Why We Need AI: Alphabet’s Moonshot To Save The World’s Electrical Grid

Humanity’s largest machine is demanding an overhaul. We built this behemoth haphazardly and yet we rely on it absolutely. Now we must fix it.

The electrical grid is the vascular system of modern society—but it wasn’t originally designed for its current immense scale. Nor was it designed to evolve, as it must, to seamlessly assimilate an inpouring of clean energy sources like wind and solar.

This brewing crisis presents one of the highest-stakes challenges imaginable for infotech as a field: Across trillions of overtaxed, aging components—most of which date back to the 1960s and 1970s—which ones demand maintenance first?

Predictive AI to the rescue—and boy do I mean it this time. Only with machine learning models will we be able to address this century’s greatest global infrastructure crisis.

Google’s parent company Alphabet has established a new paradigm to supercharge grid upkeep globally. Here’s how the paradigm works—with the help of AI—and how it not only scales critical maintenance but also sets a foundation for the grid’s next evolutionary development: full-scale support for clean energy.

Desperate Times Call For… A Moonshot

Grid operators can’t keep up. Most still perform inspections manually—workers approach individual components, clipboard in hand. Staff can only inspect a small fraction of potentially rusty, broken or otherwise failing poles, pins and transformers.

To lead a sweeping high-tech overhaul of the global grid, an organization would need large-scale instrumentation resources, a brand reputation capable of rallying worldwide participation and an aptitude and appetite for attempting moonshots.

Enter the world’s only “moonshot factory,” X, part of Alphabet and therefore a sister of Google. X’s initiative to take on this and related energy challenges is called Tapestry, described as, “Making the world’s electric grid visible so everyone can access clean, reliable energy.”

To put eyes on the problem, X taps Google’s immense image database originally collected for Street View—you know, the panoramic, street-level perspectives you can access in Google Maps.

X is also exploring the advantages of new, customized image-collection efforts. It rejiggers the hardware developed to collect Street View images, mounting it on the vehicles of energy partners to roam their terrains and collect photos of their grid components. In the last couple of years, vehicles outfitted with this hardware have covered several thousand miles in—for example— Michigan, as part of a joint initiative with the state’s largest energy utility, Consumers Energy.

Needles In A Planet-Sized Haystack

Out of the frying pan and into the fire. This scaled-up monitoring provides ideal, centralized visibility—but it’s too much to look at. The problem is we don’t know what the problem is. Among trillions of components, which ones have a defect or impending failure set to cause an outage—or even a dangerous incident like a fire?

The antidote to information overload is predictive AI, the use of machine learning models to triage and prioritize. It’s a rule of thumb for the Information Age: Warehousing data isn’t enough. To leverage any newfound source of data, you’ve got to learn from it. Machine learning has become imperative.

By targeting maintenance right where it’s needed most, predictive AI fortifies the grid. Once trained, ML models predictively score individual cases—incoming field photos that may or may not reveal an impending issue—which are then prioritized for human inspection. Acting on this great number of incoming images is largely a matter of establishing a strategic decision threshold so that the most urgent issues are properly flagged—and yet not too many flags come in, overwhelming staff.

This predictive targeting parallels another X project, Bellwether, which informs first responders where to go after climate catastrophes. Bellwether uses AI to prioritize photos captured from aircraft, whereas Tapestry prioritizes photos captured from land vehicles.

The wins are in. Even after early deployments, Consumers Energy credits Tapestry with reducing massive outages by 11.2 minutes during 2022, according to testimony given to the Michigan Public Service Commission. Each of these minutes are estimated to cost the Michigan economy $3.6M.

Scaling It Up With Global Data Sharing

As with many machine learning initiatives, manually labeling the data presents a bottleneck (although nowhere near the bottleneck of manually surveilling equipment out on the field with a clipboard). X must manually label thousands of images that reveal defects and it partners with the experts, system operators, to do so. ML needs these labeled examples—along with non-defective examples as well. Crunching the numbers, it learns how to differentiate between positive and negative cases.

As Tapestry gains traction globally, it’s enabling fruitful data-sharing cooperation between its customers. For example, Vector, which provides power to most of New Zealand’s population, is benefiting from the work done with Michigan’s Consumers Energy. After all, the two don’t compete for business.

“This global cooperation works because some patterns are universal,” Tapestry lead Page Crahan told me. Her team trained a model for Michigan and then tried it out on images from Auckland, New Zealand, where grid infrastructure looks entirely different, e.g., transformers are located on the ground rather than on a pole.

Amazingly, the model performed fairly well. “The Michigan model identifies 84% of faults on poles and cross-arms in New Zealand,” Crahan said. “To take advantage of this international crossover, we’re now developing a practice for tuning and repurposing models to transfer well across geographies. This scales our efforts by greatly reducing the investment needed to cover grids globally.”

Similar arrangements have emerged in other industries. For example, FICO’s payment card fraud detection system, called Falcon, greatly benefits from shared data across a consortium of over 9,000 banks, all cooperating for the common good. Likewise, “Flood behavior on the East Coast of the U.S. can be used to forecast floods on the West Coast,” the lead for Bellwether told me.

The Tapestry project began only a few years ago, but this singular initiative is positioned to weave a new fabric of intercontinental energy management. Several other grid operators also have signed on as customers and more are in the pipeline.

But wait, there’s more! All this grid monitoring doesn’t only serve grid maintenance—it also builds the foundational visibility needed for grid advancement and expansion. This addresses much longer-term, larger-scale concerns. Keeping the grid working is critical, but radically evolving it for a new future of energy is an even greater challenge. In my next article, I will continue with that part of the story.