A MODULAR AND HIERARCHICAL MODELLING APPROACH FOR STOCHASTIC CONTROL

We propose a novel modelling concept for stochastic control on systems which are hierarchically composed of subsystems with discrete states and stochastic continuous time. The global control structure (called decision tree) is based on the model hierarchy and can be defined by an interleaving of local control with concurrency. For model specification we review the language LARES which comprises an object-oriented modelling design in order to specify modular and hierarchically structured stochastic systems. In order to embed the control structure into the LARES framework we describe the language extension LARES.de. The main focus of the paper is a transformation to a Markov Decision Process induced by an agent-based view on the control structure. This defines the concrete language semantics and makes state-based system optimization accessible.