The availability of data and computing infrastructure in smart grids creates new challenges that invite solutions based on algorithmic techniques. A particular problem of interest in these systems is fault detection. It is difficult to characterize faults because of the number of different ways in which these faults can occur. Moreover, simulating a faulty mode of the system can be expensive. This work addresses fault detection by framing it as the task of detecting anomalies in the sensor data of smart grids. The assumption made is that faults are events of low probability. But no reliance is made on any possible availability of information classifying the sensor data as faulty or normal. Thus, the problem is solved purely based on the structure of the data, by building an anomaly detection system consisting of a multi-task Gaussian process (MTGP) model coupled with a One-Class Support Vector Machine (OC-SVM).
fault detection, anomaly detection, unsupervised learning, multi-task Gaussian process, one-class SVM, smart grids
Level of Degree
Electrical and Computer Engineering
First Committee Member (Chair)
Second Committee Member
Third Committee Member
Karra, Aman. "Anomaly detection in smart grids using multi-task Gaussian processes." (2021). https://digitalrepository.unm.edu/ece_etds/557