The calibration of building energy model is a vital part of the whole modelling process. To improve the efficiency of this work, an automation procedure has recently been introduced to the calibration process, but no generic approach has yet received the consensus of the whole community at present. The main reason is that a purely mathematics-based, automated calibration lacks physical explanation, which means that the calibrated model probably has a large error in certain single physical values despite a good overall agreement with the measurement data.

In this study, the authors design a set of procedures to automatize the calibration process of building energy model based on schedule tuning and signed directed graph (SDG) method, which codifies human experience and logic and incorporates them into the modules of computational calibration to combine the advantages of traditional and automated approach. The specific operations of calibration process are introduced through a case study. In this case, a building energy model with relatively low accuracy is finally well calibrated. The CV(RMSE) (Coefficient of Variation of Root Mean Square Error) of the original model is 42.12% for power consumption and 25.50% for gas consumption; and for the calibrated model, the CV(RMSE) is 2.21% for power consumption and 3.15% for gas consumption. In addition, the same operations are also applied to another case for further verification. In this case, the final CV(RMSE) of power consumption is reduced to 2.19% from 19.25%. This significant result reveals the applicability and effectiveness of the automated process.

建筑能耗模型的校准是整个建模过程的重要组成部分。为了提高这项工作的效率，最近在校准过程中引入了自动化程序，但是目前还没有通用方法得到整个社区的共识。主要原因是纯粹基于数学的自动校准缺乏物理解释，这意味着尽管与测量数据总体上吻合良好，但校准后的模型在某些单个物理值上可能存在较大的误差。

在这项研究中，作者设计了一套程序，以基于进度计划调整和带符号有向图（SDG）方法来自动化建筑能耗模型的校准过程，该过程将人类的经验和逻辑进行了整理，并将它们纳入计算校准的模块中，以进行组合传统和自动化方法的优势。通过案例研究介绍了校准过程的具体操作。在这种情况下，最终会很好地校准具有相对较低精度的建筑能量模型。原始模型的CV（RMSE）（均方根误差变异系数）在功耗方面为42.12％，在燃气方面为25.50％；对于校准模型，CV（RMSE）的功耗为2.21％，燃气的为3.15％。此外，同样的操作也适用于另一种情况，以进行进一步的验证。在这种情况下，最终功耗的CV（RMSE）从19.25％降低到2.19％。这一重要结果揭示了自动化过程的适用性和有效性。