A framework for knowledge discovery in massive building automation data and its application in building diagnostics

Building Automation System (BAS) plays an important role in building operation nowadays. A huge amount of building operational data is stored in BAS; however, the data can seldom be effectively utilized due to the lack of powerful tools for analyzing the large data. Data mining (DM) is a promising technology for discovering knowledge hidden in large data. This paper presents a generic framework for knowledge discovery in massive BAS data using DM techniques. The framework is specifically designed considering the low quality and complexity of BAS data, the diversity of advanced DM techniques, as well as the integration of knowledge discovered by DM techniques and domain knowledge in the building field. The framework mainly consists of four phases, i.e., data exploration, data partitioning, knowledge discovery, and post-mining. The framework is applied to analyze the BAS data of the tallest building in Hong Kong. The analysis of variance (ANOVA) method is adopted to identify the most significant time variables to the aggregated power consumption. Then the clustering analysis is used to identify the typical operation patterns in terms of power consumption. Eight operation patterns have been identified and therefore the entire BAS data are partitioned into eight subsets. The quantitative association rule mining (QARM) method is adopted for knowledge discovery in each subset considering most of BAS data are numeric type. To enhance the efficiency of the post-mining phase, two indices are proposed for fast and conveniently identifying and utilizing potentially interesting rules discovered by QARM. The knowledge discovered is successfully used for understanding the building operating behaviors, identifying non-typical operating conditions and detecting faulty conditions.