Generic visual data mining-based framework for revealing abnormal operation patterns in building energy systems.

The abnormal operation patterns in building energy systems can be revealed by analyzing their historical operational data. In practice, the amount of data is so tremendous that manual data analysis is challenging. Visual data mining is a promising solution to this problem. This study proposes a generic visual data mining-based framework for extracting abnormal operation patterns in building energy systems from their historical operational data. The framework consists of three steps. First, a kernel density estimation-based approach is utilized to preprocess the raw data. Then, a decision tree-based approach is adopted to identify the system operation conditions. Finally, a maximal frequent subgraph mining-based approach is developed to reveal the system operation patterns. The framework is applied to analyze the one-year operational data of a chiller plant. This study proves that the framework can appropriately interpret the data mining results, and can make the analysis of the results more convenient. • A visual data mining framework is proposed to reveal abnormal building operation patterns. • Prior knowledge-based graphs are constructed to visualize relations among multiple variables. • A top-down maximal frequent subgraph mining algorithm is developed. • The top-down maximal frequent subgraph mining algorithm outperforms gSpan. • The visual data mining framework makes data mining results interpretable. [ABSTRACT FROM AUTHOR]