ARTICLE IN PRESS

Energy Policy 35 (2007) 5643–5655

www.elsevier.com/locate/enpol

Fuzzy possibilistic model for medium-term power generation planning

with environmental criteria
E. Muelaa,, G. Schweickardtb, F. Garcésa
a
Instituto de Energı´a Eléctrica (IEE), Universidad Nacional de San Juan, Av. Lib. Gral. San Martı´n 1109 oeste, 5400 San Juan, Argentina
b
Instituto de Economı´a Energética—Fundación Bariloche, Bariloche, Argentina
Received 3 April 2007; accepted 18 May 2007
Available online 1 August 2007

Abstract

The aim of this paper is to apply a fuzzy possibilistic model to the power generation planning that includes environmental criteria.
Since it is not always meaningful to relate uncertainty to frequency, the proposed approach analyzes the imprecision and ambiguity into
the decision making, especially when the system involves human subjectivity. This paper highlights the subjacent differences between
fuzzy and possibilistic entities. Additionally, it illustrates the use of fuzzy sets theory and possibility theory for modeling flexibility, and
nonprobablistic uncertainty, respectively. The necessity of a new direction for the environmental problem in a power system is outlined,
an approach that attempts a greater integral quality of planning instead of searching for a simple optimal solution. This process must be
consistent with a wider and more suitable interpretation about both the problem as such and the concept of solution in uncertain
situations.
r 2007 Elsevier Ltd. All rights reserved.

Keywords: Environmental power generation planning; Fuzzy possibilistic decision making; Multi-objective optimization

1. Introduction well as the natural environment. Although these impacts

have a direct bearing on individuals’ well-being, the
Nowadays, there is a growing recognition that the impacts are not usually factored into any of the decisions
current growth of human activity cannot continue without to generate or consume electric power (Rahman and De
significantly affecting the environmental quality. Then, new Castro, 1995). Additionally, the increase of electrical
instruments are required to handle sustainability issues in demand raises concerns about the environment’s ability
different areas of human activity. to sustain this development without harm to itself; there-
All over the world, electricity remains to be a funda- fore, it is indispensable to develop adequate decision-
mental element of national development. Obviously, making tools in order to face the environmental problems
energy production and consumption is connected to from the perspective of power systems (Soderholm and
environmental pressure in many aspects. Particularly, Sundquist, 2003).
environmental damage of power generation can be quite On the other hand, power generation planning is subject
significant. In electricity generation, the emissions, dis- to important sources of uncertainty related to different
charges, and other effects of power production affect the variables, such as physical, technical, economic, environ-
health of nearby and sometimes distant populations, as mental, etc. Uncertainty is an unavoidable component of
such a procedure. This uncertainty may have essentially
two origins: randomness due to natural variability of the
Corresponding author. Tel.: +54 2644226444; fax: +54 2644210299. observed phenomenon resulting from heterogeneity or
stochasticity and imprecision due to lack of information
E-mail addresses: [email protected] (E. Muela),
[email protected] (G. Schweickardt), resulting, for example, from systematic measurement error
[email protected] (F. Garcés). or expert opinion (Dubois and Prade, 2000).

0301-4215/$ - see front matter r 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.enpol.2007.05.020
 ARTICLE IN PRESS
5644 E. Muela et al. / Energy Policy 35 (2007) 5643–5655

Nomenclature APnh hydro inflows expressed in energy for hydro

unit h in period n
Variables CEks emission coefficient of pollutant k of thermal
unit s
Cd nm unserved power cost in load level m of period n CFs fuel cost of thermal unit s
C nsm production cost of unit s in load level m of DM nm forecasting customer power demand in load
period n level m of period n
CT system’s total economic production cost dAP deviation of inflows from forecasting value
DNSnm unserved power in load level m of period n dCFs deviation of fuel cost of thermal unit s from
dv normalized deviation of an attribute decision forecasting value
Eknsm emissions of the polluting agent k of unit s in dCEks deviation of emission level of pollutant k of
load level m of period n thermal unit s from forecasting value
ETk total emissions of polluting agent k dDem deviation of customer power demand from
PH nhm output power of hydro unit h in load level m of forecasting value
period n dRe maximum reserve deviation from nominal
PT nsm output power of thermal unit s in load level m value
of period n ET maxk maximum amount of emission during the
SPnh water released in reservoir h of period n planning horizon of pollutant k
V nh hydro energy reserve of hydro unit h of period n N number of periods for optimization
pnDEMm possibility distribution of demand constraint in NB number of load levels of demand
load level m of period n NG number of thermal units of the system
pnAPh possibility distribution of hydro constraint of NH number of hydro units of the system
dynamic water balance equation for reservoir h PH maxh maximum rate capacity of hydro unit h
in period n PT maxs maximum rate capacity of thermal unit s
pCF s possibility distribution of fuel cost of unit s PT mins minimum rate capacity of thermal unit s
pCE ks possibility distribution of emission level of ReM nm nominal reserve requirement in load level m of
polluting k of unit s period n
mCT membership function of economic objective tm duration of load level m (h/month)
mCO2 membership function of environmental attri- V maxh maximum hydro energy reserve of hydro unit h
bute CO2 V minh minimum hydro energy reserve of hydro unit h
mSO2 membership function of environmental attri- Wf relative weighting of decision-maker about
bute SO2 standardized deviation between calculated va-
mNOx membership function of environmental attri- lue of attribute decision f and its ideal value f*
bute NOx
mnRes m membership value of reserve constraint in load Sets
level m of period n
F qualitative utility of planning h hydro units
k environmental attributes
Constants m load levels
n optimization periods
a1s lineal coefficient of heat rate curve of thermal s thermal units
unit s
a0s independent coefficient of heat rate curve of
thermal unit s

