The processing of data arising from connected smart grid technology is an important area of research for the next generation power system. The volume of data allows for increased awareness and efficiency of operation but poses challenges for analyzing the data and turning it into meaningful information. This thesis showcases the utility of clustering algorithms applied to three separate smart-grid data sets and analyzes their ability to improve awareness and operational efficiency.

Hierarchical clustering for anomaly detection in phasor measurement unit (PMU) datasets is identified as an appropriate method for fault and anomaly detection. It showed an increase in anomaly detection efficiency according to Dunn Index (DI) and improved computational considerations compared to currently employed techniques such as Density Based Spatial Clustering of Applications with Noise (DBSCAN).

The efficacy of betweenness-centrality (BC) based clustering in a novel clustering scheme for the determination of microgrids from large scale bus systems is demonstrated and compared against a multitude of other graph clustering algorithms. The BC based clustering showed an overall decrease in economic dispatch cost when compared to other methods of graph clustering. Additionally, the utility of BC for identification of critical buses was showcased.

Finally, this work demonstrates the utility of partitional dynamic time warping (DTW) and k-shape clustering methods for classifying power demand profiles of households with and without electric vehicles (EVs). The utility of DTW time-series clustering was compared against other methods of time-series clustering and tested based upon demand forecasting using traditional and deep-learning techniques. Additionally, a novel process for selecting an optimal time-series clustering scheme based upon a scaled sum of cluster validity indices (CVIs) was developed. Forecasting schemes based on DTW and k-shape demand profiles showed an overall increase in forecast accuracy.

In summary, the use of clustering methods for three distinct types of smart grid datasets is demonstrated. The use of clustering algorithms as a means of processing data can lead to overall methods that improve forecasting, economic dispatch, event detection, and overall system operation. Ultimately, the techniques demonstrated in this thesis give analytical insights and foster data-driven management and automation for smart grid power systems of the future.