Computational intelligence based anomaly detection for Building Energy Management Systems

In the past several decades Building Energy Management Systems (BEMSs) have become vital components of most modern buildings. BEMSs utilize advanced microprocessor technology combined with extensive sensor data collection and communication to minimize energy consumption while maintaining high human comfort levels. When properly tuned and operated, BEMSs can provide significant energy savings. However, the complexity of the acquired sensory data and the overwhelming amount of presented information renders them difficult to adjust or even understand by responsible building managers. This inevitably results in suboptimal BEMS operation and performance. To address this issue, this paper reports on a research effort that utilizes Computational Intelligence techniques to fuse multiple heterogeneous sources of BEMS data and to extract relevant actionable information. This actionable information can then be easily understood and acted upon by responsible building managers. In particular, this paper describes the use of anomaly detection algorithms for improving the understandability of BEMS data and for increasing the state-awareness of building managers. The developed system utilizes modified nearest neighbor clustering algorithm and fuzzy logic rule extraction technique to automatically build a model of normal BEMS operations and detect possible anomalous behavior. In addition, linguistic summaries based on fuzzy set representation of the input values are generated for the detected anomalies which increase the understandability of the presented results.