The Central Air-Conditioning System (CACS) in subtropical region is responsible for more than 50% of total energy consumption in public buildings. Improper operating modes of CACS often lead to abnormalities in DHECM (Diurnal Hourly Energy Consumption Mode), the detection of which is of great significance for energy conservation. However, It is difficult to detect the abnormal modes effectively by conventional feature extraction and single threshold anomaly detection methods due to its complicated operational condition. Two-year hourly energy consumption data of CACS in an office building collected by CACS monitoring and control platform are divided into types of typical working conditions by decision tree and the information entropy value is used as the characteristic parameter of uncertainty for diurnal hourly energy consumption time series to reduce their dimension. Furthermore, a clustering unsupervised algorithm was used to classify normal and abnormal DHECM which solve the problem that the threshold of the abnormal mode is difficult to determine. The abnormal detection results showed the effectiveness of this method in the field of abnormal DHECM detection of public buildings.
Keywords air conditioning system, diurnal hourly energy consumption mode, abnormal detection, information entropy, clustering analysis