Space Missions Engineering Laboratory
Decision Making Methodologies
Many situations in real life ask to make rapid and correct choices among several alternatives having in mind different judging criteria. In that framework the decision-maker experience and skill are the key element to obtain, from that decision, the maximum product return, whatever it is. In addition, it might occur to have multiple decision-makers who must intervene either on the same or on different but interdependent areas.
The Decision Making scientific area helps both experts in complex decision-making scenario and neophytes in new situations.
The Decision Making (DM) is an Artificial Intelligence branch focused on algorithms and methods quite effective to either support or simulate the human behaviour in making selections in a universe of alternatives on the basis of an even clashing criteria set; they turn to be powerful whenever both the alternative and the criterion sets are large. Scenarios with multiple decision-makers are supported too by algorithms from this specific area.
Multi-Criteria Decision Making problems fall in one of the followings:
- the choice problem – to determine a subset of the solution domain X to be the best according to a family of criteria;
- the ranking problem – to rank the element of the solution domain X from the best to the worst;
- the sorting problem – to split the solution domain X into subsets according to a given norm.
The Multi-Criteria Decision Making problem solving methods belong to two main classes, the Multi-Objective Decision Making (MODM) and the Multi-Attribute Decision Making (MADM): the taxonomy depends on whether the solution space is both infinite and unknown or finite and perfectly known.
The MODM approach can be seen as a multi-objective optimization and solved by one of the methods exposed at the optimization paragraph. MODM is well suited to solve the aforementioned choice problems.
MADM can be selected to face items both choice and ranking problems, and it is specifically suited whenever a discrete and known set of alternatives define the search space.
The core of all algorithms available to cope with Decision Making problems for multiple alternatives according to different Criteria/Attributes asks to fill the so called decision matrix. The decision matrix basically collects preferences from the users according to each alternative with respect to each criterion: as soon as the matrix is filled, specific algorithms can solve the problem of globally ranking the proposals.
The Decision Matrix elements may come either from users or by a tool dedicated to generate, according to the problem, those information. Two main issues represent the challenge of such research area:
- the Decision Matrix filling, i.e. how to define the score of each alternative according to each judging criterion;
- the Decision Matrix manipulation to get a final ranking .
Typical available techniques to solve those aspects are Multi-Attribute Utility, Outranking, weighted sum, weighted product and Analytic Hierarchical Process.
The decision matrix filling is simple whenever a group of problem experts is available and can be interviewed. The availability of former problem-solving archive can still represent a partial solution to the first issue. The matrix filling becomes harder as soon as no even partial information exist to weight alternatives according to criteria. Some learning algorithms as well as control techniques can be applied to cope with lack of data.
Our group implemented a tool useful to support or – whenever necessary – substitute the decision-makers’ team in multi-criteria multiple choices problems that evolve in time. The tool includes a Fuzzy Logic inference motor trained by dedicated ANN to fill the decision matrix according to the current decision status. A revised Analytic Hierarchical Process manipulates the data to get the final alternatives ranking. Therefore, both un-informed and experts-supported decision-making problems can be faced. The tool interacts with the user via graphical interfaces: decisional steps are completely visible and controllable, if desired.
In the past years, in this research field we successfully solved problems belonging to the following areas:
- the space system design (ESA –CDF support);
- the autonomous planning/scheduling of spacecraft activities (ESA Contract for Proba Satellite).
On going research is focused on:
- linguistic into numerical mapping of preference for the matrix fulfilment;
- control motor synthesis from the current decision process status refinement;
- control motor synthesis from data mining on former similar decisional processes;
- Analytic Hierarchical Process/ Fuzzy inference motor tuning.
Powered by CMSimple