When the environmental variable is included in genera- fuzzy logic, and possibility theory are proposed as tools
tion planning, it is necessary to deeply analyze the attached which provide a greater quantity of information for a
characteristics and uncertainty of this variable and, better decision-making.
consequently, to examine the current mechanisms for The proposed model is basically defined as a tool of
dealing with environmental issues. As a consequence of support for planning. Then, this tool can benefit electric
previous analysis, the necessity for alternative focuses as and environmental regulators, which pursue to achieve
well as the use of nontraditional techniques which could environmental objectives within a planning process. This
strengthen the decision-making process will be justified. In happens because the described model allows to take
this context, structures such as multi-criteria paradigm, advantage from the available information, specifically
 ARTICLE IN PRESS
E. Muela et al. / Energy Policy 35 (2007) 5643–5655 5645

those of qualitative origin. Additionally, this model imprecise, and there is no agreement for adopting a
supplies relevant information by means of possibility standard mechanism; moreover, the effects in their total
distributions of state vectors from the analyzed system. magnitude will only be known in the future.
This situation enables a most flexible estimation of results, Ideally, one should base decisions on a large body of
because a range of solution values is obtained, instead of reliable and measured data. In air pollution problems one
one concrete value whose reliability and relevance is generally has a choice of models and ideally one should use
comparatively minor. measurement data to choose the best model. This is a
The new mechanism can be integrated within the situation when measured data are available, where
traditional model (probabilistic) through the comparison uncertainty is due to the variability of the observed
of the results obtained by one and another model, this is, to phenomenon, and it is handled by means of probability
observe if the solutions of a method support or they theory. However, the incomplete or imprecise information,
confront to the solutions of the other method. In the case the uncertainty in the models, and paucity of measure-
of contradictions, it will be necessary to determine the ments often make this difficult. For example, to assess
causes in order to improve or even change the initial pollution from thermal power generation, the uncertainty
hypotheses for analysis. Another alternative form to about emission levels for each generation unit is consider-
integrate the two procedures is by means of the definition able. In this new situation, uncertainty is captured more
of a model that combines both mechanisms. Such proposal naturally by set-valued representations where all that is
is still in a research status of the theoretic foundations, and known is that a certain value belongs to a certain set, which
whose development escapes to the scope of this work. is possibly fuzzy. This is the idea of possibility which will be
shown later.
2. Environmental variable On the other hand, the evaluation of environmental
quality usually consists of comparing a quantity (e.g., total
Although it may seem strange, carefulness of the emissions) against an environmental criterion. Poor
environment was not a well-known activity some decades environmental quality arises when this quantity exceeds
ago. It was rather stimulated by definite scientific an objective and good environmental quality is said to arise
disciplines in a very isolated way. Today, the situation when an objective is not exceeded. This paper recognizes
has changed. Ecological problems have reached such two important facts. First, it is too difficult to set
proportions that it is perceived a growing consciousness objectives, and second, the boundaries between an accep-
that mankind is guiding the planet towards extreme table and an unacceptable situation should not be
situations in its sustaining capacity. considered as sharp, but as fuzzy, with implications for
Ecological issues outline an enormous challenge. It is to subsequent action plans. Moreover, the involved uncer-
build a way of life, a society, which must be able to develop tainty introduces bigger fuzziness into the boundary
in a harmonious and sustainable way without forgetting between good and poor environmental quality.
their environment. It does not seem that a project of such Therefore, one can sense the great difficulty to include
magnitude can be achieved by means of policies whose this criterion in decision-making. Obviously, this work
impact only reach the problem surface by introducing seeks neither an exhaustive analysis nor definitive answers,
corrective actions in the economic system and new because it is probable that a unique solution does not exist.
techniques in the productive processes. This situation Its objective is to supply some additional elements for
demands to go further on. As it has been proposed by a improving the analysis in a specific human activity that is
growing number of authors within the environmental the electrical industry.
discipline, the same structures of social organization
should be analyzed from an ecological viewpoint by 3. Fundamental concepts
searching for weak points, which can make them inop-
erative to reach the ideal of sustainable development As a natural consequence of complexity of the environ-
(Moreno et al., 2001). mental variable, it is necessary to use alternative tools
In the last years, regulations of atmospheric emissions which allow obtaining relevant information to improve
have become more formal with the application of decision-making. Next, the use of such tools will be
standards for quality air in Europe and the United States justified.
(EMEPE, 2005; EPA, 2002). Although each one of these
regulations searches for improving the environment, the 3.1. Multi-criteria approach
numbers which are used in decision-making, e.g., cost or
constraint of emissions, may not appear as clear and Human activities are related to decisions. Essentially,
consistent as it is thought. It happens because the used decision-making attempts to select the best action alter-
models could fail in describing the pervasive and inherent native among available options. As a consequence of this
uncertainty appropriately. For example, today it is process, it is expected to obtain good results which are
considered that produced damages are only perceptible, evaluated by means of utility functions. For simplicity,
but an appropriate measure of the impact is still very these decisions habitually have been guided by a single
 ARTICLE IN PRESS
5646 E. Muela et al. / Energy Policy 35 (2007) 5643–5655

