Bayesian Optimization-Based Analysis and Planning Approach for Self-Adaptive Cyber-Physical Systems

Abstract

Modern cyber-physical systems (CPSs) operate in dynamic and uncertain environments or operational contexts. Therefore, it is necessary to design systems that self-adapt according to context changes at run-time. However, making a decision on the optimal adaptation in a changing and uncertain context is a complex task. This paper proposes a modular approach for analysis and planning, which generates the optimal system adaptations based on individual sub-decisions. Each sub-decision corresponds to a model @ RT that deals with specific aspects of the context relevant for the concrete adaptation. As a proof-of-concept, we introduce a multi-robot use case to show the possible performance gains of the suggested approach compared with non-adaptive analysis and planning.