You are on page 1of 15

ZZY

sets and systems


ELSEVIER Fuzzy Sets and Systems 81 (1996) 169-183

Fuzzy sets modeling and optimization for disaster


control systems planning
Augustine 0. Esogbue*
School of lndustrial and Systems Engineering, Georgia Institute of Technology, Atlanta, Georgia 30332-0205, USA
Received June 1995

Abstract

Various forms of disasters, natural and man-made, abound. Although, to some extent, some of these have been
exacerbated by modern civilization, various technologies have been invoked to control or mitigate their adverse effects.
Modeling- and optimization-based techniques have also been employed. We outline some efforts based on fuzzy sets
theory which are utilized in conjunction with classical methods to enhance the reliability and utility of these efforts. In
particular, disasters caused by floods and storm water pollution are used as vehicles for the exemplification of our models.

Keywords: Disaster control planning; Flood control; National and regional levels; Fuzzy modeling and optimization;
Fuzzy dynamic programming; Data generation

1. Introduction appears that we are unable to assess accurately and


reliably the nature and extent of these damages,
In recent times, we have learned of the occur- both rapidly and with minimal resource consump-
rence of various forms of disasters in many parts of tion. An immediate consequence of this dilemma is
the world. Some of these are natural, while others poor and suboptimal responses in an attempt to
result from man-made systems or structures. A few mitigate the deleterious effects of these disasters.
examples of these are nuclear power plant failures, The problems sketched in the foregoing are con-
earthquakes, floods, oil spillage in the oceans, forest tinually being wrestled with by planners and vari-
fires, storm water pollution destroying surface ous agencies charged with the responsibility for
waters and ecosystems, etc. The magnitude of the their amelioration. Systems science is considered
damages due to each of these occurrences can be central to these planning and response efforts. It
immense. Confounding our attempts to cope with has become recognized that a solo employment of
these problems is mankind's apparent inability to engineering structures and devices can prove to be
predict and control accurately the incipience of quite an ineffective control strategy in mitigating
these disasters. Additionally, once they occur, it the deleterious effects of various disasters. Indeed,
this practice has, at times, led to a false sense of
protection and resulted in poor or lack of prepared-
*E mail: augustine.esogbue@isye.gatech.edu. ness for dealing with the disasters. Hence, the use of

0165-0114/96/$15.00 1996 ElsevierScienceB.V. All rights reserved


SSDI 0 1 6 5 - 0 1 1 4 ( 9 5 ) 0 0 2 4 8 - 0
170 A.O. Esogbue / Fuzzy Sets and Systems 81 (1996) 169-183

Present
National Balance ~

~resent Tol
Catastrophe
Potential

I
None X x
-- Net Benefits

Fig. 1. Trends and limits of adjustments to floods (Source: [5]).

some mix or combination of structural and non- of these problems and control policies are highly
structural measures in attacking the problem is interactive.
advocated. To illustrate, we exhibit in a matrix format (see
As an illustration of the foregoing, consider the Fig. 2) the interaction of these various adjustments.
flood control problem. In this case, the engineering The next task is to fill in the cells as indicated.
(protective) measures such as levees, flood walls, Doing so via classical approaches results in very
channel improvements and storage reservoirs are vaguely defined protocols. We feel that proper
considered structural while the nonstructural evaluation of the cells in the matrix using such
measures would typically consist of flood plain descriptors as 'high stimulation', 'little or none',
zoning, flood proofing, purchase of portions of the and 'doubtful', etc., can be systematically and sub-
flood plain, land use conversion, warning and stantially aided by fuzzy set theory. This is also the
evacuation, relief and rehabilitation, and flood in- case with the evaluation of damages or costs of
surance. As Fig. 1 indicates, while a number of these disasters, especially when these need to be obtained
may be beneficial at certain tolerable disaster levels, rather expeditiously as is usually the case in the
they are totally useless or inadequate for coping hectic and painful moments immediately following
with catastrophes of immense magnitude such as the incipience of such disasters.
those that struck the mid-west region of the US in Fuzzy mathematical programming has matured
1993 and southern Georgia in July of 1994. considerably in recent times. Applications to nu-
A further analysis of the net benefits of various merous areas have also been reported and several
measures as a function of the magnitude of the reviews, surveys, and tutorials abound. Notable
catastrophe suggests that some adroit combination among these are Verdagay [13], Lai and Hwang
of these measures would be beneficial. Reliance on [10], and Zimmermann [15]. In the sequel, we
any one of them may provide a false sense of secur- present a general mathematical statement of the
ity leading to rather unwise practices. Determ- optimal adjustment to the disaster control problem
ination of an optimal combination or policy is, and outline aspects of a fuzzy mathematical pro-
however, a very difficult task indeed. This is gramming approach which we have developed
due to many reasons. For one thing, the effects and described in various places for addressing the
A.O. Esogbue t Fuzzy Sets and Systems 81 (1996) 169-183 171

/
InitialAdjustment f_~ _ . / _~" / . / f f ~

l / //Jl
Controland 0 0 0 0 0
Protection

Flood-
Proofing 0 0 O 9 0

LandUse
Plannlng 0 9 0

Warnings 0 O O 9 0

,.snrance 0 ? ? 0
Reliefand
Rehabilitation 0 0 0 0

Key. Stimulatedbytheinitialadjustment:
- Highstimulation
D - Littleor none
9 . Doubtful
Fig. 2. Matrix of interaction of adjustments to floods (Source: [5]).

