How Machine Learning Can Transform The Energy Industry
In 2017, Bill Gates wrote an open letter to graduating college seniors across the world. He mentioned that if he had to start over in life and look for an opportunity to make a big impact on today’s world, he would consider three fields: “One is artificial intelligence. We have only begun to tap into all the ways it will make people’s lives more productive and creative. The second is energy, because making it clean, affordable, and reliable will be essential for fighting poverty and climate change.” The third field was bio-sciences. [1]
Today, as countries try to diversify their energy portfolios and effect a greater reliance on cleaner power, they are left with one major problem. The two main sources of renewable energy- solar and wind- are, in their very nature, variable. The power generated by a solar panel or a wind turbine is never uniform and depends on a range of external factors — intensity of solar radiation, cloud cover, wind speed — that can’t be controlled.
Even Germany, the world’s poster child for renewable energy, finds it hard to solely depend on renewable energy. On calm and cloudy days, when hardly any energy is produced by solar and wind farms, grid operators are forced to call upon conventional power stations to meet the energy demand. On the other hand, if it’s a sunny and windy day, and 90% of their energy requirements for the day are fulfilled, the operators need to quickly reduce the output from coal and gas-fired power stations before an overload of power suffocates the entire grid. Such requests cost German customers around US$553 million a year as grid operators have to compensate utility firms for making adjustments to their inputs. [2] Additionally, grid operators end up emitting needless carbon dioxide emissions as the extra power ends up getting wasted. To add to all this, accurately forecasting energy demand among consumers is another challenge of its own. Overall, maintaining an equilibrium across energy supply and demand can turn into an absolute nightmare.
This brings me to the role machine learning could have in the overall energy spectrum. Even though it is in its early stages of implementation, machine learning could revolutionize the way we deal with energy. Its impact ranges across the areas of renewable energy forecasting and smart grids.
Renewable Energy Forecasting
If we could accurately predict the amount of electricity generated by a wind turbine in the next 36 hours, we would be able to effectively transition to renewable energy without destabilizing the power grid. But is that possible? The answer is yes. We’re getting there.
As early as 2013, IBM, in collaboration with the US Department of Energy, started working on ways to leverage Watson, their AI engine, for cleaner power. This machine learning model was built from several forecasting models and was later fed with data on the weather and atmosphere from around 1,600 sites across the United States. As time passed, this machine learning model got better at making predictions regarding power output.
Today, IBM research has over 200 partners that use its solar and wind forecasting technology and can closely predict solar/wind conditions 15–30 days in advance. IBM’s renewable forecasting technology (called Watt-sun) is “50% more accurate than the next best solar forecasting model”, says Hendrick Hamann, a project manager at IBM. [3]
The economic potential could be massive. The perfect case study for this is DeepMind, a British AI company recently acquired by Google.
In 2018, DeepMind began applying machine learning algorithms to 700 MW of Google’s wind power capacity in central US (700 MW is enough electricity to power a medium-sized city). Using a neural network that worked on available weather forecasts and historical turbine data, it could reasonably predict wind power output 36 hours in advance. In just a year, these machine learning algorithms boosted the value of their wind energy “by roughly 20%, compared to the baseline scenario of no time-based commitments to the grid”, says Sims Witherspoon, a program manager at DeepMind. [4]
A stronger reliance on machine learning algorithms can also save the customer’s money while rescuing the planet at the same time. Xcel Energy, a utility firm that handles the highest total wind capacity in the United States, did just that. [5] Drake Bartlett, a renewable-energy analyst with Xcel Energy says:
“The number of forecasting errors has dropped since 2009, saving customers some US$60 million and reducing annual CO2 emissions from fossil-reserve power generation by more than a quarter of a million tons per year.”
Smart Grids
As predictive capabilities are improving, countries are gradually moving towards establishing a ‘smart grid’, a “fully automated power delivery network that monitors and controls every consumer and node, ensuring a two-way flow of electricity and information”. [6]
Since 2010, the US Department of Energy has invested over $4.5 billion in establishing smart grid infrastructure. They have installed over 15 million smart meters, devices on the consumer end that monitor energy demand and supply. Additionally, they are investing in devices known as ‘synchrophasers’. These brief-sized boxes measure the instantaneous voltage, current and frequency at specific locations on the grid. These sensors would communicate with the grid and modify electricity flow during off-peak times, lowering prices for the customer while also relaxing the workload of the grid. Even Google has applied this AI technology in an effort to reduce their total power consumption from its data centers, saving millions of dollars in the process. [7]
But even with a smart grid, there is a potential source of concern. A central system that collects data about the energy usage habits of millions of users can emerge as a target for malicious cyber-attacks. This could potentially destabilize a grid while also damaging precious consumer data. UK researchers feel that block-chain protocols could be the solution. Using the same technology as Bitcoin, a decentralized ledger system could avoid the security risk of having a single point of storage for user data. [8]
Key Takeaway
With improved predictive capabilities integrating into a smart grid, countries can increasingly depend on renewable energy, facing minimal disruption caused by the irregularity of solar and wind energy. The convergence of machine learning and energy could change the world like never before, and radically alter the way we thought about these two industries.
Feel free to email me at kausarp2@illinois.edu or connect with me on LinkedIn!
Inspirations: @TadasJucikas & Franklin Wolfe
