Hydro One reduces power outage duration with AI predictions
By Derek Roles | 3 minute read | January 21, 2020
Frequent severe weather events pose a challenge to the electricity industry. We not only deal with a higher number of outages, but they’re often more widespread than in the past. In addition, rising customer expectations mean that when major weather events do disrupt service, customers expect their power to be restored very quickly.
Reactive storm response isn’t enough
Hydro One’s primary task is to maintain safe, reliable and affordable power for our customers in the province of Ontario. Hydro One has always done a good job at responding to weather-related outages. But we’ve been reactionary. As a company, and as an industry, we need to find better ways to more effectively manage our resources so that we can respond faster and more proactively to severe weather.
Unlike some of our North American counterparts, we don’t have a meteorologist onsite. Our on-shift managers would monitor the weather and communicate with different providers such as the Weather Network and Environment Canada. But, in general, we couldn’t respond proactively because we didn’t know exactly where and how these events might impact our system.
The first 12 to 24 hours during and after a weather event are critical, so we wanted to find a way to get ahead of the forecast, predict the impact, and activate our emergency response protocol. By doing so, we could shorten the duration of the event by restoring power faster, while providing our customers with updated timely information.
Outage Prediction tool cuts power restoration time in half
We turned to the Outage Prediction tool from The Weather Company, an IBM Business. It provides a weather model to help us make better decisions around mobilizing resources ahead of severe weather. The AI tool takes historical data from Hydro One’s outage database and compares it against the weather patterns that caused the interruptions. When storms are predicted, it takes the previous data and creates an outage forecast.
Now, we can start to look out 24 or 48, or up to 72 hours in advance, map that against our emergency response staging and planning and operational levels for storm management and begin to activate those emergency procedures well in advance of any storm. We can also activate our incident command center to be better prepared and save time once the severe weather has passed. Instead of using the first 24 hours to reposition resources, our prepositioned people and equipment can be repairing lines and restoring power.
In 2018, we had a chance to test the Outage Prediction tool when we had four very significant weather events. A couple of them were very similar to events in 2013, and using the tool helped us cut the power restoration time almost in half compared to 2013.
Great analytics rely on good data
From the start, we worked very closely with The Weather Company and IBM to ensure that they had the right data from Hydro One to train the model so that it could provide the best analytics to Hydro One. That was critical. It’s an old saying, but “garbage in, garbage out” is still true. The analysis from the tool is only as good as the information you feed it and only as good as the information that you maintain.
In our business, we have to be able to trust any product that we’ll rely on for decision making, and this is in part achieved by building lasting relationships with our providers. Every time we’ve reached out to IBM, from early development to today, when the product is in service, they’ve been committed to making the product even better.
Watch Derek Roles discuss how Hydro One uses the Outage Prediction tool to prepare for severe weather events: