Abstract-We consider a class of hybrid systems that involve random phenomena, in addition to discrete and continuous behaviour. Examples of such systems include wireless sensing and control applications. We propose and compare two abstraction techniques for this class of models, which yield lower and upper bounds on the optimal probability of reaching a particular class of states. We also demonstrate the applicability of these abstraction techniques to the computation of long-run average reward properties and the synthesis of controllers. The first of the two abstractions yields more precise information, while the second is easier to construct. For the latter, we demonstrate how existing solvers for hybrid systems can be leveraged to perform the computation.
I. INTRODUCTIONFormal analysis of modern applications involves many characteristics, including real-time, stochastic and hybrid dynamics. Often, probabilistic behaviour is abstracted away during the verification of such systems, due to the additional dimension of complexity. This level of abstraction restricts the analysis to qualitative properties. For systems such as wireless sensing and control applications, however, quantitative and performance properties are desired, thus motivating the study of probabilistic hybrid systems.We consider a class of hybrid systems that involve random phenomena, in addition to discrete and continuous behaviour. We model these systems using probabilistic hybrid automata (PHAs), which extend standard hybrid automata with discrete probabilistic choices. In this paper, we tackle the problem of verifying two types of quantitative properties of PHAs: the minimum/maximum probability of reaching a target (e.g. "the maximum probability of the boiler's temperature exceeding its safe limit"); and the minimum/maximum long-run average reward (e.g. "the minimum average power consumption"). We also consider the problem of synthesising controllers for PHAs to achieve such optimum values.The infinite-state nature of hybrid automata necessitates the use of abstraction for their analysis. In [1], an abstraction technique for PHAs was proposed that bounds the maximum probability of reaching a target, by exploiting the construction of finite abstractions from the non-probabilistic setting. The main drawback of this approach is the lack of knowledge about how far away the computed upper bound is from the real value. In this paper, we propose and compare two types of abstraction for PHAs that allow us to give both lower and upper bounds for such quantitative properties.Our approach is based on the use of n-player stochastic games, finite-state automata incorporating decisions made