In Machine Learning

Millennia of sun watching, astronomy, and ever-improving technology allow us to predict quite accurately the day, time, location, and duration of every solar total eclipse. With that prediction, we can plan for the effects of missing rays of sunlight.

irradiance total eclipse bands and pv generation

Source: NERC

US grid operators are now planning for the expected 9,000 MW shortfall in solar output in anticipation of the August 21, 2017 total eclipse. Gas-fired generators are lining up fuel supplies to ensure they can help deliver the missing energy output. Utilities expect huge load swings to occur as solar output drops in near real time with the sun’s occlusion. Preemptive measures such as disconnecting solar generation from the grid and replacing it with other energy sources ahead of the eclipse will allow generators to moderate the effects of the eclipse. After it is disconnected, utilities will carefully plan how to bring solar generation back online. These preventative measures are possible because of the high confidence in predicting the eclipse.

Machine learning can likewise provide high confidence predictions on future asset failures, allowing utilities and renewable energy generators to take appropriate steps to moderate or even avoid the effects of an asset’s failure. Trained models (in significantly shorter time than the millennia it took to develop good eclipse predictions) can deliver predictions at different points in the future for states and conditions both anticipated and unexpected by the machine learning model.

Armed with an asset’s failure prediction, we can deploy machine learning algorithms to optimally (and even automatically) assign O&M teams to take preventative maintenance measures that will address the predicted failure. The algorithms will optimize the maintenance, repair, or replacement tasks for cost, parts availability, skills, time, and even regulatory requirements. With the right models, machine learning algorithms can be as reliable as the sun rising in the east and setting in the west.

Recommended Posts
Showing 2 comments
  • Paul Street

    You article on the USA ellipse is not a great advert for STI and machine learning. As the exact opposite happened.

  • jeff

    Actually, the eclipse was a non-event from a power standpoint because of all the planning utilities and associated entities were able to do in advance. The power did drop and the grid was properly managed. Our point was, and is, that accurate predictions allow businesses to properly plan for and avoid failures, and that using the right tools to make better predictions is the way to go. For more on how California, for example, handled the very-well-predicted eclipse, see this article:

Leave a Comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.

solar array