In Europe, the energy performance of dwellings is measured using theoretical building energy models based on the Energy Performance of Buildings Directive (EPBD), which estimates the energy consumption of dwellings. However, literature shows large performance gaps between the theoretically predicted energy consumption and the actual energy consumption of dwellings. The goal of this paper is to investigate the extent to which empirical models provide more accurate estimations of actual energy consumption when compared to a theoretical building energy model, in order to estimate average actual energy savings of renovations. We used the Dutch non-profit housing stock to demonstrate the results. We examined three empirical models to predict the actual energy consumption of dwellings: a linear regression model, a non-linear regression model, and a machine learning model (GBM). This paper shows that these three models alleviate the performance gap by giving a good prediction of actual energy consumption on sectoral cross-sections. However, these models still have shortcomings when predicting the effects of specific renovation interventions, for example newly introduced heat pumps. The non-linear and machine learning model (GBM) outperform the theoretical model in terms of estimating energy savings through renovation interventions.

在欧洲，住宅的能源性能是使用基于建筑能源性能指令 (EPBD) 的理论建筑能源模型来衡量的，该指令估计住宅的能源消耗。然而，文献显示理论上预测的能源消耗与住宅的实际能源消耗之间存在巨大的性能差距。本文的目的是调查与理论建筑能源模型相比，经验模型在多大程度上提供了对实际能源消耗的更准确估计，以估计翻新的平均实际节能。我们使用荷兰的非营利住房存量来证明结果。我们检查了三个经验模型来预测住宅的实际能源消耗：线性回归模型、非线性回归模型、和机器学习模型（GBM）。本文表明，这三个模型通过对部门横截面的实际能源消耗进行良好预测，从而缩小了性能差距。然而，这些模型在预测特定改造干预措施（例如新引入的热泵）的影响时仍然存在缺陷。非线性和机器学习模型 (GBM) 在通过改造干预估计节能方面优于理论模型。