Formal Behaviour Classification under Uncertainty

Abstract

A study was performed by the author on formal analysis of uncertain non-linear systems in the context of System Dynamics (SD). The objective of this study was to develop a more insightful method to classify model behaviour for exploratory modelling. The long term vision of this study is more efficient, more exhaustive and more insightful model behaviour exploration than the current sampling and clustering approaches. To illustrate the possibilities this, a simple predator-prey model from literature was analysed. Uncertainties were specified on the parameters and the resulting behaviour was represented in phase portraits. Through further analysis of local, linearised behaviour around equilibrium points, classes of behaviour were defined on mathematical properties of the system instead of properties of the output. For the predator-prey model, these behaviour classed resulted in well-defined boundaries in the uncertainty space. The major finding of the study is that formal analysis can analytically split the uncertainty space into sub-spaces that result in different behaviour, thereby offers an alternative to the current behaviour classification methods.