How utilities use AI to adapt to electricity demand22. April 2020
How utilities use AI to adapt to electricity demand
New York, 22.4.2020
The spread of the novel coronavirus causing COVID-19 has prompted state and local governments in the United States to set up emergency shelters and close down businesses. With millions of people suddenly finding themselves in their homes, this change has put pressure not only on Internet service providers, streaming platforms and online retailers, but also on the utilities that provide electricity to the country’s power grid.
US electricity consumption was 3% lower on 27 March 2020 than on 27 March 2019, representing a loss of about three years of revenue growth. This is the result of a report by “VentureBeat”. Peter Fox-Penner, director of the Institute for Sustainable Energy at Boston University, claimed in a recent statement that utility revenues will suffer because utilities will stop shutdowns and postpone tariff increases. In addition, according to research firm Wood Mackenzie, the increase in household demand for electricity will not compensate for the lower demand for electricity from companies, mainly because household demand accounts for only 40% of total demand in North America.
The spread of the novel coronavirus that causes COVID-19 has prompted state and local governments in the U.S. to place on-site orders and close down businesses. With millions of people suddenly finding themselves in their homes, this change has put pressure not only on Internet service providers, streaming platforms and online retailers, but also on the utilities that provide electricity to the country’s power grid.
Some utilities are using AI and machine learning techniques to manage the windfalls and fluctuations in energy consumption caused by COVID-19. Accurate load forecasting could ensure that operations are not interrupted in the coming months, which could prevent power outages. And it could also increase the efficiency of utilities’ internal processes, which could lead to lower prices and improved service long after the pandemic ends.
Innowatts, a startup company developing an automated toolkit for energy monitoring and management, counts several large U.S. utilities among its customers, including Portland General Electric, Gexa Energy, Avangrid, Arizona Public Service Electric, WGL and Mega Energy. Its eUtility platform collects data from more than 34 million smart energy meters from 21 million customers in more than 13 regional energy markets, while its machine learning algorithms analyze the data to predict short and long-term loads, deviations, weather sensitivity and more.
Beyond these table-stage forecasts, Innowatts helps to assess the impact of different tariff configurations by mapping the tariff structures of utilities using disaggregated cost models. It also creates cost curves for each customer that show the margin impact on the wider business, and validates product revenue and customer acquisition costs with models that reveal the relationship between marketing efforts and customer behavior (such as real-time load).
Innowwatts reported that it observed “dramatic” shifts in energy consumption between the first and fourth weeks of March. In the Northeast of the country, “unimportant” retailers such as salons, clothing stores and dry cleaners consumed only 35% of the energy at the end of the month (after shelter-in-place orders) compared to the beginning of the month, while restaurants (excluding pizza chains) consumed only 28%. In Texas, on the other hand, storage facilities consumed as much energy in the fourth week as in the first 142%.
Innowatts says that during these usage increases and decreases, its customers took advantage of AI-based load forecasting to learn from short-term shocks and make timely adjustments. Within three days of the shelter-in-place order, the company says its forecasting models were able to learn new consumption patterns and produce accurate forecasts that reflect changes in real time.
Sid Sachdeva, CEO of Innowatts, believes that demand forecasts in mid-March would have shown deviations of 10-20% if the utilities had not used machine learning models, which would have had a significant impact on operations.
“In these turbulent times, AI-based load forecasting is giving utilities the ability… With utilities and energy retailers experiencing a one-time drop in commercial energy consumption of more than 30%, accurate forecasting has never been more important. Without AI tools, utilities would see their forecasts fluctuate sharply, resulting in inaccuracies of 20% or more, placing a huge burden on their operations and ultimately driving up costs for businesses and consumers”.
Autogrid works with over 50 customers in 10 countries – including Energy Australia, Florida Power & Light and Southern California Edison – to provide AI-informed insights into power consumption. Its platform produces 10 million forecasts every 10 minutes and optimizes over 50 megawatts of power, enough to supply an average suburb.
The company’s flagship product, Flex forecasts and manages tens of thousands of energy resources for millions of customers by capturing, storing and managing petabytes of data from trillions of endpoints. Using a combination of data science, machine learning, and network optimization algorithms, Flex models both physics and customer behavior, automatically anticipating and adjusting supply and demand patterns.
Autogrid also provides a fully managed solution for the integration and use of end-user battery and microgrid installations. Like Flex, it automatically aggregates, forecasts and optimises the capacity of substation and transformer installations, responding to distribution management needs while providing capacity to avoid capital investment in system upgrades.
Dr. Amit Narayan, CEO of Autogrid, told VentureBeat that the COVID 19 crisis has greatly shifted the daily distribution of electricity in California, where it is exerting “significant” downward pressure on hourly energy market prices. He said that Autogrid has also heard from customers about transformer failures in some regions due to overloaded power circuits, which he believes will become a problem in the summer months (when air conditioning use increases) in heavily loaded residential areas and areas with saturated loads.
“In California in 2019, more than one million residents faced outages related to forest fire prevention in the PG&E area,” Narayan said, referring to the controversial planned outages staged by Pacific Gas & Electric last summer. “Demand remains high despite the COVID 19 crisis in 2020 as residents prepare for a similar situation this summer. If there is a recurrence in 2019, it will be even more devastating given the health crisis and the difficulties in buying food.
AI makes a difference
Artificial intelligence and machine learning are not a miracle weapon for the power grid – even if they have foresighted tools at their disposal, utilities are exposed to a stormy demand curve. But utilities say they see evidence that the tools are already helping to prevent the worst of the impact of the pandemic – especially by helping them adapt better to changing daily and weekly power load profiles.
“The social impact [of the pandemic] will continue to be felt – people can continue to work remotely instead of going to the office, they can change their commute times to avoid rush hour traffic, or they can look for alternative transportation,” Emmanuel Lagarrigue, Chief Innovation Officer of Schneider Electric, told VentureBeat, “All of this will affect the daily load curve, and this is where AI and automation can help us maintain, perform and diagnose in our homes, buildings and on the grid.