problem. In addition, we consider a data evaluation DCP: max f ( x , y )


stratagem, particularly suited to the membership
functions that appear in the various models. These s.t. (x,y) ~ @(x,y), (1)
and other aspects not treated previously are essen- xeX, y ~ Y,
tial for the model implementation.
in which x = (xl, x2, - . - , X N ) is the vector of struc-
tural measures, where xj = 1 if structural measure
2. Mathematical models of disaster control j is selected and 0 if not; y = (Yx,Y2 . . . . . Yk) is the
vector of nonstructural measures, where Yk is the
A very general model of the disaster control level of the kth nonstructural measure selected,
problem involving the use of a vector combination f(x,y) is the objective function; e.g., the discounted
of structural and nonstructural measures and net reduction in flood damages resulting from plan
which forms the leitmotif of our study is as follows: (x,y); I,(x,y) is the set of feasible plans (x,y), i.e.,
Find a combination (x,y) of structural (x) and non- those satisfying the planning, financial, engineering,
structural (y) measures so as to and social constraints, and X and Y, respectively,
172 A.O. Esogbue / Fuzzy Sets and Systems 81 (1996) 169 183

are the sets of feasible structural and nonstructural ter control problem with immense potential for un-
measures. told losses, both in terms of human and financial
In the literature, it is customarily assumed that capital. Extension to other forms of disasters may be
nonstructural measures essentially compliment made with minimal adjustment to the models. We
a given set of structural measures as far as damage next outline some of the steps followed to acquire
reduction is concerned. In keeping with this practice, and process the various data requirements of the
and assuming further that they may vary over time, models.
the general formulation given in (1) then reduces to The primary motivation for our approach is the
the determination of y* so as to minimize the ex- strong belief that, in the environmental systems
pected damages in year t. This can be expressed as analysis field, a substantive departure from the con-
ventional crisp quantitative way of modeling is
P(l, t) = min {O(I,y, t) + C(y, t)}, (2) needed both because of modeling and implementa-
ye Y~Gil)
tion exigencies. Such an approach would provide
where D(l,y, t) is the annual damage in the tth year the researcher with a more close-to-reality repre-
for a given combination I of the structural measures sentation of complex or ill-defined phenomena as
with the level of nonstructural measures aty, C(y, t) employed by planners. This should lead to more
is the annual cost incurred in the tth year with the effective common sense control policies for a wide
levels of the nonstructural measures aty, and ~/, and variety of practical problems.
Y are similar to those defined in the (DCP) prob- As stated in our previous exposition, the D C P is
lem. P(I,t) denotes the minimal sum of the net a complex web of problems involving engineering,
annual damages and the non-structural measures economic, environmental, social and management
costs in the tth year for the combination l of the aspects. It therefore deals with entities and relations
structural measures. which are often not precisely known and indeed are
difficult to quantify. A fuzzy approach appears to
be more natural and appropriate than classical
3. Fuzzy mathematical models of the disaster methods. In particular, the difficulty of dis-associat-
control problem ing crisply the impacts (benefits) of interacting con-
trol strategies, usually the case with nonstructural
Despite the utility of the above model, it possesses measures, is minimized by allowing the use of fuzzy
certain limitations which need to be addressed. variables or descriptors.
These include: its local nature and an inherent diffi- To reiterate, we have developed two versions of
culty, computational and otherwise, to apply it on a fuzzy mathematical programming formulation of
a regional or national level. A more serious concern this problem. It is noteworthy that they both view
is its inability to incorporate satisfactorily and dir- the problem as a fuzzy dynamic programming prob-
ectly certain persistent as well as pervasive systemic lem. In addition to different philosophical problem
variables which are intrinsically fuzzy and imprecise. modeling tact, there are also computational philo-
In other words, the classical approach such as sophical differences. Version 1 is solved by employing
Morin et al.'s suffers from all the well-known objec- a modification of a fuzzy branch and bound proced-
tions to the use of crisp models to represent socio- ure first proposed by Kacprzyk [8] employing as
technical and dynamical large-scale systems. simple downward search algorithm. Version 2 looks
In the present effort, we revisit a novel approach at the problem using a decomposition scheme similar
which we have proposed to address the Disaster to the one we developed in our studies of the cancer
Control Problem (DCP). In particular, we show how research appropriation process [-3].
the tools of fuzzy sets and possibility theory, can be
adroitly combined with methods of large-scale sys- 3.1. Version 1 [6]
tems problem solving theory to solve an otherwise
intractable social problem. To concretize our dis- The details of this model are discussed in the
cussion, we use floods as a manifestation of a disas- above-cited reference. Briefly, the problem is posed
A.O. Esogbue / Fuzzy Sets and Systems 81 (1996) 169-183 173

Table 1
Fuzzy hazard control model, version 1 (Two Optimal Policies)

State space: the flood damage level (no, slight, moderate, severe, disastrous)
Decision Space: the investment level for the measures (low, medium, high)

The membershipfunction of initial state


X0 = 0.1/no + 0.4/slight + 0.7/moderate + 1.0/severe + 0.8/disastrous
The membershipfunction of ooal state
G1 = 0.4/no + 0.6/slight + 0.6/moderate + 0.7/severe + 0.5/disastrous
G2 = 0.9/no + 0.7/slight + 0.5/moderate + 0.3/severe + 0.1/disastrous
G3 = 1.0/no + 0.8/slight + 0.4/moderate + 0.1/severe + 0.0/disastrous
The membershipfunction of constraint for measures
C1 = 0.45/low + 0.85/medium + 0.65/high
C2 = 1.00/low + 0.80/medium + 0.60/laigh
C3 = 0.50/low + 0.70/medium + 0.90/high
The fuzzy transform matrix for state, control and stage (transition mappings)
T 1 (low) T 1 (medium) T 1 (high)
0.3 0.8 0.5 0.3 0.1 0.6 0.9 0.4 0.1 0.0 0.4 0.6 0.8 0.2 0.1
0.2 0.3 0.8 0.5 0.3 0.1 0.6 0.9 0.4 0.1 0.3 0.4 0.6 0.8 0.2
0.1 0.2 0.3 0.8 0.5 0.0 0.1 0.6 0.9 0.4 0.2 0,3 0.4 0.6 0.8
0.0 0.1 0.2 0.3 0.8 0.0 0.0 0.1 0.6 0.9 0.1 0.2 0.3 0.4 0.6
0.0 0.0 0.1 0.2 0.3 0.0 0.0 0.0 0.1 0.6 0.0 0.1 0.2 0.3 0.4
T2(low) T2(medium) T2 (high)
0.4 0.7 0.5 0.3 0.1 0.5 0.8 0.4 0.2 0.1 0.5 0.6 0.8 0.4 0.1 I
0.3 0.4 0.7 0.5 0.3 0.3 0.5 0.8 0.4 0.2 0.3 0.5 0.6 0.8 0.4
0.2 0.3
0.1 0.2
0.0 0.1
T3(low)
0.4
0.3
0.2
0.7
0.4
0.3
0.5
0.7
0.4
0.1 0.3 0.5
0.0 0.1 0.3
0.0 0.0 0.1
T3(medium)
0.8
0.5
0.3
0.4
0.8
0.5
0.2
0.1
0.0
0.3
0.2
0.1
T3 (high)
0.5
0.3
0.2
0.6
0.5
0.3
0.8
0.6
0.5 I
0.3 0.7 0.6 0.4 0.2 0.7 0.9 0.3 0.2 0.0 0.3 0.6 0.9 0.2 0.1
0.2 0.3 0.7 0.6 0.4 0.4 0.7 0.9 0.3 0.2 0.2
0.1 0.2 0.3 0.7 0.6 0.1 0.4 0.7 0.9 0.3 0.2
0.2 0.3
0.2 0.6
0.3 0.9 0.3
0.6 0.9
0.0 0.1 0.2 0.3 0.7 0.0 0.1 0.4 0.7 0.9 0.1 0.2 0.2 0.3 0.6
0.0 0.0 0.1 0.2 0.3 0.0 0.0 0.1 0.4 0.7 0.0 0.1 0.2 0.2

as a m u l t i s t a g e d e c i s i o n m a k i n g p r o b l e m in a fuzzy d(XN, G N) b e t w e e n t h e t w o fuzzy sets:


environment with the system under control repre-
s e n t e d as a c o n d i t i o n e d fuzzy set w h i c h is r e p r e -
s e n t e d v i a its m e m b e r s h i p f u n c t i o n . S i m i l a r l y , t h e
d(XN,GN)= (1) (,=~II~x'(X,)--/~o'(Xl)I) (4)

c o n t r o l s , c o n s t r a i n t s a n d s y s t e m s d y n a m i c s a r e all
e x p r e s s e d in t e r m s o f t h e i r m e m b e r s h i p f u n c t i o n s . It is u s e d w i t h n, t h e n u m b e r of all p o s s i b l e states t h a t
is a s s u m e d t h a t at e a c h s t a g e i, a fuzzy g o a l G i w i t h t h e s y s t e m c a n be in.
membership function F o r e x a m p l e , u s i n g f l o o d as i l l u s t r a t i v e of a
m a j o r disaster, t h e fuzzy s t a t e o f f l o o d d a m a g e is
~G,(xi) (3) r e p r e s e n t e d o n five levels a c c o r d i n g to t h e f o l l o w -
is set, a n d t h e a i m o f t h e c o n t r o l / ~ , is to r e t u r n t h e i n g t e r m sets: n o d a m a g e , slight d a m a g e , m o d e r a t e
s t a t e o f t h e s y s t e m xi as close as p o s s i b l e to a p r e - damage, severe damage, and disastrous damage.
d e t e r m i n e d o n e g i v e n b y G ~. T o m e a s u r e t h e close- T h e d e c i s i o n s p a c e is t h e s p a c e o f t h r e e i n v e s t m e n t
ness b e t w e e n XN a n d G u, t h e r e l a t i v e d i s t a n c e levels for e a c h o f the t h r e e f l o o d c o n t r o l m e a s u r e s .
174 A. O. Esogbue / Fuzzy Sets and Systems 81 (1996) 169-183

Table 2
Solution

The calculations are now performed for the various stages beginning with stage 1.
Stage 1
v l ' (low) = 0.45
vl' ( m e d i u m ) = 0.85***
v l ' ( h i g h ) = 0.65
Xlm = 0.4/no + 0.7/slight + 0.9/moderate + 0.8/severe + 0.6/disastrous
1.0 - D[Xlm, G1] = 1.0 - [0.0 + 0.1 + 0.3 + 0.1 + 0 . 1 ] / 5 = 0.88
vl ( m e d i u m ) = 0.85 A 0.88 = 0.85

Stage 2
v2' (low) = 1.00 A 0.85 = 0.85***
v2' ( m e d i u m ) = 0.80 A 0.85 = 0.80***
v2' (high) = 0.60 A 0.85 = 0.60
X 2 m 1 = 0 . 7 / n o - 0 . 7 / s l i g h t + 0.7/moderate + 0 . 6 / s e v e r e + 0.4/disastrous
1.0 - D[X2ml,G2] = 1.0 - [0.2 + 0.0 + 0.2 + 0.3 + 0 . 3 ] / 5 = 0.80
X 2 m m = 0 . 7 / n o - 0 . 8 / s l i g h t + 0.8/moderate + 0 . 6 / s e v e r e + 0.5/disastrous
1.0 - D [ X 2 m m , G 2 ] = 1.0 -- [0.2 + 0.1 + 0.3 + 0.3 + 0 . 4 ] / 5 = 0.74
v2 (low) = 0.85 A 0.80 + 0 . 8 0 * * *
v2 ( m e d i u m ) = 0.80 A 0.74 = 0.74**

