A practical building energy consumption anomaly detection method based on parameter adaptive setting DBSCAN

Abstract

In order to realize the Building Energy Consumption Anomaly Detection (BECAD) for the green building assessment, the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) is adopted for data clustering. To deal with the parameter setting difficulty of the DBSCAN, a practical parameter adaptive setting method is proposed. The presented method determines values of the DBSCAN parameters, MinPts and ε, according to four distribution characteristics (average data distance, data local densities, cosine similarity, and equivalent space radius) of data, and does not need prior knowledge of the datasets. Furthermore, parameter values determined by the proposed method can improve the clustering effect of the DBSCAN on datasets with various data densities. After testing the proposed method with open datasets, DBSCAN with the parameter adaptive setting method is applied to the BECAD. Experiment results show that identified building energy utilization patterns and abnormal buildings are reasonable and the results can offer the management departments a clear understanding of building energy consumption patterns, as well as decision supports to make subsequent improvement measures.