Fully defined physics-based building energy models can accurately represent building systems; however, generating models based on high-level parameters is time consuming and simulation time of complex models can be slow. This article discusses the development of a Metamodelling Framework to create metamodels from a building energy modelling dataset. The framework generates metamodels using either linear regression, random forests, or support vector regressions. A fifth-generation district heating and cooling system analysis use case was used to motivate the development of the framework. The use case required quick and accurate representations of annual building loads reported hourly. Typical annual building modelling approaches can result in a runtime of 10 min. The metamodels runtime was reduced to less than 10 s to load and run an annual simulation with user-defined covariates. The results of the metamodel performance and an abbreviated topology analysis based on the motivating use case will be presented.

完全定义的基于物理的建筑能量模型可以准确地表示建筑系统；但是，基于高级参数生成模型非常耗时，复杂模型的仿真时间可能会很慢。本文讨论了元模型框架的开发，以从建筑能耗模型数据集中创建元模型。该框架使用线性回归，随机森林或支持向量回归生成元模型。使用第五代区域供热和制冷系统分析用例来激励框架的开发。用例需要每小时报告一次的年度建筑负荷的快速，准确的表示。典型的年度建筑建模方法可能需要10分钟的运行时间。元模型的运行时间减少到少于10秒，以加载和运行带有用户定义的协变量的年度模拟。将介绍元模型性能的结果以及基于激励用例的简短拓扑分析。