Stage 3
vY (low) = 0.50 A 0.80 = 0.50
v3' ( m e d i u m ) = 0.70/X 0.80 = 0.70
v3' ( h i g h ) = 0 . 9 0 / x 0.80 = 0.80***
X 3 m l h = 0 . 7 / n o + 0.6/slight + 0.6/moderate + 0 . 4 / s e v e r e + 0.3/disastrous
1.0 - D[X3mlh, G3] = 1.0 - I-0.3 + 0.2 + 0.2 + 0.3 + 0 . 3 ] / 5 = 0.74
v3 (high) = 0.80 A 0.74 = 0.74
v3' (low) = 0.50 A 0.74 = 0.50
v3' ( m e d i u m ) = 0 . 7 0 / k 0.74 = 0.70
v3' ( h i g h ) = 0.90 A 0.74 = 0.74***
X 3 m m h = 0 . 8 / n o + 0 . 6 / s l i g h t + 0.6/moderate + 0 . 5 / s e v e r e + 0.3/disastrous
1.0 - D [ X 3 m m h , G 3 ] = 1.0 - [0.2 + 0.2 + 0.2 + 0.4 + 0 . 3 ] / 5 -- 0.74
v3 (high) = 0.74 A 0.74 = 0.74
The optimal policies are thus both (medium, low, high) and (medium, medium, high)

The control measures could be either structural, The problem then is to determine the optimal com-
nonstructural or a combination of them. Typical bination of controls or measures together with their
examples of structural measures used in the United funding levels so as to minimize the damage levels
States are levees, flood walls, channel improvements, due to incipient floods.
and storage reservoirs while the nonstructural The approach proposed by Kacprzyk was modi-
measures include: flood plain zoning, flood proofing, fied and then applied to the flood control problem.
purchase of portions of the flood plain, land use The computerized model was exercised in two
conversion, warning, evacuation, relief and rehabili- sample problems illustrating two different results:
tation, and flood insurance. The measures then rep- (i) a situation where the optimal solution was
resent the stages of the fuzzy decision model. unique and (ii) another where it was not. We pre-
The model has three fuzzy goals, different for sent the case where more than one optimal solution
each measure, and three fuzzy constraints, as well resulted in Table 1, while the solution is given in
as a fuzzy initial system state. Each of these is Table 2. This disadvantage may be minimized by
expressed in terms of their membership functions. the simplicity of the algorithm and the possibility of
A.O. Esogbue / Fuzzy Sets and Systems 81 (1996) 169-183 175

its use in conjunction with the sieve approach of nation, while C.(j) and G.(j) are similarly defined
determine optimal policies in relatively less com- for region n. When used in regions, the symbols
plex scenarios. have the following additional meanings: 1.(i), the
As noted earlier, the essential aspects of a very membership function of initial states in region n;
general model of a fuzzy decision system solved by F.(i), the membership function of final states
branch and bound method was first proposed by in region n; G.(i), the membership function of goal
Kacprzyk [8]. Because of the simple structure of of states in region n; J., the upper bound of total
the model, the solution algorithm involved only investment for region n; and C.k(j), the membership
a single directional search down the branch of function of constraint for measure k in region n.
a decision tree. We note that this is a one level Once again, we note that I.(i), F.(i) and G.(i) are
model, whereas the approach which we propose in defined on the state space of all of the possible flood
Section 3.2 is not and employs a completely differ- damage levels for region n, while C.k(j) is similarly
ent modeling philosophy. defined on the decision space for measure k in
region n. Additionally, we need the fuzzy matrix of
3.2. Version 2 [7] state transform for measure k in region n at the
investment level j. Let this be denoted by T.kj(i, i).
This version of the problem which we proposed We note that T.kj(i, i) is an l x l matrix, where l is
in [7] seems to always provide a unique optimal the dimension of the state space (all of the possible
solution. However, it is computationally more tedi- flood damage levels for region n). It represents the
ous than the first one. We have given a detailed fuzzy relation between the membership function of
exposition of both the model and algorithm [7]. states before and after measure k has been put into
The model has two components, namely, the core use at the investment level j.
and the expanded models while the algorithm has
three levels, namely the regional, national and
a model for coordination. In the sequel, we present
3.2.1. A core fuzzy model of flood control
for the national level." phase 2
only the core model which, without loss of general-
The core model for the problem at the national
ity, gives the bare bones of the problem. The ex-
level or phase 2 may be viewed as that of a multi-
panded version is intended to more closely mirror
stage decision-making process for a nonfuzzy sys-
a given scenario incorporating, for example, certain
tem in fuzzy environment [12]. This is in contrast
realistic constraints. To avoid unnecessary duplica-
to the regional level model in which both the sys-
tion, we omit the regional model and refer the
tem and its environment were considered fuzzy. In
reader to [7]. We however, take the liberty of
this phase, the usual concepts of stage, decision,
providing the national one for ease of reference.
and state are defined as follows: stage, the region
The following symbols and definitions will be
for flood control; decision, the level of total invest-
found useful in the presentation that follows: n, the
ment for region (in US dollars) and, state, the effect
index of region; k, the index of flood control
of flood control for region.
measure; j, the index of flood control investment
As in the regional model, the following input
level; and i, the index of flood damage level.
data (all fuzzy) are necessary: C,(j), the membership
At the national level, Phase 2, the following are
function of constraint for region n (n = i, 2 . . . . ,10);
used: C(j), the membership function of constraint
G.(j), the membership function of goal for region
for the nation; G(j), the membership function of
n (n = 1, 2, ..., 10); and W., the weight or relative
goal for the nation; and Cn(j), the membership
importance of region w.r.t, control n (n = I, ..., i0).
function of investment constraint for region n; G(j),
The resultant fuzzy mathematical program may
the membership function of goal for region n: J, the
then be stated as
upper bound of total investment for the nation;
while W., the weight or criticality of region n. = V {R,(jl) + "-' + R.(j.) + ... + Rlo(J,o)}
In the foregoing, C(j) and G(j) are defined on the J~ ...)~o
set of all of the possible investment levels for the (5)
176 A.O. Esogbue / Fuzzy Sets and Systems 81 (1996) 169-183

