To achieve ambitious climate goals, it is necessary to increase the rate of purposeful retrofit measures in the building sector. As a result, Energy Performance Certificates have been designed as important evaluation and rating criterion to increase the retrofit rate in the EU and Germany. Yet, today’s most frequently used and legally required methods to quantify building energy performance show low prediction accuracy, as recent research reveals. To enhance prediction accuracy, the research community introduced data-driven methods which obtained promising results. However, there are no insights in how far Energy Quantification Methods are particularly suited for energy performance prediction. In this research article the data-driven methods Artificial Neural Network, D-vine copula quantile regression, Extreme Gradient Boosting, Random Forest, and Support Vector Regression are compared with and validated by real-world Energy Performance Certificates of German residential buildings issued by qualified auditors using the engineering method required by law. The results, tested for robustness and systematic bias, show that all data-driven methods exceed the engineering method by almost 50% in terms of prediction accuracy. In contrast to existing literature favoring Artificial Neural Networks and Support Vector Regression, all tested methods show similar prediction accuracy with marginal advantages for Extreme Gradient Boosting and Support Vector Regression in terms of prediction accuracy. Given the higher prediction accuracy of data-driven methods, it seems appropriate to revise the current legislation prescribing engineering methods. In addition, data-driven methods could support different organizations, e.g., asset management, in decision-making in order to reduce financial risk and to cut expenses.

为了实现宏伟的气候目标，有必要提高建筑部门有目的的改造措施的比率。因此，能源绩效证书已被设计为重要的评估和评级标准，以提高欧盟和德国的改造率。然而，正如最近的研究表明的那样，当今最常用和法律上要求的量化建筑能源性能的方法显示出较低的预测准确性。为了提高预测的准确性，研究社区引入了数据驱动的方法，这些方法获得了可喜的结果。但是，对于能源量化方法在多大程度上特别适用于能源性能预测，尚无任何见解。在这篇研究文章中，数据驱动的方法是人工神经网络，D-vine copula分位数回归，极端梯度增强，随机森林，与支持向量回归进行比较，并由合格的审核员使用法律要求的工程方法，通过德国住宅建筑物的真实世界能源性能证书进行验证，并对其进行验证。经过健壮性和系统偏差测试，结果表明，所有数据驱动方法在预测精度方面都比工程方法高出近50％。与支持人工神经网络和支持向量回归的现有文献相比，所有测试方法均显示出相似的预测准确性，但在预测准确性方面，极限梯度增强和支持向量回归具有边际优势。鉴于数据驱动方法的预测准确性更高，因此似乎有必要修改规定工程方法的现行法规。此外，