Developing simulation models - from conceptual to executable model and back - an artifact-based workflow approach

Abstract

Developing a model for simulation is an art and a science. The question is how this process can be suitably supported. Integrating workflows into simulation systems promises user guidance, documentation and reproducibility of this process. However, the highly interactive, partly concurrent, partly optional nature of the modeling process challenges traditional activity-based workflow approaches. This is intensified by its multiple inter-dependencies and the need for an easy extension. We will illuminate this based on a modeling example from cell biology. To support the required flexibility, we propose an artifact-based workflow approach instead. Conceptual model, formal model and different data and information sources are specified declaratively as artifacts. The life cycle of an artifact is defined by stages, guards, milestones, and sentries, following the Guard-Stages-Milestone (GSM) approach. It is shown that the declarative specification provides a better fit for the process of developing a model.

Simulation-based research in management accounting and control: an illustrative overview

“…The final phases of the roadmap proposed by Davis et al (2007) cover experimenting with the model as well as validating the results with empirical data, in order to strengthen the validity of the results. However, it is beyond dispute that developing the simulation model is the most difficult part in this roadmap (Rybacki et al 2014;Robinson 2008a). The process of modeling requires a set of diverse abilities and experience.…”

“…One can argue that building a proper model significantly contributes to "good" simulationbased research. Legitimately, modeling is frequently interpreted as "science" or "art" (Shannon 1975(Shannon , 1998Rybacki et al 2014), which is often just learnt by experience (Robinson 2008a). However, in order to structure this very important step of doing simulation-based research, some scholars elaborated on various life-cycles of simulation model development (cf, e.g., Rybacki et al 2014;Kreutzer 1986;Robinson 2008b).…”

“…Legitimately, modeling is frequently interpreted as "science" or "art" (Shannon 1975(Shannon , 1998Rybacki et al 2014), which is often just learnt by experience (Robinson 2008a). However, in order to structure this very important step of doing simulation-based research, some scholars elaborated on various life-cycles of simulation model development (cf, e.g., Rybacki et al 2014;Kreutzer 1986;Robinson 2008b). What, from a meta perspective, most of these models have in common [and this applies also to the roadmap proposed by Davis et al (2007)] is the following sequence: (i) conceptual and (ii) formal modeling, (iii) building the simulator, (iv) model validation, (v) experimenting and (vi) analyzing.…”

Towards semantic model composition via experiments

“…It serves as an additional software layer between users and simulation systems, facilitates the reuse and execution of simulation experiments, and offers various features, e.g., for experiment design and simulationbased optimization. As already mentioned, a variety of experiments are executed during the development of a simulation model, e.g., sensitivity analysis or parameter estimation [32]. The user might wish to store experiments and results together with the model.…”

“…Experiments and their results play a key role in refining and successively enriching models, which is also reflected in workflows that intertwine phases of experimenting and modeling, e.g., [32]. These experiments shall serve as a starting point to tackle the issue of conceptual interoperability and, more generally, the support of modeling by automatic experimentation.…”