Anomaly detection method of daily energy consumption patterns for central air conditioning systems

Anomaly detection of operating patterns of complex systems is an important measure to achieve building energy conservation. In this paper, a low-cost anomaly detection method is proposed to identify the anomaly energy consumption patterns of central air conditioning systems (CACS). The complex process of anomaly detection is simplified as a binary classification problem without threshold. And information entropy is used as the characteristic parameter of daily energy consumption patterns (DECP) while traditional characteristic parameters are prone to cause high miss rates or false-positive rates due to the large data fluctuation, numerous influence factors and complex operational parameters of the complex systems. Moreover, three main influence factors are analyzed to divide the complex operating conditions of CACS and the normal DECPs data-set is updated online to improve the accuracy of the abnormal patterns detection. This non-threshold detection method is also verified by site survey which indicates that the detection accuracy is higher than the traditional detection method based on conventional characteristic parameters and regular K-Means clustering method. •Information entropy is used as the characteristic parameter for anomaly detection.•Abnormal DECP of CACS is framed as a binary classification problem without threshold.•Normal DECP data-set is updated continuously to improve the detection accuracy.•Low cost data acquisition system is applied to detect abnormal operating patterns.