s.t. R,(j,) = [Gn(j,) A C,(j,)] W,, levels - regional and national, we present a linkage
program for coordinating the preceding two
n = 1, ... ,10, (6)
phases. This phase is basically a single-stage deci-
W] + ... + W , + ... + W l o = l, (7) sion-making process for a nonfuzzy system in
a fuzzy environment by standard fuzzy decision-
where is the algebraic product operator and making procedures. We note that the resultant
R,(j,) is the return function for stage n, i.e., region models can be viewed as three nested dynamic
n. Solution of the foregoing generates the output programming models thus obviating any concern
data j* and 4) where j* is the optimal investment for duality gaps or similar problems present in
level for region n (n = 1 . . . . ,10) and ~ is the opti- classical decomposition algorithms. The model, re-
mal weighted sum of effect of flood control for the quires the following as input data: C(j) and G(j)
nation. which are, respectively, the membership functions
of constraint and goal at the national level.
3.3 Fuzzy model for coordination." phase 3 The mathematical program is then

Since this is a hierarchical model in which a


= V [G(j) A C(j)]. (8)
national planning problem is decomposed into two j~ [0,J]

Start )

Input the Datafor


Regionn I
P~ sel I
v I
[ C~,eProd'am _,,.1 [Zl

J
r Save the Solution]
of Regionn

[3] Inputthe Datafor


National

PI se II [4] I MergeandPrcess ? ._J


CombinedData

I' 'J
. . . . . . . . . . . . . . .

( s, )
Fig. 3. Flow chart for data requirements and processing.
A,O. Esogbue / Fuzzy Sets and Systems 81 (1996) 169-183 177

Solution of this optimization problem leads to the Table 3


output data j* and ~, where j* is the optimal N o t e s for flow chart a of Fig. 3.
investment level for control management for the
Ill C k, k = 1 . . . . . K
nation, and ~b is the degree to which the optimal Gk, k = 1 , . . . , K
control plan satisfies the national objective. Tk(dj),j = 1 . . . . . J, k = 1, ... ,K, X
A practical algorithm for implementing Phase [2] D R, k = 1 . . . . . K, where D k is the o p t i m u m c o n t r o l level
3 model is basically a three-step procedure which a m o n g dj for kth m e a s u r e
[3] W., n = 1 . . . . . 10
again is described in our previous work and there-
G",n=l,...,lO
fore omitted here. /(, w h e r e / ( is the n u m b e r of best m e a s u r e s d e t e r m i n e d in
Phase 1
[4] (1) X = ~.11 W . X ( n ) where X(n) is the X of region n.
4. Data acquisition and processing (2) T"(d,) = Tk[Dk], k = 1 . . . . . /(, n = 1 . . . . . 10 where D k
is the o p t i m u m control level for the kth m e a s u r e in
region n. Tk[D k] is the t r a n s f o r m m a t r i x for the kth
4.1. Introduction m e a s u r e w i t h c o n t r o l level D k in region n.
(3) G" = W,G", n = 1 . . . . ,10.
The mathematical and computer models that we
have developed and implemented in these models " A l s o see the C o r e P r o g r a m (Fig. 4) for m o r e details.
require various resources for their implementation
in the real world. There are at least two primary
issues. The first deals with data requirements of the constraints, decisions and transitions in terms of
models and some suggested procedures for their their respective membership functions. The deter-
generation. The second is process oriented and mination and measurement of these functions in
addresses computational realization of the models the real world has always been somewhat of
considering computational complexity and com- a thorny problem in the use of fuzzy sets in systems
puter resource requirements. The purpose of this modeling. Various methods, mathematical and
section is to provide some tour of these problems otherwise, for their generation are discussed in
and concerns. [3-5, 15, 16], to name a few. We have suggested an
instructive and practical algorithm for their genera-
4.2. Generation of essential data tion in the case of the problem of current interest.
Essentially, the data acquired from the experts and
As a vehicle for these inquiries, we provide a glo- specialists are converted to membership functions
bal flow chart of the two-phase program omitting, through the practical algorithm displayed in Fig. 8.
without loss of generality, the linkage program.
This chart is given in Fig. 3 and followed by some
brief notes (see Table 3) isolating the kernels of the 5. An illustrative example
program. Central to the first phase is the Core
Program. The essentials of this program are sum- We sketch the basic operation of the algorithm at
marized in Fig. 4. the regional level. For its use at the national level,
The most significant data packets required to the interested reader is referred to Esogbue et al.
implement the models in each of the phases have [6], where a regional example with 5 discretization
been isolated and summarized for ease of reference levels for the state space, 4 for the control space and
in Fig. 5. Equally important is the concern about 4 flood control measures was solved. The regional
the various methods for the acquisition of these solution was [1, 2, 0, 1] or [low, medium, none,
data as well as some possible sources for them. We low] implying a low-level investment on the first
provide this information in Fig. 6. Further explana- flood control strategy, medium on the second, none
tion of data requirements is given in Fig. 7. on the third, and low on the fourth.
It must be recalled that the models represented When the same program was run with the state
the various fuzzy variables such as state, goals, space further discretized into 6 instead of 5 levels,
178 A.O. Esogbue / Fuzzy Sets and Systems 81 (1996) 169-183

Notations

(1) State: S~, i = ] ..... [; Decisioq: dj, j = l ..... J; .~ta,g.~.: k = l , . . . , K .

