Reusing simulation experiment specifications to support developing models by successive extension

Peng, Danhua and Warnke, Tom and Haack, Fiete and Uhrmacher, Adelinde M. (2016) Reusing simulation experiment specifications to support developing models by successive extension. Simulation Modelling Practice and Theory, 68, pp. 33-53. ISSN 1569-190X.

Preview

Text (Postprint) Reusing simulation experiment specifications to support developing models by successive extension.pdf - Accepted Version Download (10MB) | Preview

Abstract

Model development is a successive process of validating, revising, and extending models, and requires iterative execution of simulation experiments. While developing a model by extension, executing similar simulation experiments to those performed with the original model reveals important behavioral insights into the extended model. An automatic generation and execution of these simulation experiments can provide valuable support in the process of developing models. A prerequisite is an explicit specification of simulation experiments. Therefore, we annotate models with simulation experiments that are specified in a declarative domain specific language SESSL (Simulation Experiment Specification via a Scala Layer). Based on experiment specifications of the original model, we introduce a mechanism to automatically generate and execute simulation experiments for the extended model with necessary adaptations. Furthermore, as we experiment with stochastic models, we exploit statistical model checking and specify the expected model behavioral properties, against which the simulation results are checked. Thereby, when a model is extended, the original experiment specifications are reused, adapted, and applied to the extended model. Accordingly, the generated simulation trajectories are probed to check whether the expected properties hold with a certain probability or not. Thus, more fast and frequent feedback during model development can be provided to the modeler. Based on a model of membrane related dynamics, we show how the developed approach can be used in successively extending models.