one node to the next one continues until the destination
node is included in the path constructed by the ant. And step2 finishes after all the M ants have constructed
a path from the source node to the destination node.
1088 Y. Yuan, D. Wang / Computers & Industrial Engineering 56 (2009) 1081–1094

Step3: Pheromone updating. After all the M ants have finished their movement from the source node to the
destination node in one cycle, the pheromone on each arc is updated with (15)–(17):

Q=Lm ; if the mth ant passed arcðvi ; vj Þin this cycle
Dsmij ¼ ð15Þ
0; otherwise
XM
Dsij ¼ Dsmij ð16Þ
m¼1

sij ðNC þ 1Þ ¼ ð1  qÞ  sij ðNCÞ þ Dsij ð17Þ


Here Lm is a value related to the value of the objective function. q is the evaporation rate of the pheromone.
Lm in this step is equal to the value of objective function (12) corresponding to the path found by ant m,
that is,
f1m  f1 f2m  f2
Lm ¼ r 1 þ r 2 ð18Þ
f1 f2
Here f1m and f2m are the values of objective function (3) and (11) corresponding to the path found by ant m
respectively. And f1*, f2* are the optimal value of objective function (3) and (11) obtained by Dijkstra
algorithm.

Step4: If NC < NCmax, NC + 1 ? NC, go to step2; else the algorithm is terminated.

5. Computational experiments

5.1. Results of model I on a logistics network with 20 nodes

We carry out our computational experiments on a logistics network with 20 nodes. The structure of the
network is shown in Fig. 2. Suppose the disaster happens at node 1. The length of each arc lij, the initial travel
speed s0ij and the decrease parameters aij, bij are shown in Table 1.
To solve model I, the modified Dijkstra algorithm was implemented in C and run on a PC with 2.66 GHz
CPU and 512 MB RAM. According to the proposed path selection model for emergency logistics manage-
ment under real-time effect of disaster extension and the modified Dijkstra algorithm, we can obtain that

2
3
5
4
1
7
6 10
8 9

13
12 14
11
15
17
16 19
18
20

Fig. 2. Structure of the emergency logistics network.


Y. Yuan, D. Wang / Computers & Industrial Engineering 56 (2009) 1081–1094 1089

Table 1
Parameters of the emergency logistics network
(vi,vj) ðlij ; s0ij ; aij ; bij Þ (m, m/min,-,-) (vi,vj) ðlij ; s0ij ; aij ; bij Þ (m, m/min,,) (vi,vj) ðlij ; s0ij ; aij ; bij Þ (m, m/min,,)
(1,2) (50,100,0.64,0.19) (6,16) (100,115,0.66,0.19) (14,10) (80,115,0.89,0.08)
(1,6) (30,60,0.63,0.17) (7,8) (40,90,0.69,0.18) (14,15) (60,105,0.84,0.06)
(1,11) (70,110,0.64,0.19) (8,4) (40,85,0.72,0.13) (15,20) (30,90,0.85,0.06)
(2,3) (30,60,0.68,0.19) (8,13) (30,75,0.72,0.11) (16,17) (80,115,0.65,0.16)
(2,7) (40,70,0.62,0.16) (9,5) (70,110,0.83,0.09) (16,18) (110,120,0.63,0.15)
(3,4) (80,100,0.72,0.11) (9,14) (40,90,0.88,0.07) (17,13) (40,80,0.71,0.12)
(3,8) (60,95,0.75,0.11) (10,15) (50,105,0.89,0.07) (17,18) (40,75,0.75,0.14)
(4,5) (110,120,0.76,0.1) (11,16) (30,70,0.69,0.17) (18,13) (60,115,0.79,0.12)
(4,9) (40,75,0.74,0.12) (12,8) (30,65,0.79,0.12) (18,19) (70,110,0.7,0.11)
(5,10) (60,110,0.88,0.07) (12,17) (40,100,0.75,0.13) (18,20) (120,120,0.73,0.14)
(6,7) (30,65,0.65,0.15) (13,9) (30,80,0.8,0.15) (19,15) (40,75,0.83,0.08)
(6,12) (60,105,0.62,0.16) (13,19) (110,120,0.78,0.14) (19,20) (70,105,0.89,0.08)