criterion, but in most of real processes several criteria (Nelson, 2005), where the decision-maker looks for
influence the decision. determining satisfactory conditions instead of an optimal
Traditional models of decision-making, which come solution for the analyzed system. This situation can be
from operation research (OR), regard additional criteria interpreted by analogy like ‘‘bounded optimality’’, where
indirectly through constraints or sensibility analysis. In the satisfaction is related to the decision-maker’s perception
past, this reduction of the problem was fundamentally due about the fulfillment degree on his requirements. This
to the fact that the optimization process was restricted to concept permits a different vision of planning from the
an analysis in the space of alternatives, which is excessively traditional one, since it opens the possibility for linking
complex and even more if several criteria are considered. objective and subjective aspects of the process, and to
On the contrary, a multi-criteria approach seeks to map the determine an alternative mechanism to evaluate the quality
problem from solutions space to objectives space, and or merit of a solution.
reduce the problem complexity, which allows for inter-
mediate analysis (Hobbs and Meier, 1994). 3.3. Fuzzy mathematical programming
Analysis of environmental policies is certainly a multi-
criteria problem involving a variety of decision makers Mathematical programming forms a subclass of deci-
with different objectives, levels of understanding of the sion-making process where preferences among different
underlying economics, science and impacts, and levels of alternatives are described by means of one or several
sophistication concerning the analytical tools. Energy objective functions. The more an alternative reaches high
planning, too, has become a multi-criteria problem with values on these functions, the more it is preferred. The
the increasing complexity of energy policy formulation, results and application depend on the way that several
particularly with the addition of increasing concerns about conditions of real system are reflected in the description
the interaction between energy production and consump- of objectives and constraints. Usually, the descri-
tion, and the environment (Greening and Bernow, 2004). ptions include some parameters. Clearly, the values of
the parameters arise from multiple factors which are
3.2. Imperfect information in decision-making not completely included in the problem formulation
(as described in epigraph B).
As expressed by Bellman and Zadeh (1970), most In this context, fuzzy mathematical programming, based
decisions in the real world are carried out in situations on both fuzzy set theory and possibility theory, is
where objectives, constraints, possible actions (solutions developed for treating such situations in the setting of
space), and their consequences are not known accurately. optimization problems. Fuzzy programming supplies a
In order to deal with imprecision in a quantitative way, natural structure to model vague concepts, for example,
traditional planning appeals to techniques and concepts of bigger, smaller, equal, polluting, economic, satisfactory,
probability theory. When one does this, implicitly, it is adequate, among others, and besides, it allows reducing the
accepted that imprecision, of any nature, can be compared problems complexity by means of the systematic inclusion
with the randomness. This assumption can be debatable. It of uncertainty results from imperfect and imprecise
is not always meaningful to relate uncertainty to frequency, information, related in part with ambiguity.
because some types of events can be rare, unrepeatable, or In this point, it is necessary to make a clarification. There
statistical data may simply be unavailable (Smets, 1997). is often confusion between fuzzy and possibilistic optimi-
In this context, fundamental uncertainty is defined zation. Fuzzy and possibilistic entities have different
including not only the occurrences of results (coming from meanings/semantics. Fuzzy and possibility model different
ignoring the future), but also those which originated by the entities and the associated solution methods are also
absence of crisp limits in values that a variable can take different (Dubois and Prade, 1997, 2001).
(Schweickardt, 2004). It happens because on many occa-
sions the decision-maker cannot measure or express in an 3.3.1. Fuzzy sets and fuzzy optimization
explicit function all critical factors of decision-making Fuzzy entities are sets with nonsharp boundaries in
because of different reasons, e.g., incomplete information, which there is a transition between elements that belong
insufficient knowledge, absence or imprecision of models and elements that do not belong to the set. In this situation,
(imperfect information), or the cost for acquiring informa- there is no uncertainty (Inuiguchi and Ramik, 2000).
tion is not compensated by a substantial improvement of This sort of fuzzy mathematical programming was
results. Additionally, in some occasions it is necessary to initially developed by Bellman and Zadeh (1970). It treats
appeal to another class of information, e.g., experience, the decision-making problem under fuzzy goals and
intuition, ideology, beliefs, feelings, etc. These options have constraints. The fuzzy goals and constraints represent the
a degree of subjectivity, but not arbitrariness (Buchanan et flexibility of the target values of objective functions and
al., 1998; Hening and Buchanan, 1994). the elasticity of constraints. Fuzzy optimization selects
All antecedents let us question the concept of optimal from a set of crisp elements. The underlying sets asso-
solution, and the effort invested to determine it. The ciated with fuzzy decision-making are fuzzy, where one
described context defines a state of bounded rationality forms the decision space of crisp elements from operations
 ARTICLE IN PRESS
E. Muela et al. / Energy Policy 35 (2007) 5643–5655 5647

And the degree of general membership of the alternative

x is defined as

mD ðxÞ ¼ Min½mOi ðxÞ; mRj ðxÞ; i ¼ 1; . . . ; z and j ¼ 1; . . . ; v.

(4)
Optimal solution x* is obtained fulfilling the next
relation

mD ðx Þ ¼ Max½mD ðxÞW , (5)

x

Fig. 1. Feasible region for Aa. where W, in general, represents a structure of preference
over attributes O and R, which is included in decision-
making.
on these fuzzy sets. From this point of view, this type of
fuzzy mathematical programming is called the flexible
programming.
Next, basic definitions on fuzzy sets are introduced 3.3.2. Possibility theory and possibilistic optimization
(Zimmermann, 1996). Let X be a collection of objects Possibilistic entities are obtained from sets that are
~ in X corresponds
generally denoted by x, then a fuzzy set A classical sets (crisp), but the evidence associated with
to the group of ordered pairs: whether a particular element belongs to the (crisp) set
or not is incomplete or hard to obtain. Then, there is
A~ ¼ fðx; mA~ ðxÞÞ=x 2 X g, uncertainty (Dubois and Prade, 2000).
In Zadeh (1978), the possibility theory on the basis of
where mA(x)A[0,1] represents a membership function of fuzzy sets theory is developed. Zadeh proposed the idea
~ Its shape depends on the concept that A~ defines as
x in A. of representing an incomplete state of knowledge by means
well as objective and subjective information that it has at of a fuzzy set. If for instance, we only know about quantity
one’s disposal on this concept. X that ‘‘X is large’’, it means that the possible values for X
The a-level set of a fuzzy set A~ is defined as an ordinary are those compatible with the meaning of ‘‘large’’ in the
(crisp) set Aa for which the degree of its membership considered context. Such a label is represented by the
function exceeds the level a: membership function of a fuzzy set, i.e., px ¼ mlarge , where
px denotes the possibility distribution describing the more
Aa ¼ fx 2 O=mA~ ðxÞXag. or less possible values for X according to what is known.
A possibility distribution px on U is a mapping from U
It should be noted here that the feasible region for Aa to the unit interval [0,1] attached to the single-valued
can be denoted by the closed interval ½aLa ; aR
a , shown in variable X. The function px represents a flexible restriction,
Fig. 1. which constrains the possible values of X according to the
Additionally, for two fuzzy sets A and B: available information, with the following conventions:

mA[B ðxÞ ¼ Max½mA ðxÞ; mB ðxÞ, (1) px(u) ¼ 0 means that X ¼ u is definitely impossible,
px(u) ¼ 1 means that absolutely nothing prevents that
X ¼ u.
mA\B ðxÞ ¼ Min½mA ðxÞ; mB ðxÞ. (2)
Intermediary levels of plausibility about the possible
Union and intersection operations are represented by
values of X are modeled by letting px(u) between 0 and 1
logical operators ‘‘OR’’ and ‘‘AND’’, respectively.
for some values u. The quantity px(u) thus represents the
With these definitions, in (1) three basic concepts are
degree of possibility of the assignment X ¼ u. Then one can
introduced: fuzzy objectives, fuzzy constraint, and fuzzy
acknowledge the fact that some values of u are more
decision; and starting from them, the decision-making in
possible than others, according to available information.
fuzzy environments is explored.
From a possibility distribution, seen as a repository of
Let X ¼ (x1, x2,y, xm) be a group of m alternatives,
knowledge about a variable X, one can build different
O ¼ (o1, o2, y, oz) a group of z objectives, R ¼ (r1, r2, y,
uncertainty measures to characterize what can be said
rv) a group of v constraints, and mOi ðxÞymRj ðxÞ membership
about any event (i.e., a subset of the domain of X). The
functions of alternative x on objective i and constraints j,
most commonly used ones are the possibility and necessity
respectively. Then, a fuzzy decision function D which
measures, denoted by P and N, respectively, that are
simultaneously satisfies to some degree all objectives and
defined as follows (Dubois and Prade, 2000):
restrictions is shown:
PK ðAÞ ¼ sup min½mA ðuÞ; px ðuÞ, (6)
mD ðxÞ ¼ mO ðxÞ \ mR ðxÞ. (3) u2A
 ARTICLE IN PRESS
5648 E. Muela et al. / Energy Policy 35 (2007) 5643–5655

N K ðAÞ ¼ 1  PK ðAc Þ ¼ inf max½mA ðuÞ; 1  px ðuÞ, (7)

ueA

where Ac denotes the set complementation of A.