(2) Membershin Function ( k = l ..... K)

state: X k =(x kI ..... x~k ..... x~)T i.C.,P.x=(S)=Xk/S,+...+X~/Si+ ...+x kI / s I

constraint:
k k k
C = (c z.....c i .....c~)T i.e.,l~c,(d)=c~/d ,+...+c~/di+ ...+c~/d,

goal: O ' = ( g f ..... gf ..... g~)T i'e"~c*(S)= gl/St


, + ' " + g l, / S i + ' " + g ~ / S l

(3)

Tk(dj), j = l ..... ], k = l ..... K

yk(dj) = Tk(dj)oX k-' where o is the max-rain product operator

(4) Return Function

rj = T t[Y (dpJ-g, , j = l ..... J; k--1 ..... K

- k--1 ..... K
J

R' = mill{Rk}, R - , R = UppcrandlowerbotmdsofRrespecfivcly


k

Data (1) Inout (2) O u t o u t

C k , k = l ..... K; Dk, k = l ..... K


G k, k= 1,...,K; R ~, k= 1 ..... K
Tk(dj),j=l .....J, k = l .....K; X k,k=l .....K
X. R', R-= R

Fig. 4. Summary of essentials of the core program.

(i.e. 6 x 6 transition matrix) and different member- 6. Discussion


ship functions for both the constraint and goal
state, the same optimal decision and values were A brief overview of our efforts in modeling the
obtained. The data are given in Table 4, while the optimal control planning for disasters using an
solution is shown in Table 5. adroit combination of structural and nonstructural
The c o m p u t e r output is given in Table 5. control measures via the mathematical technique of
A.O. Esogbue / Fuzzy Sets and Systems 81 (1996) 169-183 179

~ee Notation Meanln8

state
S i, i=l,...,I Discretize
i
~control)
level
d j , J=l . . . . . J

Ck O Constraint for
Gk k=l,...,K .for each measure k uch
Goal region
Tkcdj), j = t . . . . . J Transform Matrix for each

k=t . . . . ,Z measure k a t c o n t r o l l e v e l dj

X Initial S t a t e

Gn , n=l,...,lO Goal f o r each r e s i o n , n


II
Wn, n - l , . . . , l O Weieht f o r each r e $ i o n , n

Fig. 5. Data requirements summary.

Data Method Source

Xo Predictions or forecast Hydrologic and hydraulic

(e.g., time series analysis) experts

AtlP ( A n a l y t i c Hierarchy NFIP specialists

Process)

C,G,T. Delphi FE~A r e g i o n a l o f f i c i a l s

Fig. 6. Possible methods and sources of data stipulated in the model.


180 A.O. Esogbue / Fuzzy Sets and Systems 81 (1996) 169-183

(l) C k: The degree of w i l l i n g n e s s to invest dj on the k-th measure


Ck

T
(2) Gk: T h e d e g r e e of belief that the state level will be attained
w h e n u s i n g the k-th m e a s u r e

1 . . . .

w h e r e a k, .k are th~k l o ~ t and ~ighest s t a t e levels


~spectively that w e can expect when using the k-th measure.
(3) Tk(dj): The r e l a t i o n between the state of the system before
and a f t e r the k-th measure has been e m p l o y e d at the d i
i n v e s t m e n t level, z.e.

X k'1 = Tk(dj) o Xk

w h e r e X k'1, X k are the states after and before; o is the max-min


p r o d u c t operator.

(4) X: The degree of belief that the state level will be at a


c e r t a i n v a l u e using any m e a s u r e i.e. initial state level.

(5) Wn: The relative importance of the n-th region in the scheme of
things.

I0
ZW.= l
n=l

(6) G": The desired state level to be achieved in region n

G"
i .

, , ~_ $
0 an St 8n

where an is the lowest acceDtable state level and 8n i s ~hQ

Fig. 7. E x p l a n a t i o n of d a t a requirements.

fuzzy sets is described. The use of fuzzy sets along bound variety is considered to be an efficient way
with classical dynamic programming models and to resolve, with enhanced reliability and ease, the
solved via a hybrid algorithm of the branch and problem of optimal planning for the control of
A.O. Esogbue / Fuzzy Sets and Systems 81 (1996) 169-183 181

Label for e a c h e v a l u a t o r as follows and consider two


Step 1: types of e v a l u a t i o n ( A & B )

TYPE A : I ' ' ) ~ ' ' ~ - ' L - - ' , - - N - - + - - * - - U - - * - . ~ L - . > j

o. 1h 2h . . . . . . Jh

TYPE I : - " "" " +" "~'- - L- -',- - +-- +- "U" -*- -',- - ) j

o. I t` 2 A . . . . . . Jh

TYPE C : I ......... N ......... I Ruutlty Scare


O. 1.

TYPE O : I ............. x . . . . . I Confidence Lever


O~ I00~

where A: the interval between two value points


L: the l a b e l for t h e l o w e r b o u n d of t h e v a l u e
U: the label for t h e u p p e r b o u n d of t h e v a l u e
M: the label for t h e m o s t p o s s i b l e / p r e f e r a b l e v a l u e
X: the d e g r e e of c o n f i d e n c e for v a l u e l a b e l e d

Step 2: Construct the membership function for each evaluatot as follows:

