AI/Watson

How AI can minimize storm effects on the electrical grid

Share this post:

Climate scientists believe we’re in for weather of increasing intensity—a situation that can severely challenge electric utilities. In order to minimize damage and quickly restore power after a storm, the industry must improve its ability to prepare for weather related outages.

As an example of the impact, in January of 2017, my company, NB Power in New Brunswick, Canada, experienced the most devastating weather event in our 100-year history. A severe ice storm knocked out power to 133,000 of our 400,000 customers. The damage required us to install 600 new utility poles, 150 new transformers and 52 kilometers of new distribution lines at a cost of $30 million.

If we could more accurately predict when and where such events would strike, we could proactively mobilize for faster response. Conversely, if we could forecast that an event would bypass us, we could stand pat and avoid mobilization costs.

Today, we’re testing an AI-powered tool with these capabilities. Working with IBM, we’ve developed OPRO, a system for Outage Prediction and Resource Optimization that uses machine learning to analyze historical storm data from The Weather Company, an IBM business. The resulting mathematical model can predict the effects of coming weather on our grid.

Adding machine learning to the mix

Why do we need this system when we already have tools and experts on staff to manage weather events?

Our legacy tools excel at managing outages after they’ve occurred. OPRO’s machine learning, however, looks three days ahead to forecast the location and intensity of coming weather. It helps to fill in specific details our operations people need to know. How severe will the impact be? Where should we position our resources? Which equipment should we reinforce? Will we need to bring in crews from outside the company?

Such forecasts complement the handful of employees focused on the weather beat. They’ve lived through many years of storms and can assess a range of factors for potential impacts. Still, by capturing The Weather Company’s data in great detail and analyzing it for patterns, OPRO can glean insights that our experts might miss. And the tool empowers a larger group of employees to participate in storm activity decisions.

Although it’s still early in the development cycle, the system has achieved some notable successes.

 Restoring power and reducing costs

Around Christmas 2017, a significant storm appeared to be approaching. Previously, we would have brought in additional resources “just in case.” OPRO, however, consistently predicted minimal impact. Based on that, we chose not to supplement company resources—and we were right. Most of the storm passed us by and we quickly restored the few outages that did occur.

At the other end of the spectrum, in mid-January 2018 the system indicated a coming storm would hit us hard. We heeded the warning by expanding internal resources and contracting for additional crews, positioning them in predicted trouble spots. The storm led to nearly 400 outages, but our preparations helped us restore 90 percent of customers within 24 hours–excellent performance for a major event.

Another incident shows OPRO’s contribution to a complex decision. Just before March Break week in Canada, when many of our crews take vacation, a big storm hit the northeastern U.S. Such storms typically move up the coast into New Brunswick, a concern for us because of vacationing crews and local supplemental resources assisting utilities in the U.S. Normally, we would have brought in crews from further afield at considerable cost. OPRO predicted minimal impact and we trusted it. The storm, which caused much trouble in nearby Nova Scotia, bypassed New Brunswick entirely. The accurate prediction helped to minimize our operating expenses.

These examples show the system’s potential, but we expect much more. To date, we’ve trained it with data from just 32 storms. As we work with IBM to input additional data from The Weather Company, the predictions should become more precise, truly helping NB Power maintain a resilient and cost effective electrical grid.

  

Watch Tony O’Hara of NB Power discuss the changing future of weather data:

 

CTO and VP of Engineering, NB Power

More AI/Watson stories

How to make campus life better with AI technology

In many organizations, business intelligence is the exclusive province of IT specialists and business analysts. However, those of us at Michigan State University (MSU) have a different approach. At MSU’s Division of Residential and Hospitality Services (RHS), we put IBM Watson Analytics into the hands of our operations managers. Its ease of use gives them […]

Continue reading

How to build trust in AI with blockchain technology

Machine learning and AI are all about data. We give AI-powered systems massive amounts of data to ingest and analyze, and then we trust the results. But I want to know how we can ensure the integrity of those systems, from the algorithms to the data sources. Solving the AI data-integrity issue I believe the […]

Continue reading

How AI technology can help make city life easier for all

My company, Shimizu Corp., has long followed the management principles of Shibusawa Eiichi, who held that economic activity and morality are inseparable. This harmony can be seen in the experimental system we developed with IBM Research to help people, including those who are disabled, enjoy barrier-free urban exploration through voice navigation on their iPhones. We […]

Continue reading