rare event analysis, data mining, building automation, clustering analysis, outlier detection ensembles
Today’s building automation systems (BASs) are becoming increasingly complex. A typical BAS usually stores hundreds of sensor measurements and control signals at each time step, which produces massive high dimensional data sets. Traditional analysis methods for BAS data only focus on a small subset of the data, resulting in a huge information loss. Data mining techniques are more effective in knowledge extraction of massive data. This study develops a holistic methodology for analyzing the high dimensional BAS data using advanced data mining techniques, with the aim of identifying rare events in building operation. Rare event analysis helps to identify atypical building operating patterns, detect and diagnose faults, and eventually improve the building operational performance. Two main challenges exist in performing rare event analysis of massive building operational data, i.e. the high data dimensionality and the complexity in building system operation. The former results that the conventional analytics, such as distance-based measures, lose their effectiveness, and the later negatively influences the robustness and reliability of the identification of rare events. The proposed method is specially designed to tackle these challenges by integrating the power of data mining techniques. It consists of four main steps, i.e., data preparation, rare event detection, rare event diagnosis, and post-mining. The methodology is adopted to analyze the BAS data of the tallest building in Hong Kong. Rare events are successfully detected and diagnosed, providing clues to enhance building operational performance.