137

ISBN 978-9955-28-482-6
M. Grasserbauer, L. Sakalauskas,
E. K. Zavadskas (Eds.): KORSD-2009
Selected papers. Vilnius, 2009, pp. 137142
Institute of Mathematics and Informatics, 2009
Vilnius Gediminas Technical University, 2009

A Multi-Objective Model for Integrated Sustainable
Planning under Uncertainty
Carla Oliveira
1
, Carlos Henggeler Antunes
2

1,

2
R&D Unit INESC Coimbra, Rua Antero de Quental, 199
3030-030 Coimbra, Portugal
2
Dept. Electrical Engineering and Computers, University of Coimbra,
Polo II, 3030-030 Coimbra, Portugal,
E-mail:
1
coliv@inescc.pt;
2
cantunes@inescc.pt
tinct axes of evaluation, generally conflicting and non-commensurable, of the merit of solu-
tions. In particular, multiple objective linear programming (MOLP) models based on the linear
inter/intra industrial linkages of production can be used to study the interactions between the
economy, the energy system and the environment. These models allow assessing the environ-
mental impacts resulting from changes in the level of the economic activities sustained by dis-
tinct policies. However, the uncertainty associated with the model coefficients, namely those
derived from Input-Output (I-O) analysis, may lead to conclusions that are not robust regarding
the changes of the input data. In this context a MOLP model based on I-O analysis with interval
coefficients is proposed, which allows assessing the impacts of distinct policies on the econ-
omy, the energy system and the environment, based on the levels of activity of the economic
sectors.
Keywords: MOLP, Input-Output, Economy-Energy-Environment (E3) interactions, Interval
coefficients.

1 Introduction
I-O analysis is an analytical tool adequate for the evaluation of the inter-relations between
different economic activities being often applied to assess E3 interactions (Hawdon and Pearson,
1995). I-O analysis and linear programming (LP) are closely related. In its simplest form, with
no substitute inputs, I-O analysis may be regarded as a simple particular case of LP (Dorfman et
al., 1958). The use of the I-O methodology in the framework of LP models allows obtaining
value added information, which would not be possible to achieve with the separate use of both
techniques. Inter/intra-sector relations embedded in I-O analysis allow designing the production
possibility frontier. LP models enable choosing the optimum level of activities to optimize a
given objective function, satisfying the productive relations imposed by I-O analysis.
Traditional studies, which use I-O analysis in the framework of LP, generally consider a
single objective function to be maximized or minimized, usually an aggregate economic indica-
tor. However, in most real-world problems multiple, conflicting and incommensurable axes of
evaluation of the merit of potential solutions are inherently at stake. In this context, mathemati-
cal programming models for decision support become more representative of reality if distinct
aspects of evaluation are explicitly considered.
Abstract: Multiple objective programming models allow for the explicit consideration of dis-
5
th
International Vilnius Conference
EURO Mini Conference
Knowledge-Based Technologies and OR Methodologies
for Strategic Decisions of Sustainable Development
(KORSD-2009)
September 30October 3, 2009, Vilnius, LITHUANIA

C. Oliveira, C. H. Antunes

138
Generally, in most real-world situations, the necessary information to specify the exact
model coefficients is not available, because data are scarce, difficult to obtain, uncertain and the
system being modelled may be subject to changes. Therefore, mathematical programming mod-
els for decision support must take explicitly into account, besides multiple and conflicting objec-
tive functions, the treatment of the uncertainty associated with the model coefficients. Interval
programming possesses some interesting characteristics to tackle uncertainty in mathematical
programming models, since it does not require the specification of the probabilistic distributions
(as in stochastic programming) or the membership functions (as in fuzzy programming) of the
model coefficients (Oliveira and Antunes, 2007). Interval programming just assumes that infor-
mation about the range of variation of some (or all) coefficients is available, which allows speci-
fying a mathematical programming model with interval coefficients (unknown but bounded).
In this paper, a MOLP model based on I-O analysis with interval coefficients is proposed
to study E3 interactions. The model is aimed at providing the decision-maker (DM) with infor-
mation about robust solutions, that is, solutions with good performance for different model coef-
ficient settings. In the next sections, a generic description of the model is given. Finally, some
illustrative results obtained using an algorithm developed to provide decision support in MOLP
problems with interval coefficients are analysed.

