Power-outage prediction

Using data from Hurricane Katrina and four other destructive storms, researchers from Johns Hopkins and Texas A&M universities have created a computer model that can estimate how many power outages will occur across a region as a hurricane is approaching.

The study was a collaborative effort involving Seth Guikema, an assistant professor of geography and environmental engineering at Johns Hopkins and formerly of Texas A&M; Steven Quiring, an assistant professor of geography at Texas A&M; and Seung-Ryong Han, who was Guikema’s doctoral student at Texas A&M and is now based at Korea University.

Their work, which was funded by a Gulf Coast utility company that wishes to remain anonymous, has been published in the current issue of the journal Risk Analysis.

When a hurricane is approaching, an electric power provider must decide how many repair crews to request from other utilities, a decision that may cost the provider millions of dollars. The utility also has to decide where to locate these crews within its service areas to enable fast and efficient restoration of service after the hurricane ends.

Having accurate estimates, prior to the storm’s arrival, of how many outages will exist and where they will occur, will allow utilities to better plan their crew requests and crew locations.

’If the power company overestimates [the damage caused], it has spent a lot of unnecessary money,’ said Quiring. ’If it underestimates, the time needed to restore power can take several extra days or longer, which is unacceptable to them and the people they serve. So these companies need the best estimates possible.’

In developing their computer model, the researchers looked at damage data from five hurricanes: Dennis (1995), Danny (1997), Georges (1998), Ivan (2004) and Katrina (2005). In the areas studied, Ivan created 13,500 power outages; Katrina, more than 10,000; Dennis, about 4,800; Georges, 1,075; and Danny, 620.

For the worst of the storms, some customers were without power for up to 11 days. The research team collected information about the locations of outages caused by these hurricanes, with an outage defined as permanent loss of power to a set of customers.

The researchers also included information about the power system in each area (poles, transformers), hurricane wind speeds, wetness of the soil, long-term average precipitation, the land use, local topography and other related factors. This data was then used to train and validate a statistical regression model called a Generalized Additive Model, a particular form of model that can account for nonlinear relationships between the variables.