2:00  -  3:00

May 7, 2019

305: Leveraging AMI and The Self Organizing Map for Pattern & Anomaly Detection

Session Category :  Duke Energy Showcase 

Given the large amount of data that power generation, transmission and distribution generate, the need to combine grid data with other sources of information to proactively identify patterns leading to outages and poor power quality is essential to create more business value. The Self Organizing Map (SOM), an unsupervised machine learning technique, has been successful in leveraging AMI voltage data and meter events in order to identify anomalies on the electric distribution grid that range from misconfigurations of electric components to safety-related situations. A key benefit of the SOM is that it retains the topology of the data and reduces dimensionality to a 2-dimensional space, creating spatially organized internal representations of the input space while uncovering relations between variables that are not easily identified.

Results illustrate that the SOM offers great potential in processing large and diverse amounts of data in order to quickly identify anomalies on the grid such as outages while maximizing human and technological efficiency.