Automating Simulation Output Analysis (AutoSimOA):Selection of Warm-up, Replications and Run-Length

许多组织都使用仿真模型来规划和更好地管理组织系统，例如制造工厂或服务运营。开发和使用模拟模型的过程的一个关键部分是对模型进行实验。为了获得模型性能的准确测量，必须注意从模型中获得足够好的数据。具体的问题是消除初始化偏差、运行模型足够长的时间以及执行足够的复制(使用不同的随机数流运行)。关于这些问题的决策需要许多模拟建模师所不具备的统计技能。因此，许多仿真模型可能使用不当，得出不正确的结论。这项研究旨在开发一种‘分析器’，它将自动分析模拟模型的输出，并就适当的热身期、运行长度和重复次数向模拟建模者提供建议。在研究的第一阶段，将测试分析模拟输出的现有方法，以确定列入分析器的候选方法。然后，将在必要时对候选方法进行调整，使其适合自动化。在研究的最后阶段，将开发和测试一个原型分析器。这些方法和分析器将通过实例数据、使用真实的模拟模型和模拟用户进行测试。

Simulation models are used in many organisations for planning and better managing organisational systems e.g. manufacturing plant or service operations. A key part of the process of developing and using a simulation model is to experiment with the model. In order to obtain accurate measures of a model's performance care must be taken to obtain sufficient good data from the model. Particular issues are removing initialisation bias, running the model for long enough and performing sufficient replications (runs with different streams of random numbers). Decisions regarding these issues require statistical skills which many simulation modellers do not possess. As a result, many simulation models may be used poorly and incorrect conclusions reached. This research aims to develop an 'analyser' that will automatically analyse the output from a simulation model and advise the simulation modeller on an appropriate warm-up period, run-length and number of replications. In the first stage of the research existing methods for analysing simulation output will be tested to identify candidate methods for inclusion in the analyser. Candidate methods will then be adapted where necessary to make them suitable for automation. In the final stage of the research a prototype analyser will be developed and tested. The methods and analyser will be tested on example data, using real simulation models and with simulation users.