Kennzahlenbasierte Steuerung, Koordination und Aktionsplanung in Multiagentensystemen

Abstract

To be of practical use, the implementation of flexible and modular agent-based cyber-physical systems (CPS) for real-world autonomous control applications in Industry 4.0 oftentimes requires the domain-specific software agents to adhere to the organization's overall qualitative and quantitative business goals, usually expressed in terms of numeric key performance indicators (KPI). In this thesis, a general software framework for multi-agent systems (MAS) and CPS is developed that facilitates the integration and configuration of KPI-related objectives into the agents' individual decision processes. It allows the user of an agent system to define new KPIs and associated multi-criteria goals and supports inter-agent coordination as well as detailed KPI-based action planning, all at runtime of the MAS. The domain-independent components of the proposed KPI framework are implemented as a Java programming library and evaluated in a simulated production planning and control scenario.