Abu Dhabi, UAESunday 25 October 2020


Facebook offers 'holy grail' data set to tackle renewable energy's unsolved problems

The project is a collaboration with Carnegie Mellon but researchers in the UAE said they would be cautious when using its data

Renewables account for one-third of total global power generation but major technical hurdles remain before full adoption is possible. Getty
Renewables account for one-third of total global power generation but major technical hurdles remain before full adoption is possible. Getty

Facebook AI, the social media giant’s artificial intelligence research arm, is introducing a massive data set and computing power to help find efficient methods to store and use renewable energy – joining the likes of AI firm DeepMind, Google and Microsoft in addressing the climate crisis.

In collaboration with Carnegie Mellon University in Pittsburgh, the Open Catalyst Project is a data set open to researchers everywhere to serve as “an accurate and useful foundation for future research” that Facebook claims is the biggest in the world of its kind.

The data set is meant to significantly improve machine learning models, allowing researchers to solve calculations – that would otherwise take them eight hours – in less than one second using trained AI algorithms.

“While most people instinctively think of batteries for energy storage, the cost of outfitting the power grid with enough lithium-ion batteries for days or weeks of reserve power during a cloudy, low-wind stretch is prohibitively expensive, especially on a global scale,” said Larry Zitnick, a researcher at Facebook AI.

Storing wind and solar power when the breeze isn’t blowing or the sun isn’t shining is a persistent challenge that, left unsolved, may prevent the world from shifting fully to renewable energy.

Even with a growing number of supercomputers and AI algorithms, it is still time and cost-intensive to run analyses of each chemical reaction in search of which will provide fuel cells with greater efficiency and capacity to store energy.

While computing power and AI have helped in recent years, a lack of training data was a hindrance to previous research projects.

Solving these problems is increasingly urgent as the world relies more and more on wind and solar to meet rising energy demand.

Renewables account for one-third of total global power generation. This week the International Energy Agency predicted the global response to Covid-19 will "reshape the future of energy" for years to come, with renewables emerging even stronger.

"I see solar becoming the new king of the world's electricity markets," said Fatih Birol, IEA's executive director. "Based on today's policy settings, it is on track to set new records for deployment every year after 2022."

But concerns over Facebook’s track record with users’ data and its primacy in our digital worlds are making some engineering and energy experts uneasy about its foray into addressing renewable energy's biggest challenges.

“Now that there are so many AI models available, the ethical issue is that they are open and can be used by anyone, then how might they be misused,” Daniel Choi, an associate professor in mechanical engineering at Khalifa University, told The National.

His research into battery storage would fit into the Facebook project’s remit but he said that while the initiative looked promising, he would likely continue to rely on algorithms available through Google and a supercomputer available on campus to speed up his work.

"Of course Facebook knows we care about climate change, they know our user behaviour and this [project] could support and enhance reliance on the power grid" to access more of Facebook’s services, Anoop Babu, the head of renewable energy at Ras Al Khaimah Municipality Department, told The National.

While he expressed caution, the initiative looks promising, he said.

Professor Zachary Ulissi, who leads the research team at Carnegie Mellon that worked on the Open Catalyst project, said opening it up to researchers on GitHub and introducing a leaderboard so teams can compete to come up with solutions would ultimately attract a hyper-competitive science community.

The breadth of the data and the processing speeds are unparalleled, Professor Ulissi said. "This is the holy grail."

Updated: October 15, 2020 07:06 PM

Sign up to our daily email