Study of decentralised decision models in distributed environments
Many of today's complex systems require effective decision making within uncertain distributed environments. The central theme of the thesis considers the systematic analysis for the representation of decision making organisations. The basic concept of stochastic learning automata provides a framework for modelling decision making in complex systems. Models of interactive decision making are discussed, which result from interconnecting decision makers in both synchronous and sequential configurations. The concepts and viewpoints from learning theory and game theory are used to explain the behaviour of these structures. This work is then extended by presenting a quantitative framework based on Petri Net theory. This formalism provides a powerful means for capturing the information flow in the decision-making process and demonstrating the explicit interactions between decision makers. Additionally, it is also used for the description and analysis of systems that axe characterised as being concurrent, asynchronous, distributed, parallel and/ or stochastic activities. The thesis discusses the limitations of each modelling framework. The thesis proposes an extension to the existing methodologies by presenting a new class of Petri Nets. This extension has resulted in a novel structure which has the additional feature of an embedded stochastic learning automata. An application of this approach to a realistic decision problem demonstrates the impact that the use of an artificial intelligence technique embedded within Petri Nets can have on the performance of decision models.