Seeing px as the membership function of a fuzzy set K
(available information), then the possibility measure
expresses the degree of intersection (compatibility) of A
and K, and the necessity measure expresses the degree of
inclusion of K in A, and corresponds to an evaluation of
the extent to which the state of knowledge K implies A. The
possibility of an event evaluates its degree of unsurprising
(plausibility) whiles necessity evaluates its grade of
acceptance (certainty). As shown in (7) the necessity
measure evaluates the impossibility of the opposite event.
From possibility theory, a different sort of fuzzy
mathematical programming is developed. It treats ambig-
uous or imprecise coefficients of objective functions
and constraints but does not treat fuzzy goals and
constraints. Since imprecision is quantified by means of
fuzzy sets that represent the continuous possibility
distributions for ambiguous parameters, this type of fuzzy
mathematical programming is usually called the possibili-
Fig. 2. Possibility and necessity measure of bpg.
stic programming.
Because the coefficients are ambiguous, the objective
functions and constraints are also ambiguous or possibi- Now, if we formulate a possibilistic programming
listic. Indeed, the range of a possibilistic function value is problem as
restricted by a fuzzy set since the possibilistic variable
coefficients are restricted by fuzzy sets. The fuzzy sets max Z~ ¼ C~  x;
(10)
which restrict the possibilistic function value are defined by subject to A~  xpB~ with xX0;
the extension principle.
~ B,
where x is the vector of decision variables; A; ~ and C~ are
A possibilistic function value cannot be determined
uniquely since its coefficients are ambiguous, i.e., non- fuzzy quantities.
deterministic (Inuiguchi and Ramik, 2000). Thus, the Given an appropriate level h of certainty or possibility,
objective, maximizing (or minimizing) a possibilistic func- and being the necessity and possibilistic measures applied
tion, and the constraint that a possibilistic function value in (10), the fuzzy formulation is as follows:
is not greater (or lesser) than a certain value do not max Z~ ¼ C~  x;
specifically make sense. To make them clear, we must
NesðC~  xXZÞXh
~ (11)
introduce an example which is based on a specific subject to
interpretation, particularly, fuzzy inequality or ranking ~ ~
NesðA:xpBÞXh;
relations between A and K.
Let b be a possibilistic variable restricted by a possibility
PosðC~  xXZÞXh,
~
distribution px of a fuzzy set K. Let A ¼ ð1; gÞ, i.e., A be
a crisp (nonfuzzy) set of real numbers, which is not greater PosðA~  xpBÞXh.
~ ð12Þ
than g. Then we obtain the following possibility and
Using a-level set concept, problems (11) and (12) are
necessity measures defined by (6) and (7):
equivalent to the following crisp programming problems,
PosðbpgÞ ¼ PK ðð1; gÞ ¼ supfpx ðuÞ=upgg, (8) respectively (Ramı́k, 2006):
max zh ¼ cL1h x
NesðbpgÞ ¼ N K ðð1; gÞ ¼ 1  supfpx ðuÞ=u4gg. (9) (13)
L
subject to aR
1h xpb1h ;
Pos (bpg) and Nes (bpg) show the possibility and
certainty degrees where b is not greater than g. These
indices are depicted in Fig. 2. max zh ¼ cR
hx
(14)
Since a possibilistic function value fi is a possibilistic subject to aLh xpbR
h:
variable restricted by a possibility distribution Fi, we can
substitute fi for b and Fi for K in (8) and (9). Thus, we can In these cases, the underlying sets associated with
get the possibility and certainty degrees where a possibi- possibilistic decision-making are crisp where one forms
listic function value is not greater (smaller) than a given the decision space of distributions from operations on
real number. crisp sets.
 ARTICLE IN PRESS
E. Muela et al. / Energy Policy 35 (2007) 5643–5655 5649

