PROBABILISTIC REASONING ABOUT VAGUENESS

The article presents a controller for economic processes based on the theory of Bayesian belief nets. The purpose of this controller is to detect divergencies between actual state and target state of the process and to output correcting variables with the aim of reaching the target state. First, we introduce Bayesian belief nets as a calculus for reasoning under uncertainty. Then, we show how to extend this calculus to cope with vagueness as well, thereby allowing to reason about both, vagueness and uncertainty, in a common calculus. Reasoning in this context denotes the exploration of cause-effect relations between vague and uncertain nodes. On this basis the controller itself is developed, which consists of the Bayesian belief net (BBN) and its interfaces to the environment. The interfaces are described by transformation functions from exact (scalar) input quantities into the probabilistic inputs of the net and by the back-transformation of the net's probabilistic output into exact variables. The applicability of the BBN controller is shown using a simplified logistics example, demonstrating the computer-based reasoning possibilities for application experts.