Decision SupportA practical weight sensitivity algorithm for goal and multiple objective programming
Highlights
► A weight sensitivity algorithm for goal and multi-objective programming is presented. ► Solutions from portion of weight space of interest to the decision maker are generated. ► The algorithm is demonstrated on two examples from the literature.
Introduction
A multiple objective programme involves the simultaneous optimisation of a number of normally conflicting objective functions. A subset of the field of MOP is that of distance metric based techniques, where the distance between an ideal or desired solution and solutions that are achievable in practice is minimised. Goal programming (Jones and Tamiz, 2010), compromise programming (Yu, 1973), and the reference point method (Wierzbicki, 1982) are examples of distance metric based techniques.
One issue arising in distance metric based techniques is that of choice of model parameters and consequent sensitivity analysis to determine the stability of the solutions produced. A key example of parameters to be set is the preferential weights associated with the objectives or the deviational variables in the achievement function. Kettani et al. (2004) note that the weighting factor can be divided into two components, a preferential component and a normalising component. In this paper the term ‘weight’ shall henceforth be used to refer to the preferential component of the weighting term. The most basic method of weight assignment is to assume equal decision maker preference for all objectives and hence set all weights to an equal value. This may be a valid starting point for an exploration of possibilities in the absence of any firm decision maker preference(s) or a genuine belief that all goals are indeed equally weighted. However, as a single iteration producing a final solution it is a somewhat crude way of modelling decision maker preference(s). Another possibility is to assign weights based on a priori decision maker information. This is a valid approach if such information exists and is quantifiable. Pairwise comparison methods have also been used to determine a set of weights for the goal programming model (Gass, 1986, Kahraman and Buyukozkan, 2008, Li et al., 2009, Wey and Wu, 2007). Penalty structures (Jones and Tamiz, 1995; Chang and Lin, 2009) and methodology from the field of multi criteria decision analysis such as the Promethee method (Martel and Aouni, 1990) can be used to refine the solutions produced.
All of the above methods should be seen as a starting point for a formal or informal process of weight determination. In many cases the means of interaction with the decision maker in order to produce and refine a weighting scheme is not reported in the literature so it is difficult to make assertions as to its nature. It is therefore hypothesised that the methodology of weight elicitation and exploration and reporting of the weight space is in general currently somewhat random and ad hoc in nature. Hence, Jones and Tamiz (2010, pp. 70–72) develop a systematic methodology for weight space analysis. This algorithm takes an initial solution (equal weight or decision maker specified) and searches the entire weighting space to produce a set of solutions for consideration by the decision maker. However, it many cases it is not appropriate to explore the entire weight space as it may contain areas that lead to solutions that are not in accord with the decision maker’s preferences. Generating such solutions involves an unnecessary computational effort and generates a larger than needed number of solutions for consideration by the decision maker. This paper presents a revised weight space exploration algorithm that allows the decision maker to give additional preference information in order to more effectively guide the bounds of the search and produce a set of more solutions that are in accord with their preferences.
This paper investigates the issues involved in designing a methodology for the revised weight space exploration algorithm. The remainder of the paper is divided into five sections. Section 2 defines the three spaces of the decision problem as well as discussing some theory related to the nature of weight space. Section 3 discusses the requirements of the decision maker in order to make their decision. Section 4 develops the revised weight space exploration algorithm. Section 5 illustrates the developed theory and methodology by means of two examples. Finally Section 6 draws conclusions.
Section snippets
Three spaces of the multi-objective problem
The generic multiple objective programming model (Steuer, 1986) is defined as:where x is a set of m decision variables over which the decision maker(s) has control. These can be continuous, discrete, integer, or binary in nature. They define the m-dimensional decision space X. The set F ⊆ X defines the subset of the decision space with feasible or implementable solutions. F is defined by a number of constraints given as functions of x. These may be
Aiding the decision maker(s)
Attention is now turned on how analysis of the parameter space can be used to aid the decision maker(s) in arriving at a solution in line with their preferences. As noted Section 1, sensitivity analysis for goal programming and or justification of weighting schemes is an area for potential improvement in the goal programming modelling process. Another factor to consider is the means of engagement and level of familiarity with decision making techniques of the decision maker. As is the
A revised weight sensitivity algorithm
This section presents a revised version of the weight space exploration algorithm given by Jones and Tamiz (2010, pp. 70–72). The Jones and Tamiz framework allows an exploration of the entire weight space to take place. However, many decisions already have some preference information a priori in addition to their initial weighting estimate. This preference information can restrict the area in weight space that needs to be explored by the algorithm and hence makes exploration of the entire
Illustrative examples
The revised weight sensitivity analysis algorithm is applied to two goal programmes for the purposes of illustration. Example 1 The first goal programme is taken from Jones and Tamiz, 2010, Chapter 3 and relates to a hypothetical manufacturing situation. The goal programme has four goals and the generic form of the weighted version using percentage normalisation has the following algebraic structure.
Conclusions
A revised weight sensitivity algorithm has been developed and illustrated on two examples in this paper. The algorithm has been illustrated on two goal programming models but the methodology also can be applied to any convex distance-metric based multiple objective programme. Although for some types of model this may require adaptation to include a number of other key parameters, such as the maximum permissible deviations in fuzzy goal programming (Yaghoobi et al., 2008); the parameter
References (19)
- et al.
Interval goal programming for s-shaped penalty function
European Journal of Operational Research
(2009) A new goal programming formulation
Omega
(1976)- et al.
Improving the flexibility of goal programming via preference modelling techniques
Omega
(1995) - et al.
A classification model based on goal programming with non-standard preference functions with application to prediction of cinema-going behaviour
European Journal of Operational Research
(2007) - et al.
The double role of the weight factor in the goal programming model
Computers and Operations Research
(2004) - et al.
Combining simulation and goal programming for healthcare planning in a medical assessment unit
European Journal of Operational Research
(2009) Extended lexicographic goal programming: a unifying approach
Omega
(2001)- et al.
Using ANP priorities with goal programming in resource allocation in transportation
Mathematical And Computer Modelling
(2007) A mathematical basis for satisficing decision making
Mathematical Modelling
(1982)
Cited by (50)
Incorporation of poverty principles into goal programming
2024, Omega (United Kingdom)Modelling De novo programming within Simon's satisficing theory: Methods and application in designing an optimal offshore wind farm location system
2024, European Journal of Operational ResearchA two-stage stochastic model for a multi-objective blood platelet supply chain network design problem incorporating frozen platelets
2023, Computers and Industrial EngineeringMulti-criteria mapping and prioritization of Arctic and North Atlantic maritime safety and security needs
2023, European Journal of Operational ResearchMultiple criteria model for allocating new medical robotic devices to treatment centres
2022, European Journal of Operational ResearchCitation Excerpt :Thus, the following observations relating to further research arise: There is scope for formulation of more recent goal programming variants of facility location models that facilitate trade-off and weight sensitivity analysis of the often complex underlying criteria, Jones (2011). There is also scope for investigations of the synergies between the underlying measures of distance utilised in the goal programming and facility location paradigms in order to facilitate more accurate preferential modelling.
An extended goal programming model for the multiobjective integrated lot-sizing and cutting stock problem
2021, European Journal of Operational Research