Using Data to Bring Light to the Last Mile

Some of the most exciting energy innovations today, from pendulum power to gravity lights, are being implemented down isolated dirt roads or across remote grasslands to communities living along the “last mile” of electricity delivery.

That mile, though, presents a challenge: how do would-be providers calculate the power needs of first-time consumers often living at or below the poverty level who need electricity to improve their economic status? The stakes are high. Overestimate, and a government or private company wastes precious resources; underestimate and the provider may lose disillusioned customers unwilling to pay for insufficient or unreliable electricity.

Electric machines and lights lift communities out of economic poverty as well as energy poverty by helping to create sustainable jobs, which will be even more important in a post-corona virus world. Indeed, economists predict global economic instability could hit emerging markets especially hard. Modern electricity access is essential for the resilience of poor and low-income communities to weather global crises like pandemics and economic shocks like recessions.

But to encourage providers to invest, innovation needs to extend not just to creating power generators, but to analyzing energy needs. So The Rockefeller Foundation has earmarked $3.8 million over three years to collaborate with energy-sector data scientists to develop a leading-edge algorithm that can predict future electricity demand among low-income customers.

New Times, New Solutions

The traditional approach to electrification has remained unchanged for more than a century; a centralized model in which large-scale electrical generation has been combined with the relatively slow build-out of expensive transmission and distribution lines across nations.

Historically, the rollout of national grids prioritized cities and towns along with commercial zones and corridors, reaching more remote areas only once there was sufficient revenue to subsidize them. For many economies and over decades, this approach worked, enabling most of North and South America, Europe and many parts of Asia to achieve universal electrification.

Countries still lacking universal access are now among the poorest in the world, which means they have less money to spend on electrification when it arrives—at a time when reliable energy has become increasingly central to modern economic activity and a key tool for poverty reduction.

Extending national grids would cost $2000 or more per connection for hundreds of thousands of communities around the world, making microgrids and other mixes of energy solutions a relative bargain. But still, the initial capital investment is high, so providers and others need to have an accurate sense of consumer needs. ESMAP 2019, Mini-Grids for Half a Billion People: Market Outlook and Handbook for Decision Makers.

Once public utilities and private providers realized the costly errors they were making in calculating anticipating electrical usage, they strategized that the best way to find out how much energy is enough was simply to ask. But door-to-door surveys consistently proved unreliable.

“People are not the best predictors of their own consumption,” says Ashvin Dayal, senior vice president of the Foundation’s Power Initiative. “Ask me how much bread I’m going to consume for the next month—I can’t tell you. So how are you going to predict how much you will use of something you’ve never had?”

What providers found, in fact, is that consumers over-estimated by four times the amount of electricity they would use if it were available, Dayal says. As the baseball-playing philosopher, Yogi Berra is famously quoted as saying: “It’s tough to make predictions, especially about the future.”

Making Use of Machine Learning

Enter the Electricity Growth and Use in Developing Economics (e-GUIDE) Initiative, which is aimed at teaching machines to make accurate predictions using historic consumption data and satellite imagery. In essence, known electricity consumption patterns are overlaid with new consumers’ assets, settlement patterns and access to local infrastructure. This information then develops predictions about particular communities’ future use, and those predictions are tested.

e-GUIDE is part of the Foundation’s focus on working with partners in support of UN Sustainable Goal 7, which calls for everyone to have access to affordable, reliable, and modern energy sources by 2030. But efforts to expand the reach of microgrids or other smaller-scale generators cannot occur in a silo. “If you can’t measure it, you can’t manage it,” says UMass Amherst’s researcher Dr. Jay Taneja, a man with a passion for energy data whose father and grandfathers were all electrical engineers.

The capacity to accurately predict energy demand will give providers, governments, and investors the ability to determine the best-fit methods of distributing energy and the confidence to make financially viable and sustainable decisions.