How Silicon Valley Is Helping BP Bring A.I. To The Oil Patch

Christopher Helman , FORBES STAFF Big Oil, Big Energy  

BP’s global operations emit the equivalent of 50.5 million tons of carbon dioxide per year. That’s down from 54.1 million four years ago, and by 2025 the aim is for 3.5 million more tons of “permanent, quantifiable greenhouse gas reductions.” That’s a lot of cuts -- equal to what comes out of the tailpipes of 2.6 million passenger cars. One of the best spots to reduce emissions is right in its oil and gas fields. BP figures that half of its fugitive methane emissions — a fancy way of saying natural gas leaking out of pumps and pipelines — come from its operations in the Lower 48. And a good portion of those happen in mature fields like the one near Wamsutter, in the Great Divide Basin of Wyoming.

Wamsutter has become a laboratory of sorts. BP has drilled thousands of wells there in the past two decades and plans thousands more. Keeping all those wells operating cleanly in the windy high desert is a challenge. Even after years of continuous improvements in automation and preventative maintenance, parts still break, and wells regularly get “loaded up” — where too much naturally occurring water builds up in the pipe such that the gas can no longer bubble out and the well stops flowing. The process of de-watering a well releases into the air a burp of backed-up methane — a fugitive emission.

BP has turned to Silicon Valley for help, and with vendor Kelvin they’ve outfitted hundreds of Wamsutter wells with arrays of cheap sensors that gather and transmit oceans of data into supercomputers where A.I.-driven algorithms crunch endless optimization simulations.

Brian Pugh, BP’s head of Lower 48 operations, is thrilled. “In six months we went from concept to proven. It works on every type of well — plunger, rod pump, all of them.” Early results are more than promising. BP figures the methane vented from those wells is down 74%. Even better, production volumes are up 20% while overall costs have dropped 22%. Wells don't get watered up nearly as often. “In 10 years every new well will have these sensors,” says Pugh. Maybe sooner than that.

“Simplifying complex systems, that’s what I care about,” says Peter Harding, Kelvin CEO and co-founder, who shares a patent on the process and previously worked at Palo Alto-based Technology Crossover Ventures. For his first oilfield installation, in 2015, Harding took pile of Android phones, put them in plastic boxes and ziptied them all around an iconic “nodding donkey” beam pump in an old oilfield in Ponca City, Oklahoma. “What’s in that Android phone is very powerful,” he says.

Sensors like accelerometers and “hyperspectral” methane-detecting cameras can feed data (via cellular network) about every aspect of the field into Kelvin’s A.I. network. They’ve also incorporated decades of maintenance records and even weather data. Paired with LIDAR scans of the field, BP has built a “digital twin” of the field, pinpointing where every bit of kit is located. A production pad gathers the oil and gas volumes from numerous wells. There’s dozens of pads in a field. In a field like Wamsutter, what happens at one pad often impacts others. So to figure out how best to have all these pads and wells working together, Kelvin’s algorithms seek to solve a “pad optimization model.”

The system learns in part by tinkering — opening and closing valves and watching to see what happens to the pressures and flow rates, not just in that well or that pad, but also cross-referencing the effect that a tweak on one side of the field might have on well pressures miles away. “Adjusting, watching, adjusting, watching — oops went too far, let’s dial it back,” explains MJ Maloof, v.p. of sales. Kelvin’s software runs endless simulations and regressions, the better to prioritize preventative maintenance. It’ll only get smarter. “Every time the system learns something it’s applied to everything we’re connected to,” says Maloof. Kelvin also has deployments in the Barnett and Eagle Ford shales of Texas and the Marcellus of Pennsylvania.

Oil companies have been adding sensors and automation gear to their fields for decades. In the past, a big part of optimizing field operations meant relying on the hard-earned knowledge of field engineers or the awareness of engineers watching datafeeds. But humans simply can’t internalize a 4-dimensional view of an entire field like a supercomputer can. “It isn’t a solution that could be determined by workers out in the field turning wrenches,” says Pugh, whose Lower 48 division produces the equivalent of 318,000 barrels per day and plans $250 million in capital investment this year.

Kelvin's biggest backers are Energy Innovation CapitalFS Investors and Evok Innovations. Chief technical officer Adam Guetz has a PhD from Stanford’s Institute for Computational & Mathematical Engineering. To be sure, Kelvin is not the only digital oilfield vendor BP works with, and it is just one of myriad cognitive computing ventures making inroads throughout the energy and industrial sectors. The trend is clear says Rakshita Agrawal of Boston Consulting Group — after a decade of promises cognitive computing is finally transforming oil and gas. “They’ve finally broken through the mistrust that existed between Silicon Valley and Houston.” The oil price downturn helped in that it forced traditional oilfield services companies to cut back. “Startups have stepped into the opportunity,” Agrawal says. Through its “Intelligent Operations” initiative, BP’s Lower 48 operations have also been deploying cutting edge augmented reality-enabled smartglasses like these.

All told, BP has found digital advances can help its people get the same work done in 40% less time — all while aiding the low-carbon transition. This means jobs that are “not fewer, but different,” says Pugh. Someone’s got to service the robots. “Now field techs are getting trained in Linux and Python.”


Senior Editor Chris Helman has been based in Houston, Texas since 2004. Read his magazine profiles of billionaires ClineHammMcClendonPickensPerotRees-Jones. On Twitter @chrishelman.