Adaptive Gridding for Abstraction and Verification of Stochastic Hybrid Systems

Abstract

This work is concerned with the generation of finite abstractions of general Stochastic Hybrid Systems, to be employed in the formal verification of probabilistic properties by means of model checkers. The contribution employs an abstraction procedure based on a partitioning of the state space, and puts forward a novel adaptive gridding algorithm that is expected to conform to the underlying dynamics of the model and thus at least to mitigate the curse of dimensionality related to the partitioning procedure. With focus on the study of probabilistic safety over a finite horizon, the proposed adaptive algorithm is first benchmarked against a uniform gridding approach from the literature, and finally tested on a known applicative case study.