A Framework for Discovering Discrete Event Models

Abstract

Complex systems are frequently reasoned about by their discrete event models. A discrete event model of a system represents its behavior by a set of rules describing temporal and causal relationships. The process of constructing such models has been essentially an art. In this paper we present a framework for discovering discrete event models. The framework includes the fundamentals of discrete event modeling, specification of a system's observations necessary for model discovery, the form of causal and temporal rules that constitute discrete event models, and a method for discovering a discrete event model from observations. The discovery algorithm partitions observations into subsets and incrementally discovers a model. The observations are partitioned according to the type of temporal relationship to be discovered. The algorithm can discover three types of temporal relationships typified by the connectives whenever, after and unless. It uses the FAHRENHEIT discovery system to discover functional relationships among variables and events. The capability of FAHRENHEIT to find regularities separated by boundaries is important in detecting complex functional relationships that typically occur in discrete event models. We discuss the limitations of our algorithm and point out the richness of data and forms of knowledge in the domain of discrete event modeling to be managed by the future discovery systems.