Cooler transformers could help electric grid
sciencedaily.com
sciencedaily.com
A good while ago, I got hired as part of a team by a local grid operator to analyze all the data they could get from their new transformers. These transformers were deployed to neighborhoods, and could report, like, 200 operating parameters in real-time.
This was pretty cool! So we quickly created an infrastructure to collect all these stats and feed them into a then-newfangled Machine Learning model.
Said model exercised its perceptrons for a while, correlating all possible parameters against the probability of experiencing an outage, and then came up with <drum roll> transformer oil temperature as the only reliable indicator of failure.
Which was sort-of a disappointment to the grid operator, as they already knew that, but it enabled them to not pay the license fee for all the 200 new metrics, and just continue with the one they already had, albeit with a bit more focus.