the best path from node 1 to node 20 is: 1 ? 6 ? 7 ? 8 ? 13 ? 9 ? 14 ? 15 ? 20 and the total travel time
along this path is 6.490727 min. While if we don’t consider the effect of disaster extension, according to clas-
sical Dijkstra algorithm, the static shortest path from node 1 to node 20 is 1 ? 11 ? 16 ? 18 ? 20. From the
parameters shown in Table 1, we can obtain that the travel time along this path is 7.210173. The travel time
along the static shortest length path is much longer than the best path obtained through model I. The travel
time along the best path obtained by model I is 9.98% shorter than the travel time along the static shortest
path. These simulation results show that it is necessary to consider the effect of disaster extension in path selec-
tion for emergency logistics management.
Table 2 shows the paths with shortest travel time from node 1 to all the other nodes. From the results
shown in Table 2, we can see that Model I will also work when the number of destination node is not limited
to one. For example, node 15 is the best destination node when nodes 10, 15, and 20 are all the alternative
destination nodes.

5.2. Results of model I under different extent of disasters

Different extent of disasters will affect the travel condition on the arcs of the logistics network differently.
Here we use different set of the decrease parameters aij and bij to reflect the extent of disasters. Decrease
parameter aij can reflect the instantaneous influence of disasters on the travel conditions of the arcs, littler
aij means greater influence of disasters. Decrease parameter bij can reflect the influence of the disasters in a
period of time after the disaster happened, bigger bij means quicker decrease of the travel speed, that is,
the greater extent of disasters. Furthermore, different set of aij and bij can also reflect the influence of disasters
on arcs in different positions.

Table 2
Best paths from the source node to all the other nodes and the corresponding travel time
Path Travel time (min) Path Travel time (min)
1?2 0.845697 1 ? 6 ? 12 2.009719
1?2?3 1.788836 1 ? 6 ? 7 ? 8 ? 13 3.444037
1?2?3?4 3.253471 1 ? 6 ? 7 ? 8 ? 13 ? 9 ? 14 4.978389
1?2?3?4?5 5.080585 1 ? 6 ? 7 ? 8 ? 13 ? 9 ? 14 ? 15 5.921667
1?6 0.852553 1 ? 11 ? 16 1.903324
1?6?7 1.712650 1 ? 6 ? 12 ? 17 2.735475
1?6?7?8 2.666789 1 ? 6 ? 12 ? 17 ? 18 3.862882
1 ? 6 ? 7 ? 8 ? 13 ? 9 4.280119 1 ? 6 ? 12 ? 17 ? 18 ? 19 5.371893
1 ? 2 ? 3 ? 4 ? 5 ? 10 5.993719 1 ? 6 ? 7 ? 8 ? 13 ? 9 ? 14 ? 15 ? 20 6.490727
1 ? 11 1.102048
1090 Y. Yuan, D. Wang / Computers & Industrial Engineering 56 (2009) 1081–1094

Here we divide the logistics network with 20 nodes into three areas, the division of the network is shown in
Fig. 3 (arcs traveled through the boundary of two areas is treated as arcs in the former area).
In each area the decrease parameters are generated randomly in different intervals. So we can get five grade
of disasters severity as shown in Table 3. In the data of Table 3, decrease parameters aij is set litter in higher
disaster grade than in lower disaster grade in corresponding areas and decrease parameters bij is set larger in
higher disaster grade than in lower disaster grade in corresponding areas. Disaster grade 0 strands for the sit-
uation when there is no disaster happened and then Model I becomes an ordinary shortest path problem.
In Tables 4–8, the value of decrease parameters and the best path obtained by the modified Dijkstra algo-
rithm are presented. The contrast of the best path and the static shortest path are also presented.
From Tables 4–8 we can see that the result of path selection differs with the disaster grade, that is, with the
set of decrease parameters. Fig. 4 shows the total travel time along the best path obtained by model I under all
the disaster grades. From Fig. 4 we can see that the total travel time along the best path increases with the
disaster grade.
Fig. 5 shows the contrast of the travel time along the best path obtained through model I and the static
shortest path. From Fig. 5 we can see that the travel time along the best path is shorter than the static shortest
path in bigger percentage under higher disaster grade. So it is more necessary to consider the real-time effect of
disaster extension when the disaster is serious.

Area Area Area

2
3
5
4
1
7 10
6
8 9
13
12 14
11
15
17
16 19
18
20

Fig. 3. Division of the emergency logistics network.

