Domestic hot water heat use prediction modelling is an important instrument for increasing energy efficiency in many buildings. This article addressed hourly domestic hot water heat use prediction, using a Norwegian hotel as a case study. Since the information available for buildings may vary, two widespread situations with different input variables were studied. For the first situation, the prediction is based only on data obtained from historical measured domestic hot water heat use. For the second situation, additional variables that affect domestic hot water heat use were applied. These variables were determined using the Wrapper approach. The Wrapper approach showed that factors related to the guests presence have the most significant influence on the domestic hot water heat use in the hotel. Nevertheless, daily data about the number of guests booked at the hotel did not appear to be informative enough for precise hourly modelling. Therefore, to improve the accuracy of the prediction, it was proposed to use an artificial variable. This artificial variable explained the hourly intensity of the guests domestic hot water use. In order to select the best model for the domestic hot water heat use prediction, ten advanced time series and machine learning techniques were tested based on the criteria of models adequacy. For both considered situations, the Prophet model showed the best results with R2 equal to 0.76 for the first situation, and 0.83 the second situation.

家用热水热利用预测模型是提高许多建筑物的能源效率的重要工具。本文以挪威一家酒店为案例研究了按小时进行的家庭热水供热预测。由于可用于建筑物的信息可能有所不同，因此研究了两种具有不同输入变量的普遍情况。对于第一种情况，预测仅基于从历史测得的生活热水热量使用获得的数据。对于第二种情况，应用了影响生活热水热量使用的其他变量。这些变量是使用包装方法确定的。包装方法表明，与客人的存在有关的因素对酒店的生活热水供热使用影响最大。不过，关于酒店预订客人人数的每日数据似乎不足以提供精确的小时建模信息。因此，为了提高预测的准确性，提出了使用人工变量。这个人工变量解释了客人家庭热水使用的每小时强度。为了为家用热水使用预测选择最佳模型，根据模型适当性的标准，测试了十种高级时间序列和机器学习技术。对于这两种情况，Prophet模型均显示出最佳结果，第一种情况的R2等于0.76，第二种情况的R2等于0.83。这个人工变量解释了客人家庭热水使用的每小时强度。为了为家用热水使用预测选择最佳模型，基于模型适当性的标准，测试了十种高级时间序列和机器学习技术。对于这两种情况，Prophet模型均显示出最佳结果，第一种情况的R2等于0.76，第二种情况的R2等于0.83。这个人工变量解释了客人家庭热水使用的每小时强度。为了为家用热水使用预测选择最佳模型，根据模型适当性的标准，测试了十种高级时间序列和机器学习技术。对于这两种情况，Prophet模型均显示出最佳结果，第一种情况的R2等于0.76，第二种情况的R2等于0.83。