3.4. Qualitative decision-making Optimistic criterion:


E ðxÞ ¼ supfminðpðuÞ; mðxðuÞÞÞg. (16)
In the field of decision-making under uncertainty, several u2U
important results that justify the use of various criteria
These criteria are clearly based on the necessity and
for the comparison of alternatives have been published.
possibility of the fuzzy event with membership function m (v).
Each decision criterion is justified by a set of axioms
The following representational conventions are assumed
putting constraints on the choice behavior of the decision
for possibility distributions. p(u) ¼ 0 means that u is definitely
maker.
impossible according to what is known, and m(v) ¼ 0 that v is
The most classical results are found in the seminal works
unacceptable as a consequence. The greater p(u) (resp. m(v)),
of Von Neumann and Morgenstern (1947) and Savage
the more plausible u as being the real state of the world
(1972). Such approaches rely on the use of a quantitative
(respectively, the more acceptable v as a consequence).
criterion for the comparison of alternatives (e.g., expected
p(u) ¼ 1 means u is among the most plausible (normal)
utility.) The justification of this criterion requires
states and there may be several u’s such that p(u) ¼ 1, and
several strong assumptions. For instance, the axiom
m(v) ¼ 1 that v is among the most preferred consequences.
system usually implies that the subjective value attached
to each consequence, as well as the degrees of confidence
of the possible events, can be quantified. However, in 4. Fuzzy possibilistic model for multi-objective generation
practical applications, the elicitation of the information planning
required by a quantitative model is often not an easy
task. Uncertainty and preferences are not always directly 4.1. Problem description
assessed by means of probability or utility functions, but
are rather expressed through collections of pieces of Traditionally, power generation planning is a medium-
information (qualitative) which implicitly constraint such term problem, which determines basically: future costs of
functions. exploitation; requirements of energy in order to cover the
That is why some alternative models have been proposed future demand; fuel consumption; and level of production
in artificial intelligence (AI), relying on a more ordinal for different technologies of the system among other
(qualitative) representation of preferences and uncertainty. results. In this classical planning, a simple objective
This qualitative nature agrees with the fact that only poor function of economic cost associated with the operation
and incomplete information is available in many practical of generation is optimized, subject to a set of technical
situations. This is the case of decision-making that involves constraints. If environmental constraints or costs exist,
both fuzzy and possibilistic entities, e.g., when we face they are included without problem in the model. However,
ambiguous coefficients as well as vague decision-maker’s as it has been exposed, such vision could be very limited.
preferences. In this kind of model, the vague preference is Thus, this paper provides a fuzzy possibilistic multi-
represented by a fuzzy satisfactory region, and a possibi- objective model for power generation planning. Such
listic function value is required to be included in the given model includes environmental criteria as an objective in
fuzzy satisfactory region. decision process and deals with nonprobabilistic uncertain-
One of these models is the qualitative utility function, ties in power generation planning.
which was proposed by Dubois and Prade (1995). In In order to clarify the proposed model, first, traditional
contrast to expected utility, qualitative utility makes sense formulation of single objective is introduced, and next,
if information is qualitative, i.e., we represent both the fuzzy-possibilistic version of the problem is presented.
incomplete knowledge on the actual state by a possibility
distribution p on U (set of states) with values in a 4.2. Traditional formulation
plausibility scale L, and the decision-maker’s prefe-
rence on V (set of possible consequences) by means  Objective function (minimize economic cost):
of another possibility distribution m with values on a
preference scale T. X
N X
NB X
NG

The utility of a decision x whose consequence in state CT ¼ fa0s EOnsm þ a1s PT nsm gCF s tm
n¼1 m¼1 s¼1
u is v ¼ xðuÞ 2 V for all states u, can be evaluated by
combining the plausibilities p(u) and the utilities m(v) in a
N X
X NB
þ fCd nm DNS nm gtm . ð17Þ
suitable way. Two qualitative criteria that evaluate the n¼1 m¼1
worth of decision x have been put forward in the literature
of fuzzy sets, provided that a commensurability assum-
ption between plausibility and preference is made, they are  System constraints (Demand):
as follows:
Pessimistic criterion: X
NG X
NH

E  ðxÞ ¼ inf fmaxð1  pðuÞ; mðxðuÞÞÞg. (15) Dnm ¼ U n n

sm PT sm þ PH nhm þ DNSnm . (18)
u2U s¼1 h¼1
 ARTICLE IN PRESS
5650 E. Muela et al. / Energy Policy 35 (2007) 5643–5655

Reserve:

X
NG X
NH
Dn n
m ð1 þ Rem Þp U n
sm PT maxs þ PH maxh . (19)
s¼1 h¼1

 Hydro scheduling constraints (Water balance equation):

X
NB
V nþ1
h ¼ V nh þ APnh  PH n n
hm tm  SPh . (20)
m¼1
Fig. 3. Representation of possibilistic demand.

 Physical limits:
only to nonfeasible or uneconomic solutions, but also to
V minh pV nh pV maxh . (21)
catastrophic consequences if the future does not come as
In this work volume V, inflow AP and release SP are expected.
expressed in its energy equivalent (MWh). In this case, planners are looking for a planning as
 Thermal unit constraint: robust as possible given the possibilistic descriptions over
noncontrollable variables (fuel cost, emissions, demand,
PT min s U nsm pPT nsm pPT max s U nsm . (22) and inflows). Here, the satisfaction degree of a solution is
evaluated as a necessity measure, which is motivated by the
 Environmental constraints (If exist): attempt to get a robust feasibility of the planning, facing
hazardous events. For example, an uncertain demand
N X
X NB X
NG
ET k ¼ fEknsm ðPT nsm Þg tm pET maxk , (23) (Dem) will be represented by a fuzzy set which is in fact a
n¼1 m¼1 s¼1 possibility distribution over the real set. Fuzzy sets may
where have a variety of shapes but for simplicity we will deal
mainly with data represented by triangular membership
Eknsm ¼ CE ks fa0s EOnsm þ a1s PT nsm g. (24)
functions. A linguistic declaration such as ‘‘demand
values’’ will surely not occur below DDW or above DUP,
and the best estimate (maximum possibility), is to say, DM
4.3. Fuzzy possibilistic multi-criteria formulation will be translated into a fuzzy set such as that in Fig. 3.
DM could come from a forecast, and the extremes DDW
When planning a power system operation, several types and DUP are taken as deviations (forecast error) from DM.
of uncertainties must be dealt with. Some of these A similar procedure can be carried out for the rest of
uncertainties are adequately taken care of by probabilistic possibilistic variables.
methods. But on several occasions such an approach does On the other hand, if we analyze constraint (19), the
not produce adequate results because the existence of power system requires a power reserve level (Res) that is
nonprobabilistic uncertainty stems from ambiguity in defined by previous studies. Unlike the last situation, this is
definition of variables as well as vagueness on subjective a controllable variable whose value can be relaxed for
representation of decision criteria. This sort of uncertainty avoiding infeasibility due to a tight requirement when a
is called possibilistic, which is not related to the event slight relaxation of it would produce a worthwhile solution.
occurrence (frequency), but to the event as such that This is a situation related to flexible constraints, where we
cannot be described without ambiguity. want to take advantage of control over the variable for
If we consider the objective function (17), and con- increasing the feasible space solution. Specifically, the
straints (18) and (20), they are related to forecasts of fuel variable (Res) will be represented by a fuzzy set which is in
cost, demand, and hydraulic inflows, respectively. Prob- fact a membership function associated with the label
ability allows a crisp treatment of them because they are ‘‘satisfactory’’ that evaluates the reserve level qualitatively.
exactly or crisply satisfied all the time, even though this Here, the planner sets a nominal reserve value ReM of
objective and constraints are imprecise due to the varia- maximal satisfaction, and a reserve deviation value dRe
tions of predicted values (rough values). Additionally, if we that defines a point with minimal satisfaction. Fig. 4
examine constraint (24), the emission levels for each power presents the fuzzy set associated with this variable.
unit depend on several different factors, e.g., historical Finally, since the planner is often not able to express his
maintenance, quality of utilized fuel, and real power requirements about decision criteria precisely, he may
production level, etc. Then, it is very hard to build a prefer to express preference degrees about specific attri-
probability distribution on this variable, thus the planner butes, e.g., absolute values of objective functions. Particu-
may only define a set of values (limits) where real value can larly in this work, we use the normalized deviation ‘‘dv’’ of
be located. Therefore, a probabilistic analysis may lead not each objective (which is evaluated with respect to a given
 ARTICLE IN PRESS
E. Muela et al. / Energy Policy 35 (2007) 5643–5655 5651

