Anomaly Detection in Complex Power Systems


Project Description

An anomaly is an observation that does not match the patterns inferred from data that is considered normal and cannot be explained by known contextual information. Anomaly detection is vital in various applications of the power system, including detection of an intentional attack, technical fault, and disturbance, etc. Cyber-physical energy systems of the future are increasingly complex systems with depth integration of computation, communications, and control technology. However, the diverse nature of these components, their interlinked topology, and the sheer size of the system lead to an unprecedented level of complexity and quantity of data, which makes anomaly detection a more difficult task.

Anomaly detection methods can be broadly categorized into statistical, proximity-based, and deviation based methods. To tackle the challenges brought by the above-mentioned power system, this project aims at developing appropriate methods and optimizing its parameters to detect the relatively high-level anomaly that conforms to the physical laws but does not match the spatial and temporal patterns inferred from the data that is considered normal in high-dimensional power systems. In addition, methods for locating anomalies in complex power systems with massive data will be studied. Ultimately, a metric to assess the extent of the anomaly will be developed based on considering both the anomaly score and security score and the metric’s calibration will be investigated as well.


Project Team:

C. Wang

Chenguang Wang is a doctoral researcher in the Intelligent Electrical Power Grids (IEPG) group at Delft University of Technology. He obtained his bachelor degree in Electrical Engineering from Wuhan University of Technology in 2014 and master degree from Xi’an Jiaotong University in 2017. He is now working on the topic of machine learning for risk assessment and threat detection. His research interests include anomaly detection, data mining technologies and machine learning algorithms.

For further information, please contact: