Language-independent look-ahead for checking multi-perspective declarative process models

摘要

Declarative process modelling languages focus on describing a process by restrictions over the behaviour, which must be satisfied throughout the whole process execution. Hence, they are well suited for modelling knowledge-intensive processes with many decision points. However, such models can be hard to read and understand, which affect the modelling and maintenance of the process models tremendously as well as their execution. When executing such declarative (multi-perspective) process models, it may happen that the execution of activities or the change of data values may result in the non-executability of crucial activities. Hence, it would be beneficial to know all consequences of decisions to give recommendations to the process participants. A look-ahead attempts to predict the effects of executing an activity towards possible consequences within an a priori defined time window. The prediction is based on the current state of the process execution, the intended next event and the underlying process model. While execution engines for single-perspective imperative process models already implement such functionality, execution approaches, for multi-perspective declarative process models that involve constraints on data and resources, are less mature. In this paper, we introduce a simulation-based look-ahead approach for multi-perspective declarative process models. This approach transforms the problem of a context-aware process simulation into a SAT problem, by translating a declarative multi-perspective process model and the current state of a process execution into a specification of the logic language Alloy. Via a SAT solver, process trajectories are generated that either satisfy or violate this specification. The simulated process trajectories are used to derive consequences and effects of certain decisions at any time of process execution. We evaluate our approach by means of three examples and give some advice for further optimizations.