Xcel Energy – Utilizing Machine Learning to Efficiently and Reliably Incorporate Renewable Energy into the U.S. Energy Grid
Given the importance of addressing climate change and adding renewable generation to the energy grid, Xcel Energy is using machine learning to overcome the challenges of the intermittent nature of wind power.
Increasing the amount of power generation from renewable energy sources is critical to slowing carbon emissions and limiting the negative effects from climate change. However, incorporating renewable generation in larger quantities is surprisingly difficult because wind and solar resources are inherently intermittent and difficult to forecast. A cloud blocking the sun or the wind not blowing decreases the power output and has proven very hard to predict. The grid operator, who ensures the supply of energy in the system is enough to meet demand, must replace this energy loss with idling reserve generation that is typically conventional gas and coal power plants (n.b., costly and dangerous blackouts are caused by a mismatch in supply and demand). This uncertainty requires operators to have a reserve of generation on standby in order to account for forecasting error and often results in a surplus of renewable energy supply because the operator is running more predictable carbon-intensive generation instead.
Machine Learning at Xcel Energy
To address the issue of intermittency and forecasting error, Xcel Energy, a large Midwest utility, has partnered with the National Center for Atmospheric Research (“NCAR”) to utilize machine learning to provide more accurate wind forecasts, reducing the margin of error. Xcel Energy provides NCAR data from sensors on hundreds of wind turbines and NCAR utilizes this data to develop high-resolution wind forecasts through artificial intelligence techniques. These new forecasts have been able to decrease the forecast margin of error by 40%. Because of the increased accuracy, Xcel Energy has been able to reduce costs to end customers by $60MM and saved a quarter of a million tons of carbon emissions per year through higher renewable energy utilization and less re-dispatching of conventional coal and gas generation.
Other Potential Energy Applications
Xcel Energy has been able to sustain the highest proportion of renewable generation (20% of total) out of all U.S. utilities, in part due to machine learning and the ability to more accurately predict wind resource and the resulting renewable generation available to the grid. However, this is just one of many applications of machine learning that can be used to optimize the energy grid. IBM has utilized similar techniques with Watson to improve solar forecasting, showing accuracy improvements of up to 30%. Numerous companies are applying machine learning techniques to the demand response sector (e.g., a smart home appliance such as a dishwasher not running during peak power demand hours with the highest prices and instead running overnight when prices are much lower). Xcel Energy is currently funding a collaborative investment platform called Energy Impact Partners that is investing in companies that aim to optimize energy consumption and improve renewable energy generation, but bringing these technological advancements in-house will be important. Going forward, Xcel Energy and other utilities must recognize how the future of the U.S. energy grid will be dependent on machine learning. The concept of a “smart grid” is a U.S. Department of Energy national policy goal which envisions an energy grid that utilizes machine learning and artificial intelligence to balance both sides of the supply and demand equation. On the supply side, battery storage will automatically capture surplus renewable generation and dispatch it efficiently back into the grid and accurate solar and wind forecasting will increase reliability and lessen the dependence on conventional fossil fuel generation. On the demand side, customers (i.e., residential, commercial, and industrial) will automatically reduce their demand during peak hours by relying on machine learning incorporation into demand response technology. By embracing these additional applications, Xcel Energy can continue to be on the cutting edge of grid modernization.
Machine learning is paving the way for a lower carbon footprint for the energy sector through the expansion of renewable energy generation. However, the topic merits discussion of potential unintended consequences. One consideration is the fact that conventional fossil generation and nuclear plants employ significantly more people than solar and wind farms. Indian Point, a nuclear plant in New York that is being shut down has 1,000 full-time workers; Wind Catcher, one of the largest wind projects in the U.S., will only need 80-90 operators when completed. How can Xcel Energy and other utilities mitigate the impact of job losses? Another important consideration is that the U.S. energy grid is critical infrastructure to national security. Any flaws in the implementation of the smart grid, whether security breaches or fundamental errors in the system, could cause catastrophic damage to the economy and potential loss of life – how can companies ensure the technology is safe because the stakes are so high?
