California’s AI-powered wildfire prevention efforts face data challenge

California-based utility companies are increasingly investing in artificial intelligence to mitigate wildfires, but say they struggle to collect enough data to train the AI.

Southern California Edison, San Diego Gas & Electric Co. and PG&E Corp.

say they see promise in AI algorithms that use images captured by drones and other means to detect anomalies in infrastructure that could lead to wildfires. However, they say it will likely take years to collect enough data to deploy the algorithms at scale in their infrastructure, where they would increase ongoing manual inspections.

It’s a common problem for niche industries that don’t have easy access to large datasets, said Gary Marcus, co-author of “Rebooting AI,” adding that effective use cases of AIs today generally work because we “beat them to death” with data.

“I think we’re probably a few years away from, we’ll call it, greater penetration,” said Todd Inlander, senior vice president of information technology and chief information officer of Southern California Edison, adding that the company’s AI algorithms are currently able to identify a number of specific conditions on which they were trained.

Detecting conditions from images is a two-step process, he said, which first involves the algorithm learning to identify what the object is (pole, cross arm, transformer, insulation), then identifies if a certain condition, such as rust or some other form of deterioration, is present.

Each of these scenarios requires hundreds or thousands of images to train the model, he said, adding that there are “hundreds of thousands and millions of images that we need to collect depending on the various circumstances”.

The company is expanding its fleet of drones in order to capture more images. “For me, that’s a really big priority,” he said.

The implications could be huge for these utility companies. Southern California Edison said it has 118,000 miles of distribution and transmission lines. PG&E has 106,681 miles of low voltage distribution lines and 18,466 miles of high voltage transmission lines, many of which pass through high wildfire risk areas, the company said.

Two years ago, PG&E agreed to use cash and stock to fund a $13.5 billion trust to compensate about 70,000 people who lost homes, businesses and family members. in fires started by its equipment.

The California Public Utilities Commission requires state utilities to follow a schedule of manual infrastructure inspections, but AI could supplement that by making inspections more frequent and thorough, the companies said.

San Diego Gas & Electric said it has 75 working models designed to detect specific conditions or damage to company assets or third-party equipment. Gabe Mika, the group’s senior product manager, said each is trained on 100 to 5,000 images. SDG&E leveraged several machine learning and computer vision tools from Amazon Web Services to help create the models, the company said.

San Diego Gas & Electric uses images captured by drones and other means to detect problems, including cracks in infrastructure. Here, an image identifies damage to equipment in a high fire risk neighborhood that has since been repaired.


Photo:

San Diego Gas and Electric

“It’s one of those technologies that has such powerful transformative potential,” said Swami Sivasubramanian, vice president of database, analytics, and machine learning for AWS, adding, ” The ability to solve the problem preemptively is much higher.”

Mr Mika said the models were built primarily to identify conditions from aerial imagery taken by drones, but recently the company has started investing in cameras on fleet vehicles to capture more pictures. He said initial models may require additional training to identify conditions captured by vehicle cameras, which are taken from different angles.

According to Dr. Marcus, current AI algorithms cannot generalize or reason from small amounts of information like humans can. They compensate for this by ingesting massive amounts of data. That said, he added, the more similar the incoming data is to the data the algorithm was originally trained on, the less you need it.

Dr. Marcus is also the founder of software development startup Robust. AI.

PG&E says it has had some success with computer vision models designed to find things like insulation contamination with each model trained over thousands to tens of thousands of images. Small anomalies like cracks are harder for models to recognize because they can be obscured by dirt, guano or water, said Andy Abranches, senior director of wildfire risk management at PG&E.

The goal, he said, is to prioritize construction models that recognize anomalies that can lead to equipment failure and cause ignitions. To do this, each time there is a failure, the company must go through years of prior images of that piece of equipment in order to build a model that recognizes the signs.

“The depth of data you need will probably last for a few more years [collecting] datasets,” said Abranches.

Write to Isabelle Bousquette at Isabelle.Bousquette@wsj.com

Copyright ©2022 Dow Jones & Company, Inc. All rights reserved. 87990cbe856818d5eddac44c7b1cdeb8

Leave a Reply