Table 3
Interval of travel speed decrease parameters under different grade of disasters
Area I Area II Area III
Disaster grade 0 a = 1, b = 0 a = 1, b = 0 a = 1, b = 0
Disaster grade 1 a e (0.9,1.0)b e (0.00,0.05) a = 1, b = 0 a = 1, b = 0
Disaster grade 2 a e (0.8, 0.9)b e (0.05, 0.10) a e (0.9, 1.0)b e (0.00, 0.05) a = 1, b = 0
Disaster grade 3 a e (0.7, 0.8)b e (0.10, 0.15) a e (0.8, 0.9)b e (0.05, 0.10) a e (0.9, 1.0)b e (0.00, 0.05)
Disaster grade 4 a e (0.6, 0.7)b e (0.15, 0.20) a e (0.7, 0.8)b e (0.10, 0.15) a e (0.8, 0.9)b e (0.05, 0.10)
Disaster grade 5 a e (0.5, 0.6)b e (0.20, 0.25) a e (0.6, 0.7)b e (0.15, 0.20) a e (0.7, 0.8)b e (0.10, 0.15)
Y. Yuan, D. Wang / Computers & Industrial Engineering 56 (2009) 1081–1094 1091

Table 4
Parameters and path selection result under disaster grade 1
Area (ti, tj) (aij, bij) (ti, tj) (aij, bij) (ti, tj) (aij, bij)
I a e (0.9, 1.0)b e (0.00, 0.05) (1,2) (0.9222,0.0381) (2,7) (0.9857,0.0305) (7,8) (0.9153,0.0436)
(1,6) (0.9193,0.0143) (6,7) (0.9577,0.0123) (11,16) (0.9967,0.0089)
(1,11) (0.9022,0.0499) (6,12) (0.9060,0.0399) (16,17) (0.9640,0.0491)
(2,3) (0.9904,0.0277) (6,16) (0.9049,0.0353) (16,18) (0.9294,0.0327)
II a = 1, b = 0 (3,4) (1.0000,0.0000) (8,13) (1.0000,0.0000) (17,13) (1.0000,0.0000)
(3,8) (1.0000,0.0000) (12,8) (1.0000,0.0000) (17,18) (1.0000,0.0000)
(4,5) (1.0000,0.0000) (12,17) (1.0000,0.0000) (18,13) (1.0000,0.0000)
(4,9) (1.0000,0.0000) (13,9) (1.0000,0.0000) (18,19) (1.0000,0.0000)
(8,4) (1.0000,0.0000) (13,19) (1.0000,0.0000) (18,20) (1.0000,0.0000)
IIIa = 1, b = 0 (5,10) (1.0000,0.0000) (10,15) (1.0000,0.0000) (15,20) (1.0000,0.0000)
(9,5) (1.0000,0.0000) (14,10) (1.0000,0.0000) (19,15) (1.0000,0.0000)
(9,14) (1.0000,0.0000) (14,15) (1.0000,0.0000) (19,20) (1.0000,0.0000)
Best path 1 ? 6 ? 12 ? 17 ? 18 ? 20 Travel time 3.136548 min
Static shortest path 1 ? 11 ? 16 ? 18 ? 20 Travel time 3.196073 min
Contrast (3.1960733.136548)/3.196073  100% = 1.86%

Table 5
Parameters and path selection result under disaster grade 2
Area (ti, tj) (aij, bij) (ti, tj) (aij, bij) (ti, tj) (aij, bij)
I a e (0.8, 0.9)b e (0.05, 0.10) (1,2) (0.8288,0.0510) (2,7) (0.8238,0.0698) (7,8) (0.8731,0.0746)
(1,6) (0.8339,0.0776) (6,7) (0.8907,0.0734) (11,16) (0.8760,0.0959)
(1,11) (0.8926,0.0669) (6,12) (0.8940,0.0889) (16,17) (0.8652,0.0660)
(2,3) (0.8047,0.0910) (6,16) (0.8910,0.0980) (16,18) (0.8183,0.0680)
II a e (0.9, 1.0)b e (0.00, 0.05) (3,4) (0.9153,0.0216) (8,13) (0.9999,0.0222) (17,13) (0.9077,0.0350)
(3,8) (0.9562,0.0114) (12,8) (0.9345,0.0044) (17,18) (0.9470,0.0344)
(4,5) (0.9890,0.0281) (12,17) (0.9846,0.0228) (18,13) (0.9669,0.0202)
(4,9) (0.9984,0.0386) (13,9) (0.9094,0.0264) (18,19) (0.9530,0.0059)
(8,4) (0.9746,0.0376) (13,19) (0.9326,0.0151) (18,20) (0.9278,0.0021)
III a = 1, b = 0 (5,10) (1.0000,0.0000) (10,15) (1.0000,0.0000) (15,20) (1.0000,0.0000)
(9,5) (1.0000,0.0000) (14,10) (1.0000,0.0000) (19,15) (1.0000,0.0000)
(9,14) (1.0000,0.0000) (14,15) (1.0000,0.0000) (19,20) (1.0000,0.0000)
Best path 1 ? 6 ? 12 ? 17 ? 18 ? 20 Travel time 3.480311 min
Static shortest path 1 ? 11 ? 16 ? 18 ? 20 Travel time 3.694391 min
Contrast (3.6943913.480311)/3.694391  100% = 5.79%

