Concurrent Model-Based Anomaly Detection and Diagnosis

Abstract

Early anomaly detection and diagnosis in commercial buildings can significantly decrease energy wastage and occupancy discomfort, and improve system reliability and equipment life.

Recent advances in connected physical systems, information technology, and statistical methodology have made it feasible to provide comprehensive data-enabled anomaly detection techniques that facilitate building operations in large-scale cyber-physical systems such as modern commercial buildings.

This study introduces a novel, real-time anomaly detection framework, called concurrent anomaly detection and diagnosis (C-ADD). C-ADD operates at three levels to detect anomalies, namely: (i) pointwise anomaly detection, (ii) behavior anomaly detection, and (iii) contextual and collective anomaly detection. The detected anomalies are then classified using the building operators’ activity log.

To increase building resilience and prevent costly incidents, C-ADD predicts anomaly behavior by detecting the conditions that usually lead to them and also predicts the potential future impacts on other equipment and features. The developed anomaly detection and diagnosis technology is applied to detect anomalies in HVAC systems of high-rise commercial buildings equipped with cyber-physical systems.

Presented By

Ricardo Cid
Vice President of Engineering, Research and Development
Prescriptive Data

Ricardo Cid is a vice president of Engineering, Research and Development at Prescriptive Data. He specializes in engineering design, big data analytics, data modeling, data-driven models, cryptocurrencies and blockchain, with focus on building energy, building automation system, and cyber-physical system. He has worked extensively on techno-economic modeling and optimization of commercial buildings in New York City.