emission levels ETk, simultaneously, where k ¼

fCO2 ; SO2 ; NOx g. For each objective, an exponential
membership function has been selected as follows:
(
1 f pf  f f
mf ¼ W f dv  with dv ¼ , (25)
e f Xf f
where f represents the objective function value, f* is the
reference value, and Wf its weighting.
 Fuzzy-possibilistic constraints: For the case of the
constraints, we choose fuzzy triangular numbers.
Fig. 4. Representation of fuzzy reserve.
Let

X
NG X
NH
y1nm ¼ PT nsm þ PH nhm .
s¼1 h¼1
X
NB
y2nh ¼ PH nhm  tm ,
m¼1
X
NG X
NH
y3 ¼ PT maxs þ PH maxh .
s¼1 h¼1

Fig. 5. Representation of planner’s preferences. Possibilistic demand 8m and 8n:

8
>
> y1nm þ DNSnm  DM nm
>
>
> 1  n si DM nm py1nm pDM nm þ DnUP m ;
>
< D UPm
pDem ðy1nm Þ ¼ y1nm  DNS nm þ DM nm (26)
>
> 1 si DM nm  DnDW m py1nm pDM nm ;
>
> DnDWm
>
>
: 0 in other case;

reference value) instead of its absolute value of attainment. where DnUPm ¼ DnDWm ¼ dDem DM nm .
As in the previous situation, preference can be represented Possibilistic inflows 8h and 8n:

8
> V nþ1  V nh þ y2nh þ SPnh  APnh
>
> 1  h
si APnh py2nh pAPnh þ APnUP h ;
>
> AP n
< UPh
pAP ðy2nh Þ ¼ V nþ1 þ V n
 y2nh  SPnh þ APnh (27)
>
> 1  h h
si APnh  APnDW h py2nh pAPnh ;
>
> APnDWh
>
:
0 in other case:

by assigning a fuzzy set to each objective function. In this Fuzzy reserve 8m and 8n:
case, a fuzzy set is a membership function which could be
also associated to the label ‘‘satisfactory’’. Such a label
depends on the context, thence the membership function 8
>
> 1 si y3  y1nm XRe M  y1nm ;
may adopt different shapes. >
> n
> 
< 1  Re M y1m  y3
> þ y1nm
Depending on the stringency of objective attainment, it  n ;
n
mRes ðy1m Þ ¼ dRe y1m
is possible to design three basic kinds of membership >
>
> siðRe M  dReÞ y1m py3  y1m pRe M y1m ;
 n n  n
>
functions as shown in Fig. 5 (Tomsovic and Chow, 2000). >
>
: 0 si y3  y1 pðRe M  dReÞ y1 ;
n  n
Here, type L is for the case of low priority, type H is for m m

more a rigorous case, and type N represents an inter- (28)

mediate position between the two previous cases.
In the following, a fuzzy possibilistic model is presented: where APnUP h ¼ APnDW h ¼ dAP APnh .
In this work, dAP and dDem have been selected
 Fuzzy objective functions: The objective defined in as a multiple integer (two) of the standard deviation
this work is: to minimize economic cost CT and of the forecasting value:
 ARTICLE IN PRESS
5652 E. Muela et al. / Energy Policy 35 (2007) 5643–5655

 Fuzzy optimization problem: Using a pessimistic ‘‘ideal’’ or reference point. Theoretically, at this point all
criterion (15), the problem is formulated as follows: criteria reach their optimal value. A good candidate for this
max F ¼ maxfminfmaxð1  pE ðuÞ; mD~ ðxðuÞÞÞgg, (29) point is obtained by solving for each objective a single
x2X u2E criterion problem with nominal requirements for constraints
where
Table 4
pE ðuÞ ¼ minfpnDem m ðuÞ; pnApo m ðuÞ; pCF s ðuÞ; pCE ks ðuÞg, Forecasting demand of system
(30)
tm Load level

Period 1 2 3 4
mD~ ðxðuÞÞ ¼ minfmCT ðxðuÞÞ; mET CO ðxðuÞÞ; mET SO ðxðuÞÞ, 150 (MW) 215 (MW) 178 (MW) 177 (MW)
2 2

mET NO ðxðuÞÞ; mnRe s m ðxðuÞÞg. ð31Þ 1 2.000 1.700 1.300 1.200

x
2 2.010 1.709 1.307 1.206
3 2.020 1.717 1.313 1.212
Here, ‘‘x’’ represents decision variables, i.e., thermal 4 2.030 1.726 1.320 1.218
power generation (PT) and hydraulic power generation 5 2.040 1.734 1.326 1.224
6 2.051 1.743 1.333 1.230
(PH), ‘‘u’’ represents the value that an uncertain variable
7 2.061 1.752 1.339 1.236
can take, ‘‘p’’ is a possibility distribution that is assigned to 8 2.071 1.760 1.346 1.243
each uncertain variable, and ‘‘m’’ is the satisfaction level of 9 2.081 1.769 1.353 1.249
the decision attributes (membership function). 10 2.092 1.778 1.360 1.255
In order to solve this problem a fuzzy compromise 11 2.102 1.787 1.366 1.261
12 2.113 1.796 1.373 1.268
programming has been chosen. This technique allows the
definition of satisfactory solutions which are nearest to an

Table 1 Table 5
Weighting of criteria Hydro inflows

Criterion/attribute Period Hydro unit

WCT W CO2 W SO2 W NOx 1 (MWh) 2 (MWh) 3 (MWh)