5.3. Results of model II on network with 20 nodes

To solve model II, the ACO algorithm was implemented in C and run on a PC with 2.66 GHz CPU,
512 MB RAM. Parameters of ACO are set as: the number of ants M = 15, the maximum number of cycles
NCmax = 100, the initial pheromone value c = 1000, a = 2, b = 1, Q = 1, q = 0.2. When r1 = 0.9, r2 = 0.1,
the ideal point and the weak efficient solution to the multi-objective model are shown in Table 9. From the
results shown in Table 9 we can see that the noninferior solution is a trade-off between the solutions corre-
sponding to the ideal point.

6. Conclusions

In this study, two path selection models are built for emergency logistics management. Model I mainly
considered the real-time effect of disaster extension and the travel speed on each arc is modeled as a con-
1092 Y. Yuan, D. Wang / Computers & Industrial Engineering 56 (2009) 1081–1094

Table 6
Parameters and path selection result under disaster grade 3
Area (ti, tj) (aij, bij) (ti, tj) (aij, bij) (ti, tj) (aij, bij)
I a e (0.7, 0.8)b e (0.10, 0.15) (1,2) (0.7544,0.1026) (2,7) (0.7527,0.1277) (7,8) (0.7242,0.1080)
(1,6) (0.7666,0.1394) (6,7) (0.7001,0.1062) (11,16) (0.7378,0.1281)
(1,11) (0.7561,0.1276) (6,12) (0.7145,0.1209) (16,17) (0.7048,0.1197)
(2,3) (0.7071,0.1089) (6,16) (0.7234,0.1361) (16,18) (0.7639,0.1365)
II a e (0.8, 0.9)b e (0.05, 0.10) (3,4) (0.8912,0.0844) (8,13) (0.8425,0.0987) (17,13) (0.8191,0.0778)
(3,8) (0.8242,0.0571) (12,8) (0.8383,0.0543) (17,18) (0.8555,0.0781)
(4,5) (0.8773,0.0552) (12,17) (0.8785,0.0786) (18,13) (0.8965,0.0653)
(4,9) (0.8678,0.0990) (13,9) (0.8551,0.0707) (18,19) (0.8967,0.0694)
(8,4) (0.8801,0.0670) (13,19) (0.8730,0.0570) (18,20) (0.8551,0.0948)
III a e (0.9, 1.0)b e (0.00, 0.05) (5,10) (0.9476,0.0343) (10,15) (0.9937,0.0091) (15,20) (0.9517,0.0363)
(9,5) (0.9806,0.0484) (14,10) (0.9334,0.0486) (19,15) (0.9547,0.0265)
(9,14) (0.9984,0.0219) (14,15) (0.9966,0.0220) (19,20) (0.9849,0.0290)
Best path 1 ? 6 ? 12 ? 17 ? 18 ? 19 ? 20 Travel time 4.544814 min
Static shortest path 1 ? 11 ? 16 ? 18 ? 20 Travel time 4.954273 min
Contrast (4.9542734.544814)/4.954273  100% = 8.26%