Type A: )l(j)
I

. . . . _ o t ~ j

o. l& 2A ... ... Jh


0. j<L
[j - (L - A)]/tM - (L - A) ] L _<j<M
or ~l(j) = 1. j=M
t(u + A) - 511[(0 + A) - M] M<j<_U
O. u<j
Type B: /uCj)
I
1.1
~ i " " + ....
II
; _ i i 4
O. 1h ZA I~
.. ... ,I A
i
j<L
or t.(U + A) - J ] I t ( U + A) - (L - All L_<j<U
u<j

Fig. 8. An algorithm for the construction of membership functions.

a complex socio-technical system such as that en- more rational planning of efforts to obviate the
countered in disaster control systems. consequences of such horrible human experiences.
Computerization of these models will aid the
generation of first cut, approximate solutions which
are useful inputs to the decision making process. References
The frequency of occurrence of these disasters to- [1] E.M. Babiker and H. Fujihara, Constructing membership
gether with the enormity of their damages warrant functions for relative distances using an empirical method,
that fresh and useful aids be provided planners for FTT '92 (1992) 103-108.
182 A.O. Esogbue / Fuzzy Sets and Systems 81 (1996) 169-183

Step 3: Data P r o c e s s i n g : Amalgamation o f Q u a l i t y S c a l e and


Confidence Level

S t e p 3.1: Combine the factors, quality scale and


confidence level, i.e., C+D as follows:

Type C+D: /t(i)

|. - . . . . . o o . t . . . . . . .

o. L M u
O. i<L
- LI/[M - L 1 L < i<M
ifM
or /a(i)-1:_ill[U_M l M < i _< U
U<i

where U - M + (I. - X) zad L = M - (i. - X)

step 3.2 s Combine | b membenhip hmctiou from ~ ~sluston u follows:


/a(j) m ~ l ( j ) 4-... +/Up(j) 4-... +pp{J)l/2
where tip(j) : the membership function for pth evlduator
R : total numknr d ~alustors

Fig. 8. Contd.

[2] M.R. Civanclar and H.J. Trussell, Constructing member- [8] J. Kacprzyk, A branch-and-boundalgorithm for the multi-
ship functions using statistical data, Fuzzy Sets and Sys- stage control of a fuzzy system in fuzzy environment,
tems 18 (1986) 1-13. Kybernetes $ (1979) 139-147.
[3] A.O. Esogbue, Dynamic programming, fuzzy sets and the [9] J. Kacprzyk and A.O. Esogbue, Fuzzy dynamic program-
modelling of R&D management control systems, IEEE ming and its applications: a survey, Fuzzy Sets and Systems
Trans. System Man Cybernet. SMC-13 (1983) 18-30. (this issue).
[4] A.O. Esogbue, Dynamic Pro#rammingfor Optimal Water [10] Y.J. Lai and C.L. Hwang, Fuzzy MathematicalProgramming:
Resources Systems (Prentice-Hall, Englewood Cliffs, NJ., Methods and Applications (Springer, Heidelberg, 1992).
1989).
[11] T.L. Morin, W.L. Meier and K.S. Nagaraj, Dynamic pro-
[5] A.O. Esogbue, Computational aspects and applications of gramming for flood control planning: the optimal mix of
a branch and bound algorithm for fuzzy multistage deci- adjustments to floods, in: A.O. Esogbue, Ed., Dynamic
sion processes, Comput. Math. Appl. 21 (1991) 117-127. Programmingfor Optimal Water Resources Systems Analy-
[6] A.O. Esogbue and R.E. Bellman, Fuzzy dynamic program- sis (1989).
ming and its extensions, Fuzzy Sets and Decision Analysis, [12] T.L. Saaty, The Analytic Hierarchy Process(McGraw.Hill,
TIMS Stud. Management Sci. 20 (1984) 147-167. New York, 1980).
[7] A.O. Esogbue, M. Theologidu and K. Guo, On the ap- [13] J.L. Verdegay, Fuzzy mathematical programming, in:
plication of fuzzy sets theory to the optimal flood control M.M. Gupta, E. Sanchez, Eds., Approximate Reasoning in
problem arising in water resources systems, Fuzzy Sets and Decision Analysis (North-Holland, Amsterdam, 1982)
Systems 48 (1992) 155-172. 231-236.
A.O, Esogbue / Fuzzy Sets and Systems 81 (1996) 169-183 183

Table 4
A flood control problem using version 2 model: the data for the regional allocation problem

State space {1, 2, 3, 4, 5, 6} or (no, slight, moderate, grievous, severe, disastrous}


