Energy systems are becoming increasingly complex as developments such as sector coupling and decentral electricity generation increase their interconnectedness. At the same time, energy system models that are implemented to depict and predict energy systems are limited in their complexity due to computational constraints. Thus, a trade-off has to be made between high degrees of detail and model runtimes. As a first step towards efficiently managing the complexity of energy system models, we examine the relationship between the purpose of models and their complexity. Using fact sheets on 145 models, we manually cluster these models based on their purpose and underlying research questions. Further, we conduct mathematical clustering using several clustering methods to investigate the reproducibility of our results. For our study, we define the complexity of a model as the level of detail in which it represents reality. We distinguish the level of detail into the four dimensions of temporal, spatial, mathematical and modeling content complexity. The differences between the clusters found in these dimensions are verified statistically using confidence intervals. 112 out of 145 models can be allocated to one out of four major clusters possessing clearly distinguishable complexity profiles: unit commitment, electrical grids, policy assessment, and future energy systems. In each of these profiles, high complexity in one dimension or subdimension is compensated by low complexities in other dimensions. We therefore conclude that when creating a model, modelers allocate complexity in order of priority on those features and properties that are particularly important for fulfilling the model's purpose. Our results provide a necessary basis for the emerging field of complexity management in energy system modeling and are therefore of high interest for the scientific community and the interpreters of model results such as decision makers from policy and industry.

能源系统正变得越来越复杂，例如部门耦合和分散发电的发展增加了它们的互连性。同时，由于计算限制，被实现为描绘和预测能量系统的能量系统模型的复杂性受到限制。因此，必须在高细节度和模型运行时之间进行权衡。作为有效管理能源系统模型复杂性的第一步，我们研究了模型目的与其复杂性之间的关系。我们使用145个模型的概况介绍，根据它们的目的和潜在的研究问题手动将这些模型聚类。此外，我们使用几种聚类方法进行数学聚类，以研究结果的可重复性。为了我们的学习 我们将模型的复杂性定义为代表现实的详细程度。我们将详细程度分为时间，空间，数学和建模内容复杂度的四个维度。使用置信区间对在这些维度上发现的聚类之间的差异进行统计验证。145个模型中的112个可以分配给四个具有明显可区分复杂性的主要集群中的一个：单位承诺，电网，政策评估和未来能源系统。在这些配置文件的每一个中，一个维度或子维度中的高复杂度都由其他维度中的低复杂度来补偿。因此，我们得出结论，在创建模型时，建模人员按照对那些对于实现模型的目的特别重要的特征和属性的优先级分配复杂性。我们的结果为能源系统建模中新兴的复杂性管理领域提供了必要的基础，因此，对于科学界和模型结果的解释者（例如，来自政策和行业的决策者）具有很高的兴趣。