Table 7
Parameters and path selection result under disaster grade 4
Area (ti,tj) (aij, bij) (ti,tj) (aij, bij) (ti,tj) (aij, bij)
I a e (0.6, 0.7)b e (0.15, 0.20) (1,2) (0.6396,0.1783) (2,7) (0.6580,0.1674) (7,8) (0.6619,0.1864)
(1,6) (0.6815,0.1934) (6,7) (0.6481,0.1515) (11,16) (0.6821,0.1645)
(1,11) (0.6204,0.1647) (6,12) (0.6717,0.1957) (16,17) (0.6146,0.1858)
(2,3) (0.6244,0.1698) (6,16) (0.6706,0.1672) (16,18) (0.6625,0.1918)
II a e (0.7, 0.8)b e (0.10, 0.15) (3,4) (0.7606,0.1184) (8,13) (0.7073,0.1320) (17,13) (0.7769,0.1329)
(3,8) (0.7454,0.1348) (12,8) (0.7911,0.1081) (17,18) (0.7269,0.1441)
(4,5) (0.7276,0.1178) (12,17) (0.7222,0.1170) (18,13) (0.7110,0.1483)
(4,9) (0.7686,0.1192) (13,9) (0.7869,0.1274) (18,19) (0.7779,0.1018)
(8,4) (0.7416,0.1484) (13,19) (0.7775,0.1477) (18,20) (0.7046,0.1270)
III a e (0.8, 0.9)b e (0.05, 0.10) (5,10) (0.8334,0.0652) (10,15) (0.8225,0.0678) (15,20) (0.8935,0.0502)
(9,5) (0.8208,0.0519) (14,10) (0.8736,0.0859) (19,15) (0.8001,0.0811)
(9,14) (0.8827,0.0574) (14,15) (0.8742,0.0662) (19,20) (0.8535,0.0919)
Best path 1 ? 6 ? 12 ? 8 ? 13 ? 9 ? 14 ? 15 ? 20 Travel time 6.365086 min
Static shortest path 1 ? 11 ? 16 ? 18 ? 20 Travel time 7.468316 min
Contrast (7.4683166.365086)/7.468316  100% = 14.77%

tinuously decrease function with respect to time in order to simulate the real-time effect of disaster exten-
sion. The proposed problem is different from classical shortest path problem in the representation of travel
speed, so a modified Dijkstra algorithm is designed to solve the model. Proof and simulation results show
the correctness and effectiveness of the algorithm. Based on model I, we take path complexity into con-
sideration and a multi-objective path selection model is built. An ant colony optimization algorithm is
proposed to solve the multi-objective optimization model. Simulation results show the effectiveness and
feasibility of the model and algorithm. There still remain quite a lot of complex factors in emergency sit-
uations to be considered. Here we just take path selection as a fundamental problem in emergency logis-
tics management. Building models taking more actual factors into account will be one of our future work
directions.
Y. Yuan, D. Wang / Computers & Industrial Engineering 56 (2009) 1081–1094 1093

Table 8
Parameters and path selection result under disaster grade 5
Area (ti, tj) (aij, bij) (ti, tj) (aij, bij) (ti, tj) (aij, bij)
I a e (0.5, 0.6)b e (0.20, 0.25) (1,2) (0.5335,0.2015) (2,7) (0.5049,0.2283) (7,8) (0.5212,0.2217)
(1,6) (0.5102,0.2404) (6,7) (0.5423,0.2267) (11,16) (0.5772,0.2161)
(1,11) (0.5288,0.2376) (6,12) (0.5016,0.2193) (16,17) (0.5936,0.2269)
(2,3) (0.5413,0.2201) (6,16) (0.5914,0.2336) (16,18) (0.5795,0.2271)
II a e (0.6, 0.7)b e (0.15, 0.20) (3,4) (0.6196,0.1560) (8,13) (0.6613,0.1619) (17,13) (0.6382,0.1995)
(3,8) (0.6887,0.1668) (12,8) (0.6174,0.1741) (17,18) (0.6247,0.1973)
(4,5) (0.6400,0.1504) (12,17) (0.6164,0.1711) (18,13) (0.6339,0.1579)
(4,9) (0.6957,0.1961) (13,9) (0.6271,0.1899) (18,19) (0.6754,0.1805)
(8,4) (0.6951,0.1611) (13,19) (0.6139,0.1851) (18,20) (0.6687,0.1729)
III a e (0.7, 0.8)b e (0.10, 0.15) (5,10) (0.7393,0.1177) (10,15) (0.7584,0.1300) (15,20) (0.7585,0.1131)
(9,5) (0.7387,0.1479) (14,10) (0.7348,0.1106) (19,15) (0.7808,0.1341)
(9,14) (0.7823,0.1425) (14,15) (0.7513,0.1019) (19,20) (0.7903,0.1040)
Best paht 1 ? 2 ? 3 ? 4 ? 9 ? 14 ? 15 ? 20 Travel time 12.323639 min
Static shortest path 1 ? 11 ? 16 ? 18 ? 20 Travel time 19.297973 min
Contrast (19.29797312.323639)/19.297973  100% = 36.14%

D 5
i
s 4
a
s 3
t
e
r 2

g 1
r
a 0
d
e 0.000 2.000 4.000 6.000 8.000 10.000 12.000 14.000
Travel time (min)
Fig. 4. Travel time under different disaster grade.

