Renewable energy has an issue with intermittency: the sun does not generate electricity at night, and winds might abruptly quit. By storing energy when the wind and sun are strong, better battery storage is seen as a key to alleviating the intermittency issue. Current storage technologies, such as lithium-ion batteries and pumped hydro, are, however, costly and difficult to scale. What if excess renewable energy could instead be stored as computation?
Sustainable energy sources might rescue the planet from catastrophic climate change while also lowering electricity expenses. However, renewable energy has an intermittency problem: the sun does not generate electricity at night, and winds might abruptly quit.
Furthermore, electricity networks must balance supply and demand to avoid surges and blackouts. As a consequence, during periods of surplus production, renewable energy is discarded, while power stations burn fossil fuels to satisfy system gaps.
According to Barath Raghavan, an associate professor of computer science at the USC Viterbi School of Engineering, “the amount of renewable electricity squandered in California each year will be similar to the amount of power L.A. needs each year in five years.”
Better battery storage is seen as a key to fixing the intermittency issue by storing energy when the wind and sun are strong, which is a holy grail for scientists all over the globe. Current storage technologies, such as lithium-ion batteries and pumped hydro, are, however, costly and difficult to scale.
What if excess renewable energy could instead be stored as computation? That’s the idea behind “information batteries,” a novel technology devised by Raghavan and Jennifer Switzer, a UC San Diego Ph.D. student, and published in the ACM Energy Informatics Review recently.
Calculations that can be predicted
The underlying concept of information batteries is straightforward: When renewable energy is abundant, it is utilized to execute calculations in huge, energy-intensive data centers on a speculative basis. According to the Office of Energy Efficiency and Renewable Energy, data centers utilize 10 to 50 times the energy of a normal commercial structure, from Google and Facebook to Hollywood movie production. When green energy is in short supply, the calculated results may be utilized later.
“We realized that if we can forecast future calculations, we can execute those computations now, while there is energy available, and store the results, which now contain embodied energy,” Raghavan, whose research focuses on systems and sustainability, said.
Every day, for example, YouTube data centers transcode around 700,000 hours of video to various resolutions. Many of these calculations are foreseeable and can be carried out when there is a surplus of renewable energy. The data is then saved on servers for later use when there is less renewable energy available on the grid, thereby shifting power use from one period to the next.
So, how does this operate in the same way as a battery works? Batteries, in the scientific sense, are potential energy reservoirs that may be used to do beneficial labor, whether electrical or otherwise, according to Raghavan. The majority of energy stored in batteries is converted from one sort of potential energy to another, such as electrical to gravitational. Because electrical energy is converted into “informational potential energy,” information offers energy in the same manner that a battery does.
The system is not just versatile, but it also takes use of job predictability: calculations conducted in advance do not have to match perfectly with computations finished later.
“We allow pre-computing multiple fragments of computation and then picking and choosing tiny portions of computation done before, like puzzle pieces, and putting them together to swiftly compute a whole new computational job,” Raghavan said.
According to Raghavan, the information battery technology is more efficient than lithium-ion batteries for certain workloads. The particular efficiency is determined by a number of parameters, including the sorts of computations performed and the power predictability. However, unlike lithium-ion batteries, data storage is both cost-effective and energy-efficient. This might aid in the reduction of dependency on fossil fuels, which account for three-quarters of global greenhouse gas emissions.
A potentially viable future option
The problem, according to the researchers, is identifying what calculations to run, where and when they should be performed, and how these computations should be done to effectively obtain the findings afterwards.
Raghavan and Switzer present a design and proof of concept implementation of the zero-carbon system, which includes recurrent neural networks for predicting future renewable energy availability and upcoming tasks in data centers, in their paper “Information Batteries: Storing Opportunity Power with Speculative Execution.”
It also comes with a cache for storing functions and a customized compiler for automatically modifying code to save and retrieve results. The infrastructure would be scattered geographically, with several tiny, distributed data centers positioned in areas of the nation where wind or solar generation is known to be strong.
“With this approach, firms will consume electricity that would otherwise be dumped, and everyone else would gain since the grid operator will not have to ramp up natural gas power in the nighttime hours to meet demand,” Raghavan explained.
There are certain limits that the researchers investigate in the study, such as the fact that it is only achievable under particular workloads and settings. However, Raghavan feels that with better prediction and integration into huge systems, the technology might be a viable future option for storing renewable energy. “We need every tool we can acquire in the civilization-scale problem of sustainability,” Raghavan added.