Weighting 3 1/3 1/3 1/3 1 313.854,91 15.692,75 22.320,45

2 250.108,99 12.505,45 16.405,20
3 290.981,38 14.549,07 20.253,39
Table 2 4 177.551,14 8.877,56 34.746,21
Hydro unit characteristics 5 145.959,41 7.297,97 22.657,92
6 103.962,10 5.198,10 5.198,10
Hydro PHmax Vmax Vmin Vinitial Vfinal 7 263.514,38 13.175,72 13.175,72
unit (h) (MW) (MWh) (MWh) (MWh) (MWh) 8 446.408,93 22.320,45 8.877,56
9 328.104,00 16.405,20 7.297,97
1 1.000 1000.000 100.000 500.000 500.000 10 405.067,82 20.253,39 5.198,10
2 250 100.000 0 50.000 50.000 11 694.924,27 34.746,21 6.000,00
3 200 100.000 0 80.000 80.000 12 453.158,50 22.657,92 18.000,00

Table 3
Thermal unit characteristics

Id Characteristic Capacity Capacity Heat rate Fuel cost Emission coefficients (kg/GJ)
PTmax (MW) PTmin (MW) (GJ/MW) ($/GJ) CFs
Factor als CE CO2 CE SO2 CE NOx

1 Carbon 2.37% S 300 100 10.53430 1.42 91.13 1.375 0.166

2 Carbon 1.83% S 300 100 10.53430 1.42 91.13 1.214 0.352
3 Carbon 0.82% S 200 50 10.53430 1.71 70 0.543 0.062
4 Fuel oil 1.5% S 150 20 10.18870 2.56 70 0.671 0.259
5 Fuel oil 1.3% S 150 20 10.18870 2.56 70 0.588 0.116
6 Fuel oil 1.0% S 150 20 10.18870 2.65 70 0.447 0.116
7 Fuel oil 0.75% S 100 20 10.18870 2.89 70 0.331 0.079
8 Fuel oil 0.7% S 100 20 10.18870 2.89 70 0.315 0.116
9 Fuel oil 0.3% S 100 20 10.18870 2.98 70 0.133 0.124
10 Gas 100 20 8.54542 2.61 48.46 0 0.111
11 Gas 100 20 8.54542 2.61 48.46 0 0.111
 ARTICLE IN PRESS
E. Muela et al. / Energy Policy 35 (2007) 5643–5655 5653

(maximum possibility or membership). Additionally, in The proposed model is a nonlinear integer mixed
order to determine the importance Wf of each criterion we problem, which can be transformed into an iterative linear
use a mechanism defined in Mikhailov (2003). integer mixed problem by means of approach described in
Sakawa (1993). Then, it is solved by commercial optimiza-
Table 6 tion software.
Deviation level from maximum possibility value

Id Desviation from maximum possibility value 5. Illustrative example

Fuel cost Emission Emission Emission The model has been implemented in GAMS (Brooke et
dCFs (%) dCE CO2 s (%) dCE SO2 s (%) dCE NOx s (%) al., 1998), a high-level language for programming and
1 5 10 15 20 mathematical optimization. It is applied to a test system
2 5 5 10 20 defined in Tables 1–6.
3 5 5 5 5 The results of single criterion optimizations are shown in
4 10 20 25 50 Table 7. From this table, it is possible to obtain reference
5 10 10 15 10
points for multi-criteria optimization.
6 10 5 15 10
7 10 5 5 2 Now, using (29) we define a solution X1 that regards
8 10 20 10 20 nonprobabilistc uncertainty of demand, inflows, fuel cost,
9 10 20 10 30 and emissions. The results are presented in Table 8.
10 15 5 5 5 Considering a pessimistic attitude in the occurrence of
11 15 10 10 10
possibilistic variables, the obtained solution is interpreted
in the following way: the biggest values that possibilistic
Table 7
variables can take are limited by the values that correspond
Single criterion optimization to a membership level of (1–0.498) in their respective
possibility distributions. For such variables status, it is
Attribute results possible to guarantee with a certainty degree N ¼ 0.498
Economic Emission CO2 Emission Emission that the satisfaction level, of the less-satisfied attribute, will
CT ($  106) (t  103/year) SO2 (t  103/ NQx (t  103/ be at least equal to 0.498. It is not possible to guarantee
year) year) such levels of achievement beyond this certainty degree.
Single objective In Table 9, we show a comparison between the last
Economic 143 6305.90 71.16 15.26 solution X1, and a deterministic solution X2 of simple
CO2 190.25 5566.20 42.76 12.28 economic criterion, which is evaluated in a world state ‘‘u’’
SO2 186.14 5584.70 39.53 10.37
defined by a certainty degree of 0.498 on the realization of
NOx 180.90 5591.90 42.54 10.17
possibilistic variables.

Table 8
Fuzzy possibilistic multi-criteria optimization

Attributes

Economic CT Emission CO2 Emission SO2 Emission NOx

($  l06) (t  103/year) (t  103/year) (t  103/year)

Necessity measure (certainty degree N ¼ 0.498) Ideal point 143.00 5566.20 39.53 10.17
Multi-criteria 179.93 6852.40 71.73 15.57
approach
Satisfaction degree m 0.665 0.498 0.665 0.654

Table 9
Comparison fuzzy possibilistic model vs deterministic model

Attributes

Economic CT Emission CO2 Emission SO2 Emission NOx

($  106) (t  103/year) (t  103/year) (t  103/year)

Solution of fuzzy possibilistic muticriteria formulation XI 179.93 6852.40 71.73 15.57

Certainty degree N ¼ 0.498
Solution of single economic criterion X2 168.83 7095.30 80.33 18.56
Percent deviation (%) 6.57 3.42 10.71 16.10
 ARTICLE IN PRESS
5654 E. Muela et al. / Energy Policy 35 (2007) 5643–5655

Fig. 6. Possibility distribution of decision attributes.

