Eignung schwarmintelligenter Verfahren für die betriebliche Entscheidungsunterstützung (Volume 60) (English shop)

Preview

Table of Contents, Datei (49 KB)
Extract, Datei (110 KB)

This study examines the two swarm intelligence methods Ant Colony Optimization and Particle Swarm Optimization, which in recent years have gained increasing importance, particularly in comparison with established metaheuristics such as Genetic Algorithms. Since both methods have so far been applied only insufficiently in a business management context, the properties of the methods are examined and evaluated on the basis of two exemplary problems from different problem classes.

The continuous problem considered is the model of a stochastic inventory system, for which optimal safety stocks and order quantities for two products have to be determined in order to maximize the total capital employed at the end of the planning horizon.

The second problem examined deals with the design of university timetables for teacher training with as few overlaps as possible, based on a current and practical example from the University of Göttingen.

For both problems, suitable parameter settings are first determined for the ACO as well as for the PSO, with which the methods are then subjected to more detailed investigations. On the basis of evaluation criteria, the quality of the results for both SI methods when applied to the two problems is assessed, in order to derive an overall appraisal of the results. It turns out that both methods not only achieve good results for the problem domain for which they were originally developed, but are also suitable for adaptation to and application in the respective other problem domain.

On the basis of the results obtained, a behaviour of both SI methods can be observed which enables the user, in advance of similar investigations, to select one of the methods on the basis of two properties. Thus, for the objective of solution quality, the ACO is better suited, as it achieves very good solutions in both investigations at comparatively high computational effort. If, on the other hand, quickly reaching a good solution has higher priority, the PSO shows advantages, since within a shorter time and with less effort it provides solutions that can be described as good.

The results of this study show, using two examples of different problems, that SI methods can be successfully employed for finding solutions to and optimizing business management problems and can be used for decision support.

It becomes clear that the two methods examined possess different qualities. When both problem domains are considered together, these differences prove to be consistent for the PSO as well as for the ACO.