Knowledge Discovery in Scheduling Systems Using Evolutionary Bilevel Optimization and Visual Analytics

摘要

Scheduling systems are subject to a variety of influencing factors, some of which (e.g. number of vehicles or employees) can be determined by the company itself. Since these framework conditions can have a major impact on the scheduling system's performance, their determination is an important management task. The difficulty of this task increases when conflicting objectives have to be considered, such as costs and performance. Even though evolutionary bilevel optimization can be used to solve this kind of strategic multi-objective problems, it remains hard to gain deeper insights into the scheduling system's behavior by only analyzing the obtained set of Pareto optimal solutions. In this paper, we propose an approach for knowledge discovery in scheduling systems by applying visual analytics on the whole set of evaluated individuals during the evolutionary algorithm. The proposed concept of bilevel innovization is demonstrated by using a nested NSGA-Ⅱ to solve a strategic personnel planning problem and subsequently applying visual analytics to support decision making regarding the number of employees and implemented shifts. The results show that bilevel innovization can be used to get a better understanding of a scheduling system's behavior and to support the decision making process in a strategic planning context.