Data- and goal-driven sequential decision making for time-dynamic logistics systems

点击翻译按钮获取中文摘要

The operational control of logistics systems consists of making decisions repetitively over time. Characteristic requirements comprise constantly changing data subject to uncertainty about future developments. Due to the complexity and dynamics of occurring decision situations, it is not sufficient to repeatedly execute methods from the two areas of data analysis and decision making one after another in order to ensure effective and efficient operations of logistics systems. Rather, an overlapping and mutually complementary coordination and integration of data analysis and optimization processes to method pipelines is required to facilitate goal- and data-driven decision making over time according to the specifics of the respective logistics application.Existing research lacks a consistent approach that would adequately take into account the described interactions between data and decision making in a dynamic context, as it is often found in operational logistics. Rather, these issues have been addressed in different communities so far: The analysis of (raw) data is done in the fields of statistics, machine learning and data engineering; time-dynamic optimization problems are treated in the mathematical disciplines of online optimization, multi-stage robust optimization, or multi-stage stochastic programming; logistics issues are mostly investigated by means of a fixed methodology for decision making without special or only insufficient consideration of associated dynamic data processes. Thus, the main objective of the research project is on developing the Dynamic Data-Driven Decisions for Logistics (4D4L)-metamodel that enables an integrated consideration of the entire chain of available data, overall goals, and methods of data analysis and decision making including their interaction effects in the context of time-dynamic logistics applications. The 4D4L-metamodel combines data analysis and decision making methods to method pipelines in a an adaptive and feedback-coupled manner that enables the data- and goal-oriented control of logistics systems. This results in the possibility to provide a structured method repository for time-dynamic logistics systems to support decision processes in a goal-oriented way linked with suitable data preparation methods. The metamodel is general enough in the sense that it allows adaptation to different types of logistics systems. To validate the metamodel, this project focuses on the area of warehouse operations which is considered to be a representative example of time-dynamic logistics systems. In the long run, the research project contributes to the realization of an automated data- and goal-driven composition of method pipelines for decision support.