Improving emergency storm planning using machine learning

• Main Authors: Angalakudati, Mallikarjun (Author), Calzada, Jorge (Author), Gonynor, Jonathan (Author), Raad, Nicolas (Author), Schein, Jeremy (Author), Warren, Cheryl (Author), Williams, John (Author), Papush, Anna Michelle (Contributor), Monsch, Matthieu Frederic (Contributor), Farias, Vivek F. (Contributor), Perakis, Georgia (Contributor), Whipple, Sean David (Contributor)
• Other Authors: Massachusetts Institute of Technology. Institute for Data, Systems, and Society (Contributor), Massachusetts Institute of Technology. Operations Research Center (Contributor), Sloan School of Management (Contributor)
• Format: Article
• Language: English
• Published: Institute of Electrical and Electronics Engineers (IEEE), 2018-06-11T19:12:22Z.
• Subjects: Article
• Online Access: Get fulltext

 Extreme weather events pose significant challenges to power utilities as they require very rapid decision making regarding expected storm impact and necessary storm response efforts. In recent years National Grid has responded to a large number of events in its Massachusetts service territory including Tropical Storm Irene and Hurricane Sandy. National Grid, along with MIT, has built a statistical model which predicts localized interruption patterns based on weather forecasts, asset information, historical damage patterns, and geography. National Grid expects that this will become an important tool in its emergency response preparations. This paper will discuss the predictive model which will aid National Grid in its preventative emergency planning efforts. A machine learning predictive algorithm was built by considering physical properties of the network, historical weather data, and environmental information to predict outages, and ultimately damage, based on weather forecasts. The machine learning algorithm will continuously improve in granularity and accuracy through its continued use and the incorporation of additional information. As a data-driven model it provides an invaluable tool for decision making before a storm, which is currently motivated primarily by intuition from industry experience.