A data-driven framework for abnormally high building energy demand detection with weather and block morphology at community scale

Buildings are one of the most important energy use sectors in cities, and forecasting the abnormal increase in building energy demand in certain climatic conditions is necessary to adjust building energy operations and implement energy policy. Accordingly, this research proposes a data-driven abnormally high energy demand detection framework in urban buildings based on their design parameters and local weather data, with the support of machine learning techniques. In this study, 71 public buildings with energy records in Jianhu city, Jiangsu province, China, were selected to abstract urban morphologies at community scale. The weather profile for the city was obtained from year 2015–2018 to create weather characteristics. Three machine learning algorithms—random forest, support vector machine, and artificial neural network—were applied to identify the months of abnormally high electricity consumption in different building types. This framework also explores key variables in the data and provides the basis for a system that prioritizes the acquisition of variables when complete data is unavailable. The results show that, with complete data, the accuracy score of the system in this study can reach 0.854 with the SVM algorithm, and the model returned an accuracy of 0.865 with the RF model after the key variable selection. Based on those results, the framework in this study can generate preemptive warnings for months with an expected abnormally high energy consumption in target buildings as a prerequisite of energy policy.