(Word Count: 756)
 Bullis, Kevin. “Smart Wind and Solar Power.” MIT Technology Review, 20 Nov. 2014, www.technologyreview.com/s/526541/smart-wind-and-solar-power/.
 “NCAR to Develop Advanced Wind and Solar Energy Forecasting System for Kuwait.” NCAR & UCAR News, 4 May 2017, news.ucar.edu/126802/ncar-develop-advanced-wind-and-solar-energy-forecasting-system-kuwait.
 Proctor, Cathy. Xcel’s Forecast Rosy for Wind and Solar. Denver Business Journal, 4 May 2018, www.bizjournals.com/denver/news/2018/05/04/cover-storyxcel-s-forecast-rosy-for-wind-and-solar.html.
 Schiermeier, Quirin. “Germany’s Renewable Revolution Awaits Energy Forecast.” Nature, vol. 535, no. 7611, 13 July 2016, pp. 212–213.
 “EERE Success Story-Solar Forecasting Gets a Boost from Watson, Accuracy Improved by 30%.” Energy.gov, Office of Energy Efficiency and Renewable Energy, 27 Oct. 2015, energy.gov/eere/success-stories/articles/eere-success-story-solar-forecasting-gets-boost-watson-accuracy.
 “Corporate Responsibility Report: Advanced Technology.” Xcel Energy, www.xcelenergy.com/company/corporate_responsibility_report/library_of_briefs/advanced_technology.
 “Grid Modernization and the Smart Grid.” Energy.gov, Office of Electricity, energy.gov/oe/activities/technology-development/grid-modernization-and-smart-grid.
 Hough, Jack. How Batteries Will Change the Power Business. Barron’s, 11 June 2018, www.barrons.com/articles/how-batteries-will-change-the-power-business-1528509035.
Student comments on Xcel Energy – Utilizing Machine Learning to Efficiently and Reliably Incorporate Renewable Energy into the U.S. Energy Grid
A very interesting topic to explore! I think demand forecasting has huge importance in the utilities industry. With the advent of smart sensors and IoT, we will definitely see increasing importance ML will play in making sense of the data produced by the network of smart devices. I would envision that someday we will have real demand-side management solutions through a combination of remote devices and predictive ML algorithms.
I agree with you that jobs in the utilities sector may be affected, but displaced workers may be retrained to operate and maintain the smart device network and the verdict is still out on the net impact on jobs in the sector.
Better forecasting of wind availability is amazing technology — as utilities can better forecast the supply of renewable sources, less fossil-fuel baseload will be required. As Lingxi mentioned, the Internet of Things can also have a huge impact on the way that consumers use energy, by delaying appliance use to off-peak time, and cycling A/C demand during critical peaks. I wonder how much of the current problem is due to lack of technology, and how much of it is due to behavior: do households do laundry in the mornings because it’s an efficient use of time, or is it because they’re managing their work clothes with JIT delivery? Similarly, is the dishwasher being run after dinner (in which case it can be delayed), or is being run in advance of dinner, because clean cookware is needed? There’s probably some variation and some households for which significant behavior change will be needed in order to realize the benefits of machine-learning based IoT shifted load.
The point that you raise on displacing jobs in the utility industry is an interesting one, and can be generalized to other industries where machine learning is replacing human intervention. This problem is exacerbated by the fact that machine learning will then lead to further automation of grid management, wherein industrial components can conduct maintenance more efficiently. I believe that retraining should be the logical first option: as the business model of the utility changes so should its workforce. In addition to grid software management, workers could also be retrained to work with customers on numerous aspects, including sales and data analytics. As machine learning opens the door toward more decentralized electricity generation, workers will be shifted from working on power plants and grid infrastructure toward the construction of local generation plants.
As you mentioned, the need to secure the internet networks of this essential infrastructure should be the utility’s highest priority. In addition to investing in sophisticated cyber security infrastructure, the utility should also create robust contingent plans in case of an attack. The algorithms need to be built with a road map of how to react in the event of a hack. This may involve rerouting energy supplies, creating back-up components that do not sit on the main network, or decentralizing its technology hubs to mitigate proximity risk.