A City Turns to an Algorithm to Prioritize Streetlight Repairs
Isogawyi / Getty Images
Connecting state and local government leaders
Rather than repairing thousands of streetlights in the order outages are reported, San Diego weighs data on residential density, the presence of schools or parks, traffic collisions and other factors in deciding which ones to fix first.
With 5,000 broken streetlights and a repair staff short 10 workers, San Diego developed an algorithm to help prioritize fixes.
“My team looks at ways to improve the workflow in a way that outcomes are better for employees and also shorten repair and wait times for city residents,” said Kirby Brady, the city’s chief innovation officer and director of the Performance and Analytics Department. One problem the team noticed is that most repairs are made based on the order in which issues were reported.
“The problem with that [is]…they can be spread out all across the city, and there was no way to really understand the urgency or the nature of that repair in relationship to the surroundings,” she said, not to mention “the inefficiency of a crew driving all over the city throughout a day to get to repairs.”
The algorithm, which the city has been using for about two months, takes data from about 10 datasets focused on location, residential density, the presence of schools or parks, traffic collision information and streetlight work orders from the city’s Transportation Department. A script combines the data and does the calculations.
Factors such as equity, increased traffic collisions or the presence of a park that needs to be well-lit are weighted to produce a ranking of all 5,000 malfunctioning streetlights. In all, San Diego has 65,000 to 70,000 streetlights.
Another element is clustering, or the proximity of broken lights to one another. “Each of those factors is weighted differently, so obviously it’s more of a priority or less of a priority in the overall methodology,” Kirby said. “We’ll continue to refine it. How you weight things obviously can have an impact on the outcomes.”
Her team works with the city’s geographic information system team to apply the resulting data to a map. “What shows up on the map are these clusters of lights across the city” that need to be fixed, she said.
The team works with San Diego Chief Data Officer Andrell Bower, who codes everything in the Python 2 programming language and posts it to GitHub. The team uses a cloud-based Snowflake database to store data and Amazon Web Services for computing power, Kirby said.
The city saw the streetlight repair problem spike in 2021 in part because of supply chain bottlenecks, but especially because of staffing shortages. Although the budget calls for 18 repair workers, only eight are working right now, mainly because of attrition, and it can take a year to fill some positions, Kirby said. As a result, it takes 10 months on average to repair a streetlight.
It’s too soon to say whether the data-driven approach has increased the speed of repairs, but anecdotal feedback from the repair crews indicates that it’s helping morale because they feel that they are getting more accomplished, Kirby said. “They love this new approach,” she said.
Now, she and her team are looking at ways to make their internal-facing map public to help increase the city government’s transparency and give residents a way to track when repairs are likely in their neighborhood. Kirby said she expects the map to be ready in the next six weeks.
She’s also looking at ways to apply the same algorithmic approach to other maintenance services.
“Potholes is a fun one because it’s very visible. It connects to the most visible piece of infrastructure that we have, which is our roads,” Kirby said. The team is trying to identify streets that repeatedly need repairs to determine whether something longer-lasting than simply filling in the pothole is required.
“Because we are the team that maintains our 311 system, which is called Get It Done, we have the benefit of really having the ability to take that 30,000-foot look across all the services that are provided to understand what the pain points are for residents, where we’re taking longer than we should to complete some work,” Kirby said. “We look at the entire picture and say, ‘You’ve got these problems going on with attrition, you’ve got these other problems we’ve heard about with significant challenges around parts delay [and] supply chain issue,’ and then usually where we dig deep … is around the process improvement.”
Stephanie Kanowitz is a freelance writer based in northern Virginia.
NEXT STORY: How Data-driven Policies Can Help States and Localities