Clustering analytics for streaming smart grid datasets

This paper presents an application of clustering algorithms for streaming smart grid datasets. The authors use a framework that includes a combination of an openPDC and statistical software (R) platform to acquire Synchrophasor data to carry out the clustering process. The clustering operations on streaming synchrophasor data enables operators to detect any anomalies or outages for better decision making. This will reduce the outage and failure rates, and enhance situational awareness of the grid. We show the application of hierarchical clustering to organize the sensor data into clusters. Additionally, several distance metrics (cluster linkage methods) are tested for enhanced hierarchical cluster separation. Experimental results on parameters such as frequency, voltage, and current demonstrate the novelty and effectiveness of our application of hierarchical clustering. The results indicate that the hierarchical clustering with single linkage distance metric is a good choice for sudden surge or sag values. On the other hand, the average distance metric is less sensitive to outliers and can detect small deviations in parameters.