350 European Journal of Operational Research31 (1987) 350-357
North-Holland
MOLP with an interactive assessment
of a piecewise linear utility function
Eric JACQUET-LAGRI~ZE and Rachid MEZIANI
Universitd Paris-Dauphine LAMSADE, Place du Mal de Lattre de Tassigny, 75016 Paris, France
Roman SLOWINSKI
Institute of Control Engineering, Technical University of Poznan, Poland
Abstract: The paper presents a methodology for Multi-Objective Linear Programming (MOLP) problems.
It relies on three steps: (1) Generation of a subset of feasible efficient solutions (from 10 to 50) as
representative as possible of the efficient set. (2) Assessment of an additive utility function using an
interactive method (PREFCALC). (3) Optimization of the additive utility function on the original set of
feasible alternatives. Following this methodology enables the user to find compromise solutions which can
be different from the vertices. It is particularly adapted for large scale linear programs where traditional
multiobjective methods would be too costly to use, since the interactive phase is limited to step 2, using
PREFCALC on a micro-computer. A micro-computer version of the method (PREFCHAT) is available.
Keywords: Multiple criteria linear programming, interactive method, utility function, decision making
1. Outline of the method
This paper presents a new method to support
Decision Makers (DMs) in solving Multi-Objec-
tive Linear Programming (MOLP) problems:
max
g,(x) = E c jxj
j=l
gK(x) = ~., crjxj
j=l
s.t. ~aijx j<~bi, i=l ..... m,
j=l
xj>~0, j=l ..... n.
Why a new method ? Many have been pro-
posed, most of them are interactive, many are
operational under the form of computer programs
(see for instance Slowinski [13], Vincke [15]).
Most of the previous methods, when they are
interactive, use the concept of 'local preferences'.
As a consequence, one has to solve one or more
optimizations of the initial linear program at each
iteration. This can be costly and time consuming,
preventing quick answers in the process, and mak-
ing the method less interactive.
The method proposed here relies on a com-
pletely different principle. It consists of three steps.
(1) Generation of a subset of efficient solutions
Preferences do not always pre-exist. It is there-
fore important to show some feasible alternatives
to the DM and their values on the criteria in order
to enable him to react entering so in a learning
process of his preferences. In order to show him
interesting solutions, we suggest to generate a few
efficient solutions. This technical step is not inter-
active and does not require the presence of the
DM.
Received June 1985
0377-2217/87/$3.50 © 1987, Elsevier SciencePublishers B.V. (North-Holland)