2 A MOLP model to study E3 interactions
A comprehensive MOLP model to deal with E3 interactions has been proposed elsewhere
(Oliveira and Antunes, 2004). Some significant changes are now incorporated into the model
herein proposed: recent changes in the System of National Accounts consistent with the Euro-
pean System of Accounts (ESA95); 80 (real and artificial) activity branches; volume and price
components of the economy; interval coefficients for the energy use within the I-O coefficients
matrix, as well as in other constraints; emissions not just arising from the combustion processes
but also from other sources; consideration of the acidification equivalent potential (AEP) and the
tropospheric ozone potential (TOP), besides the global warming potential (GWP).
The model includes two types of constraints: coherence constraints and defining con-
straints. The first, based on I-O analysis, guarantee that intermediate consumption and final de-
mand of goods or services of each activity sector shall not exceed the total amount available
from national production and competitive imports of that same good or service:
Ax + a
cptf
(cptf)+ a
csf
(csf)+ a
g
(g) + a
gfcf
(gfcf) + a
sc
(sc) +
a
aldv
(aldv) + a
exp
(expstcif) x + imp
c
, (1)
where A is the interval technical coefficients matrix (product-by-product) and each of its ele-
ments, a
ij
, is the amount of good or service i needed to produce a unit of good or service j; x is
the vector of total output for each activity branch; a
cptf
is the interval vector with the weight of
each good or service aimed at household consumption on the total household consumption; cptf
is the total consumption of the resident and non-resident households on the territory; a
csf
is the
interval vector with the weight of each good or service aimed at the consumption of non-profit
institutions serving households (NPISH) on the total NPISH consumption; csf is the total con-
sumption of the NPISH; a
g
is the interval vector with the weight of each good or service aimed
at public consumption on the total public consumption; g is the total public consumption; a
gfcf
is
the interval vector with the weight of each good or service aimed at gross fixed capital formation
(GFCF) on the total GFCF; gfcf is the total GFCF; a
sc
is the interval vector with the weight of
each good or service aimed at changes in inventories on the total changes in inventories; sc is the
total changes in inventories; a
aldv
is the interval vector with the weight of each good or service
aimed at acquisitions less disposals of valuables (ALDV) on the total ALDV; aldv is the total
A MULTI-OBJECTIVE MODEL FOR INTEGRATED SUSTAINABLE PLANNING UNDER UNCERTAINTY

139
ALDV; a
exp
is the interval vector with the weight of each good or service aimed at exports at
CIF (cost, insurance and freight) prices on the total exports (excluding tourism); expstcif is the
total exports at purchasers CIF prices (excluding tourism); imp
c
is the vector of competitive
imports, where the elements which correspond to non-energetic goods or services have null
value.
The last type of constraints, including economic, social and environmental indicators, is
described in detail in Oliveira and Antunes (2009a).
The economic indicators include several consumption representations: the households
consumption on the territory, the residents (households and NPISH) consumption and, finally,
the tourism imports. In what regards to foreign trade, it is possible to obtain total exports and
imports at constant FOB (free on board) prices, exports at constant purchasers prices; total ex-
ports and imports including tourism and total imports at CIF prices. The employment level for
each activity branch is obtained by dividing the output by the corresponding expected labour
gross productivity for each branch. GDP is computed according to the production approach and
to the expenditure approach.
gdpprod = gav + ts, (2)
where gdpprod is the GDP at constant market prices (production approach), gav is the total gross
added value (GAV) and ts are net taxes on goods and services.
gdp = cpr + g + gfcf + sc + aldv + expfob mfob, (3)
where gdp is the GDP at constant market prices (expenditure approach), cpr is the total con-
sumption of resident households and NPISH, expfob and mfob are total exports and imports at
FOB purchasers prices (including tourism), respectively.
GDP at current prices is obtained from the distinct components of GDP at constant prices
according to the expenditure approach, which are multiplied by the corresponding deflators.
Other economic indicators such as disposable income of households and NPISH, the public debt
and the public administrations global balance are also obtainable.
The environmental sphere of the model considers several kinds of emissions: CO
2
and non-
CO
2
emissions from fuel combustion; fugitive emissions; emissions from industrial processes,
from solvent and other products use, from agriculture activities, from waste management and
from waste water handling. CO
2
emissions from fuel combustion are easily obtained from the I-
O table, where the total fuel use is the total amount of fuel production plus imports. The emis-
sion factors used in the computation of other pollutant emissions are highly dependent on the
technology used. In this case a tier 2 methodology has been used whenever possible, with data
from the National Programme for Emission Ceilings and the National Inventory Report.
The allocation of energetic resources should be made having in mind that the energy sector
is a part of the economic system as a whole and that energy planning requires the incorporation
of economic, social, energy and environmental objectives. In this way, the model herein pro-
posed considers the following objective functions: maximization of GDP, a measure of econ-
omy performance; maximization of the total level of employment, a measure of social well-
being; minimization of GWP, measured through the emission of green house gases (GHG);
minimization of energy imports, a measure of the countrys foreign energy dependency.

3 An interactive method for interval MOLP models
The interactive approach used to obtain compromise solutions to the MOLP model based
on I-O analysis with interval coefficients is based on Oliveira and Antunes (2009b).
C. Oliveira, C. H. Antunes