Since we consider the coefficients of objective functions majority of possible scenarios, i.e., it is a highly satisfactory
as expressed by means of possibility distribution, then any and robust solution. On the other hand, the solution X2
solution can have a possibility distribution associated to behaves adequately only from the economic perspective,
each one of attribute decision. Thus, in order to achieve a while the achievement of environmental attributes depends
better understanding of the results, we constructed the on the occurrence of favorable events, i.e., low emissions of
possibility distributions of both X1 and X2, which are thermal units, then, it is possible that power system will not
shown in Fig. 6. get the proposed objectives. Thence, if a planner really
If we consider as an objective reaching the absolute desires to achieve environmental objectives it is important
values related to X1, then we can conclude as follows. to regard the nonprobabiltic uncertainty associated to
For the economic objective, according to available power generation emissions.
information, it is possible to affirm with complete certainty
(necessity degree N ¼ 1) that the solution X2 will not
exceed the specified value for this attribute given by the 6. Conclusion
solution X1 (179,93  106$). On the other hand, the
solution X1 will not exceed this value with a certainty In this paper, a fuzzy possibilistic model for power
degree of 0.498. generation planning was presented. In order to consider
For emissions of SO2, it is not possible to assign any adequately environmental criteria a multi-objective opti-
certainty degree N that the solution X2 will not overstep mization has been implemented.
the specified level for this attribute given by the solution X1 The proposed model can be viewed as an integrative
(71.73  103 t), it is only possible to give a plausibility focus, where decision criteria are not considered like
degree P ¼ 0.62 about such a situation. On the contrary, isolated attributes but are analyzed within a process of
we can provide a certainty degree of 0.498 that solution conjoint evaluation. Here, a fuzzy possibilistic model looks
X1will not overstep such a value. for reflecting the imprecision, ambiguity, and vagueness
Similar analyses can be done for emission of CO2 and present in power system planning. Fuzzy sets and
NOx. If we examine each solution as a whole, we can possibility theory are an alternative to do this, because
conclude that the solution X1 has an acceptable level of they allow including this sort of imperfect information into
simultaneous fulfillment of attributes decision in the the problem.
 ARTICLE IN PRESS
E. Muela et al. / Energy Policy 35 (2007) 5643–5655 5655

The model can be extended to estimate the possibility for Inuiguchi, M., Ramik, J., 2000. Possibilistic linear programming: a brief
achieving predetermined crisp constraints on environmen- review of fuzzy mathematical programming and a comparison with
stochastic programming in portfolio selection problem. Fuzzy Sets and
tal attributes. Another extension is to apply the proposed
Systems 111, 3–28.
model for evaluating discrete options in power generation Mikhailov, L., 2003. Deriving priorities from fuzzy pairwise comparison
expansion planning, of course considering its own char- judgements. Fuzzy Sets and Systems 134 (3), 365–385.
acteristics of new problem. Moreno, J.M., Aguarón, J., Escobar, M.T., 2001. Metodologı́a cientı́fica
en valoración y selección ambiental. Pesquisa Operacional Junio 21
(1), 1–16.
References Nelson, R., 2005. Bounded rationality, cognitive maps, and trial and
error learning. LEM Papers Series from Laboratory of Economics
Bellman, R., Zadeh, L.A., 1970. Decision making in fuzzy environment. and Management (LEM), Sant’Anna School of Advanced Studies,
Management Science 17 (4), 141–164. Pisa, Italy. Available on /http://www.lem.sssup.it/WPLem/files/
Brooke, A., Kendrick, D., Meeraus, A., Raman, R., 1998. GAMS: 2005-28.pdfS.
A User’s Guide. GAMS Development Corporation, Washington, DC. Rahman, S., De Castro, A., 1995. Environmental impacts of electricity
Buchanan, J.T., Henig, E.J., Henig, M.I., 1998. Objectivity and subjectivity in generation: a global perspective. IEEE Transactions on Energy
the decision making process. Annals of Operations Research 80, 333–345. Conversion 10 (2), 307–314.
Dubois, D., Prade, H., 1995. Possibility theory as a basis for qualitative Ramı́k, J., 2006. Duality in fuzzy linear programming with possibility and
decision theory. In: Proceedings of the IJCAI-95, Montreal, Quebec, necessity relations. Fuzzy Sets and Systems 157 (10), 1283–1302.
pp. 1924–1930. Sakawa, M., 1993. Fuzzy sets and Interactive Multiobjective Optimiza-
Dubois, D., Prade, H., 1997. The three semantics of fuzzy sets. Fuzzy Sets tion. Plenum Press, New York.
and Systems 90, 141–150. Savage, L.J., 1972. The Foundations of Statistics. Dover, New York.
Dubois, D., Prade, H., (Eds.), 2000. Fundamentals of fuzzy sets. In: The Schweickardt, G., 2004. Segmentos Disputables y Monopolio natural: Los
Handbooks of Fuzzy Sets Series. Kluwer Academic Publishers, Boston. problemas en Distribución Eléctrica en relación a los Sistemas
Dubois, D., Prade, H., 2001. Possibility theory, probability theory and Económicamente Adaptados. Tesis de Maestrı́a. Universidad Nacio-
multiple-valued logics: a clarification. Annals of Mathematics and nal de Comahue, Facultad de Economı́a.
Artificial Intelligence 32, 35–36. Smets, P., 1997. Imperfect information: imprecision—uncertainty. In:
EMEP/CORINAIR Emission Inventory Guidebook, 2005. Technical Motro, A., Smets, P. (Eds.), Uncertainty Management in Information
Report No. 30. European Environmental Agency, Combustion in Systems. From Needs to Solutions. Kluwer Academic Publishers,
energy and transformation industries, available on: /http://reports Dordrecht, pp. 225–254.
.eea.europa.eu/EMEPCORINAIR4/en/page010.htmlS. Soderholm, P., Sundquist, T., 2003. Pricing environmental externalities in
EPA US Office of Air Quality Planning & Standards, AP-42 Emission the power sector: ethical limits and implications for social choice.
Factors Supplement Documents, update 2002, available on: /http:// Ecological Economics 46, 333–350.
www.epa.gov/ttn/chief/ap42/ap42supp.htmlS. Tomsovic, K., Chow, M.Y., 2000. Tutorial on fuzzy logic applications in
Greening, L.A., Bernow, S., 2004. Designed of coordinated energy and power systems. In: Proceedings of the IEEE Power Engineering
environmental policies: use of multi-criteria decision-making. Energy Society Winter Meeting, Singapore, January.
Policy 32, 721–735. Von Neumann, J., Morgenstern, O., 1947. Theory of Games and
Hening, M.I., Buchanan, J.T., 1994. Decision making by multiple criteria: a Economic Behavior. Princeton University Press, Princeton, NJ.
concept of solution. In: Eleventh International Conference on Multiple Zadeh, L.A., 1978. Fuzzy sets as a basis for a theory of possibility. Fuzzy
Criteria Decision Making, Coimbra, Portugal, August 1–6, p. 47. Sets and Systems 1, 3–28.
Hobbs, B.F., Meier, P.M., 1994. Multicriteria methods for resource planning: Zimmermann, H.J., 1996. Fuzzy Set Theory and its Applications, third ed.
an experimental comparison. IEEE Transactions on Power Systems 9 (4). Kluwer Academic Publishers, Dordrecht, pp. 291–310.