Thesis
Characterising power grid events using unsupervised learning
Washington State University
Master of Science (MS), Washington State University
05/2016
Handle:
https://hdl.handle.net/2376/102167
Abstract
Recent advances in modern sensing, communication, and information technologies have lead to the Smart Grid as an emerging field with its own challenges. One of the most critical challenges is a need for automatic solutions to data analysis, driven by the continuous, high-frequency data collection on the grid. Clustering is an unsupervised machine learning technique that can be applied to gain greater understanding of data characterization by categorizing data into groups based on their similarities. In this thesis, we present four main contributions in using unsupervised clustering for characterizing line faults and generator fault anomalies in a smart grid. First, we present features that when paired with filtering and the DBSCAN algorithm accomplish a homogeneity score of 0.939. Second, we show that instantaneous clustering under the same features is far superior to that of time series clustering. Third, we explore cluster-specific classifiers and show that with only a fraction of training data, they perform just as well as regular classifiers trained on the whole dataset, yet in some cases, clustering provides most of the classification. Finally, we present a novel set of features that can be used for generator fault discovery. We show that these novel features along with reasonable data filtering can provide a homogeneity score of at least 0.96. These findings shed light on applying unsupervised learning techniques to data collected on the smart grid for characterizing known events and identifying unknown anomalies.
Metrics
9 File views/ downloads
10 Record Views
Details
- Title
- Characterising power grid events using unsupervised learning
- Creators
- Eric Bryan Klinginsmith
- Contributors
- Xinghui Zhao (Chair)Scott Wallace (Committee Member) - Washington State University, Engineering and Computer Science (VANC), School ofSarah Mocas (Committee Member)
- Awarding Institution
- Washington State University
- Academic Unit
- Electrical Engineering and Computer Science, School of
- Theses and Dissertations
- Master of Science (MS), Washington State University
- Publisher
- Washington State University; [Pullman, Washington] :
- Number of pages
- 66
- Identifiers
- 99900525091701842
- Language
- English
- Resource Type
- Thesis