Towards generating and executing automatically simulation experiments

仿真研究正成为许多应用领域中不可缺少的工具。然而，进行仿真研究不仅需要对要建模和分析的系统有深入的了解，还需要对仿真实验的设计和涉及的方法有详细的了解。“面向模拟实验的自动生成和执行--GRASE”项目旨在通过自动生成和执行实验来支持系统的离散事件随机模拟研究。模拟研究包括模型的迭代改进和不同实验的连续执行，例如灵敏度分析或优化，对于这些实验，同样有不同的方法可用。为了自动生成和执行仿真实验，我们将重点解决以下问题：需要什么样的关于仿真实验的知识、方法、目标和关于当前仿真研究的知识，这些知识如何被表示、使用和组合，仿真实验的图式、关于方法的本体、概念模型以及先前执行的仿真实验和来源在这方面发挥了什么作用？我们将采用两种策略来生成模拟实验。两者都依赖于上述知识来源的有效组合。然而，他们的出发点和方法各不相同。一种策略侧重于在以前已经执行过具有类似目标的实验的基础上，或者在某些实验已经用密切相关的模型进行的基础上，生成和执行特定的模拟实验。因此，来源地将成为这一战略的起点。第二种策略旨在从头开始生成模拟实验，并将依靠推理规则来选择实验类型、适当的方法，并填写相应的实验模板。在这里，概念模型将是至关重要的，因为它提供了关于模拟模型的上下文信息。几个具体的模拟研究将有助于评估所开发的方法、策略及其组合。

Simulation studies are becoming more and more an indispensable tool in many application areas. However, executing simulation studies requires not only in-depth knowledge about the system to be modeled and analyzed, but also detailed knowledge about the design of simulation experiments and the involved methods. The project "towards GeneRating and Executing Automatically Simulation Experiments - GrEASE" aims at supporting systematic discrete-event stochastic simulation studies by automatically generating and executing experiments. Simulation studies involve the iterative refinement of models and the successive execution of diverse experiments, e.g., sensitivity analysis or optimization, for which again different methods are available. To automatically generate and execute simulation experiments, we will focus on the following questions: what kinds of knowledge about simulation experiments, methods, goals, and about the current simulation study are needed, how can such knowledge be represented, used, and combined, and what role can schemas of simulation experiments, ontologies about methods, the conceptual model, as well as previously executed simulation experiments and provenance play in this endeavor? We will pursue two strategies to generate simulation experiments. Both rely on an effective combination of the above knowledge sources. However, their starting points and approaches vary. One strategy focuses on generating and executing a specific simulation experiment on the basis that experiments with similar goals have been executed before, or on the basis that certain experiments have been done with closely related models. Thus, provenance will form the starting point for this strategy. The second strategy aims to generate simulation experiments from scratch and will rely on inference rules to select an experiment type, appropriate methods, and to fill the respective experiment templates. Here, the conceptual model will be crucial as it provides context information about the simulation model. Several specific simulation studies will help evaluating the developed methods, strategies, and their combination.