Decision space {0, 1, 2, 3} or {no, low, medium, high}
Initial state:
l = 0.13/no + 0.35/slight + 0.57/moderate + 0.78/grievous + 0.95/severe + 0.69/disastrous
Goal state: G = 1.00/no + 0.85/slight + 0.65/moderate + 0.40/grievous + 0.10/severe + 0.00/disastrous
Constraint: C1 = 1.00/no + 0.80/low + 0.57/medium + 0.31/high
C2 = 1.00/no + 0.35/low + 0.89/medium + 0.19/high
C3 = 1.00/no + 0.59/low + 0.68/medium + 0.36/high
C4 = 1.00/no + 0.75/1ow + 0.60/medium + 0.27/high
Transform matrix:
T 1 (low) T1 (medium) Tl (high)
0.3,0.9,0.4,0.2,0.1,0.0 0.4, 0.7, 0.3, 0.1,0.0, 0.0 0.4, 0.6, 0.2, 0.0, 0.0, 0.0
0.1,0.3,0.9,0.4,0.2,0.1 0.2, 0.4, 0.7, 0.3, 0.1, 0.0 0.3, 0.4, 0.6, 0.2, 0.0, 0.0
0.0,0.1,0.3,0.9,0.4,0.2 0.1,0.2,0.4,0.7,0.3,0.1 0.2, 0.3, 0.4, 0.6, 0.2, 0.0
0.0, 0.0, 0.1, 0.3, 0.9, 0.4 0.0,0.1,0.2,0.4,0.7,0.3 0.1,0.2,0.3,0.4,0.6,0.2
0.0,0.0,0.0,0.1,0.3,0.9 0.0,0.0,0.1,0.2,0.4,0.7 0.0,0.1,0.2,0.3,0.4,0.6
0.0,0.0,0.0,0.0,0.1,0.3 0.0, 0.0, 0.0, 0.1, 0.2, 0.4 0.0,0.0, 0.1, 0.2, 0.3, 0.4
T2 (low) T2 (medium) T2 (high)
0.4,0.7,0.4,0.2,0.1,0.0 0.3,0.9,0.3,0.1,0.0,0.0 0.4, 0.6, 0.3, 0.0, 0.0, 0.0
0.1,0.4,0.7,0.4,0.2,0.1 0.2,0.3,0.9,0.3,0.1,0.0 0.3, 0.4, 0.6, 0.3, 0.0, 0.0
0.0,0.1,0.4,0.7,0.4,0.2 0.1,0.2,0.3,0.9,0.3,0.1 0.2, 0.3, 0.4, 0.6, 0.3, 0.0
0.0,0.0,0.1,0.4,0.7,0.4 0.0, 0.1, 0.2, 0.3, 0.9, 0.3 0.1,0.2,0.3,0.4,0.6,0.4
0.0,0.0,0.0,0.1,0.4,0.7 0.0,0.0,0.1,0.2,0.3,0.9 0.0,0.1,0.2,0.3,0.4,0.6
0.0,0.0,0.0,0.0,0.1,0.4 0.0,0.0,0.0,0.1,0.2,0.3 0.0,0.0,0.1,0.2,0.3,0.4
T3 (low) T3 (medium) T3 (high)
0.6,0.5,0.4,0.3,0.1,0.0 0.4,0.7,0.3,0.1,0.0,0.0 0.5, 0.6, 0.2, 0.0, 0.0, 0.0
0.1,0.6,0.5,0.4,0.3,0.1 0.2,0.4,0.7,0.3,0.1,0.0 0.3,0.5,0.6,0.2,0.0,0.0
0.0,0.1,0.6,0.5,0.4,0.3 0.1,0.2,0.4,0.7,0.3,0.1 0.2, 0.3, 0.5, 0.6, 0.2, 0.0
0.0,0.0,0.1,0.6,0.5,0.4 0.0,0.1,0.2,0.4,0.7,0.3 0.1,0.2,0.3,0.5,0.6,0.2
0.0, 0.0, 0.0, 0.1,0.6, 0.5 0.0,0.0,0.1,0.2,0.4,0.7 0.0,0.1,0.2,0.3,0.5,0.6
0.0, 0.0, 0.0, 0.0, 0,1, 0.6 0.0,0.0,0.0,0.1,0.2,0.4 0.0,0.0,0.1,0.2,0.3,0.5
T4 (low) T4 (medium) T4 (high)
0.3,0.9,0.4,0.1,0,0,0.0 0.4,0.7,0.3,0.1,0.0,0.0 0.4, 0.5, 0.2, 0.0, 0.0, 0.0
0.1,0.3,0.9,0.4,0.1,0.0 0.2,0.4,0.7,0.3,0.1,0.0 0.3,0.4,0.5,0.2,0.0,0.0
0.0,0.1,0.3,0.9,0.4,0.1 0.1,0.2,0.4,0.7,0.3,0.1 0.2, 0.3, 0.4, 0.5, 0.2, 0.0
0.0,0.0,0.1,0.3,0.9,0.4 0.0,0.1,0.2,0.4,0.7,0.3 0.1,0.2,0.3,0.4,0.5,0.2
0.0,0.0,0.0,0.1,0.3,0.9 0.0,0.0,0.1,0.2,0.4,0.7 0.0,0.1,0.2,0.3,0.4,0.5
0.0,0.0,0.0,0.0,0.1,0.3 0.0, 0.0, 0.0, 0.1,0.2, 0.4 0.0,0.0,0.1,0.2,0.3,0.4

Note: The problem is solved via the hybrid dynamic programming-branch and bound algorithm whose flowchart is
provided in the cited reference. The optimum solution is 0, 75 and the associated policy for this region is [0,1, 2, 0,1] or
[none, low, medium, low]. In other words, the recommendation is not to invest in control strategies 1 and 3, to make low
investments on strategies 2 and 4, and then medium or moderate investment on the 4th strategy.

Table 5

Stage, Constraint, Goal, Decision, State:X(1),X(2) .... ,X(I)

0 0.4383 0.130 0.350 0.570 0.780 0.950 0.690


1 0.8000 0.6000 1 0.400 0.570 0.780 0.900 0.690 0.300
2 0.8900 0.7433 2 0.570 0.780 0.900 0.690 0.300 0.300
4 0.7500 0.8483 1 0.780 0.900 0.690 0.300 0.300 0.300

[Optimal solution]0.7500 = [constraint]O.7500][goal] 0.8483

[14] R.R. Yager a n d L.A. Zadeh, Eds., An Introduction to Fuzzy Zadeh, Eds., An Introduction to Fuzzy Logic Applications
Logic Applications in Intelligent Systems (Kluwer Aca- in Intelligent Systems (Kluwer Academic, Boston, 1992)
demic, Boston, 1992). 97-120.
[15] H.J. Z i m m e r m a n n , M e t h o d s a n d applications of fuzzy [16] H.J. Z i m m e r m a n n , Fuzzy Sets and its Applications,
m a t h e m a t i c a l p r o g r a m m i n g , in: R.R. Yager and L.A. (Kluwer Academic, Boston, 2nd edn., 1992).

You might also like