140
The method starts by formulating two surrogate deterministic problems, by considering the
minimization of the worst possible deviation of the interval objective functions from their
corresponding interval ideal solutions (Inuiguchi and Kume, 1991) and considering satisfaction
levels on the constraints (Urli and Nadeau, 1992). The interval ideal solutions are computed con-
sidering both the extreme versions of the objective functions and the feasible region (Chinneck
and Ramadan, 2000). In order to obtain the first compromise solution, the DM starts by consid-
ering the most constraining feasible region. Let the first compromise solutions be given by x
1U

and x
1U
, which are obtained from the two surrogate deterministic problems obtained, according
to pessimistic or optimistic perspectives, in case the DM wants to minimize the upper bound or
the lower bound of the worst possible deviation, respectively. If the first compromise solution
satisfies the DM, then the algorithm stops; otherwise, the algorithm proceeds. The other com-
promise solutions at iteration m are given by x
m
= x
mU
and/or x
mU
. The interactive phases are
briefly described next (for details see Oliveira and Antunes, 2009b). For each compromise solu-
tion obtained, the following data is shown to the DM: the centre and the width of the interval
objective values for the solutions obtained, the distance between the interval objective values for
the solutions obtained and the interval ideal solutions, and the acceptability index (see Sen-
gupta and Pal, 2000) of the interval objective value for the solution obtained being inferior to the
interval ideal solution; the achievement rate of Z
k
(x
m
) with respect to the interval ideal solution;
and the impact of different values for
i
(constraint satisfaction level) on the compromise solu-
tion. After providing this information to the DM, he/she is asked to reveal his/her preferences
regarding the solution under analysis. If the DM is not yet satisfied with the solution obtained,
then the algorithm proceeds. The DM is then asked to choose the objective function he/she
wishes to improve. If the problem obtained with the additional constraints has an empty feasible
region, then information is provided on the amount he/she should relax the different objective
reference values, in order to restore feasibility (see Chinneck and Dravnieks, 1991). In this phase
the DM may also choose the objective function(s) he/she is willing to relax in order to improve
the other objective function(s) and solve the problem with the additional constraints. The
exhaustiveness of the solution search process depends on the DM, who may decide to end the
procedure when he/she considers to have gathered enough information about the problem.

4. Conclusions
In order to obtain the compromise solutions which best suit the DMs preference structure,
we started our analysis by computing the individual optima of each objective function with the
best and worst case scenarios, respectively. The surrogate model of the original interval MOLP
model is obtained by considering the minimization of the worst possible deviation of each
interval objective function regarding each interval ideal solution.
The solution search process has been driven by considering a hypothetical DM which ex-
presses his/her preferences regarding the information presented. In this case, we have opted to
perform the solution search process by considering distinct decision alternatives according to a
more or less conservative stance of the DM. By analyzing globally the solutions obtained with
both formulations we can conclude that, in general, the more conservative stance allows obtain-
ing better results for the energy and environmental objectives; on the other hand, with the less
conservative stance better outcomes are obtained for the economic and social concerns (see
Fig. 1).
A MULTI-OBJECTIVE MODEL FOR INTEGRATED SUSTAINABLE PLANNING UNDER UNCERTAINTY

141
55000
60000
65000
70000
75000
80000
85000
90000
GWP
Lower bound
Upper bound
Coefficients setting
Target 2010


14000
16000
18000
20000
22000
24000
26000
28000
30000
AEP
Lower bound
Upper bound
Coefficients setting
Target 2010
Gotemburg Protocol


Figure 1. Ranges of variation of GWP and AEP in the solutions analysed

All the solutions obtained indicate the need to reduce the energy intensity of the economy
in order to overcome the deficit regarding the Kyoto Protocol fulfillment. Then again, it might
be said that the improvement of energy efficiency is not enough to attain the necessary emis-
sions reduction in order to achieve the targets imposed on the acidifying substances, being nec-
essary to operate substantial changes namely on the electricity generation sector.

References
Chinneck, J. W.; Dravnieks, E. W. 1991. Locating minimal infeasible constraint sets in linear programs,
ORSA Journal on Computing 3(2): 157168.
Chinneck, J. W.; Ramadan, K. 2000. Linear programming with interval coefficients, Journal of the Opera-
tional Research Society 51(2): 209220.
C. Oliveira, C. H. Antunes

142
Dorfman, R.; Samuelson, P. A. and Solow, R. M. 1958. Linear programming and economic analysis. New
York: Dover Publications, Inc.
Hawdon, D.; Pearson, P. 1995. Input-output simulations of energy, environment, economy interactions in
the UK, Energy Economics 17(1): 7386.
Inuiguchi, M.; Kume, Y. 1991. Goal programming problems with interval coefficients and target intervals,
European Journal of Operational Research 52(3): 345360.
Oliveira, C.; Antunes, C. H. 2004. A multiple objective model to deal with economy-energy-environment
interactions, European Journal of Operational Research 153(2): 370385.
Oliveira, C.; Antunes, C. H. 2007. Multiple objective linear programming models with interval coeffi-
cients an illustrated overview, European Journal of Operational Research 181(3): 14341463.
Oliveira, C.; Antunes, C. H. 2009a. Energy-environment sustainability - a multi-objective approach with
uncertain data, INESC Coimbra Research Report 4/2009. http://www.inescc.pt/documentos/
4_2009.pdf.
Oliveira, C.; Antunes, C. H. 2009b. An interactive method to tackle uncertainty in interval multiple objec-
tive linear programming, Journal of Mathematical Sciences (in print).
Urli, B.; Nadeau, R. 1992. An interactive method to multiobjective linear programming problems with
interval coefficients, INFOR 30(2): 127137.