40
35
30
25
(%)

20
15
10
5
0
1 2 3 4 5
Disaster grade
Fig. 5. Contrast of the best path and the static shortest path under different disaster grade.
1094 Y. Yuan, D. Wang / Computers & Industrial Engineering 56 (2009) 1081–1094

Table 9
Ideal points and noninferior solution obtained by ACO when r1 = 0.9, r2 = 0.1
Path Travel time Path Fitness value of the noninferior
(min) complexity solution
Ideal point f1 1 ? 6 ? 7 ? 8 ? 13 ? 9 ? 14 ? 15 ? 20 6.490727 8 –
Ideal point f2 1 ? 11 ? 16 ? 18 ? 20 7.210173 4 –
Noninferior 1 ? 6 ? 12 ? 17 ? 18 ? 20 6.716635 5 0.056324
solution

Acknowledgments

This work is supported by the Key Program of National Natural Science Foundation of China (Grant No.
70431003), the Science Fund for Creative Research Group of National Natural Science Foundation of China
(Grant No. 60521003) and National Key Technology R&D Program (Grant No. 2006BAH02A09).

References

Chandre, Monerawela, Baskett, Peter, & Gallagher, Gerry (2007). The southeast Asian tsunami disaster—how resuscitation helped the
recovery program. Resuscitation, 72(1), 6–7.
Dai, Jianshe, Wang, Shuning, Yang, Xiaoyin, & Lv, Qiang (1995). A model for time-critical transport problem and its computerized
implementation. Control and Decision, 10(2), 143–147.
Ergonul, S. (2005). A probabilistic approach for earthquake loss estimation. Structural Safety, 27, 309–321.
Kambiz Farahmand. (1997). Application of simulation modeling to emergency population evacuation. In Proceedings of the 1997 winter
simulation conference, Atlanta (pp. 1181–1188). Piscataway: IEEE.
Ahuja Ravindra, K., Magnanti Thomas, L., & Orlin James, B. (1993). Network flows: Theory, algorithms and application. New York:
Pearson Education.
Mathieu, Gorge (2006). Crisis management best practice – where do we start from? Computer Fraud & Security, 2006(6), 10–13.
Chang, Mei-Shiang, Tseng, Ya-Ling, & Chen, Jing-Wen (2007). A scenario planning approach for the flood emergency logistics
preparation problem under uncertainty. Transport Research Part E, 43(6), 737–754.
Dorigo, Macro, Maniezzo, Vittorio, & Colomi, Alberto (1996). Ant system: Optimization by a colony of cooperating agents. IEEE
Transactions on Systems, Man, and Cybernetics, 26(1), 29–41.
Nezih, Altay, & Walter, G Green III, (2006). OR/MS research in disaster operations management. European Journal of Operational
Research, 175(1), 475–493.
Özdamar, L., Ekinci, E., & KÜÇÜKYAZICI, B. (2004). Emergency logistics planning in natural disasters. Annals of Operations Research,
129, 217–245.
Tufekci, Suleyman, & Wallace, William A (1998). The emerging area of emergency management and engineering. IEEE Transactions on
Engineering Management, 45(2), 103–105.
Sheu, J.-B. (2007). An emergency logistics distribution approach for quick response to urgent relief demand in disasters. Transport
Research Part E, 43(6), 687–709.
Tufekci, Suleyman (1995). An integrated emergency management decision support system for hurricane emergencies. Safety Science, 20(1),
39–48.
Cova, Thomas J., & Johnson, Justin P. (2003). A network flow model for lane-based evacuation routing. Transportation Research Part A,
37, 579–604.
Yi, Wei, & Özdamar, Linet (2007). A dynamic logistics coordination model for evacuation and support in disaster response activities.
European Journal of Operational Research, 179(3), 1177–1193.
Kuwata, Yasuko, & Takada, Shiro (2004). Effective emergency transportation for saving human lives. Natural Hazards, 33, 23–46.

You might also like