Argonne Lab Using AI for Smarter Electric GridJune 17, 2019 by Scott McMahan
Argonne National Laboratory researchers are striving to make optimization models that employ machine learning to simulate the electric system and the severity of various problems.
Answering the question, "How much electricity will you need tomorrow?" is somewhat predictable, but the answer also varies considerably. Well designed computer models are intended to manage this inherent uncertainty in predicting power needs and avoid surprises. Electric grid operators use such computer models that help estimate everything from power demand to traffic patterns.
"Argonne's approach decides whether the current conditions of the system are expected based on past behavior, or whether something is new and different. This information can be used to alert the operator that they may have something they don't expect on the grid." — Mihai Anitescu, senior computational mathematician at Argonne National Laboratory
Factoring in both the certain and the unknown to deliver electricity throughout all kinds of scenarios requires a series of amazingly complex math problems.
With the aid of artificial intelligence (AI), researchers at the U.S. Department of Energy's (DOE) Argonne National Laboratory are devising new methods of finding insights from mountains of data about the electric grid.
The goal of such models is ensuring more reliability, resilience, and efficiency.
The work brings together Argonne's grid expertise with its experts and its advanced computing facilities.
A Better Understanding of Uncertainties
Uncertainties power demands are not a new phenomenon. Grid operators have always had to face challenges and some amount of uncertainty from factors including extreme weather, or equipment failures.
However, fluctuating and intermittent supplies of renewable energy, some of it flowing from residential rooftop solar panels outfitted with smart meters, are expanding the number of variables grid operators must examine.
The researchers note that in a region with 1,000 electric power assets, an outage of only three assets can produce nearly a billion scenarios of potential failure. Which of those potentialities will require the most attention?
Solving such complex modeling problems is time-consuming. With resources including the Argonne Leadership Computing Facility (ALCF), a DOE Office of Science User Facility, researchers can model multiple scenarios in parallel, accelerating the process.
"The idea is to generate a large number of scenarios and train the machine learning model to tell us the answer," said Kibaek Kim, an assistant computational mathematician at Argonne. "Instead of solving a number of difficult optimization models over several hours or days, we train the model ahead of time and then get the answer right away."
The researchers can use supervised learning to train the machine by feeding it a set of data that includes the solutions. This is analogous to a student studying previous "exams" before trying new ones.
Another method, unsupervised learning, feeds a computer raw data and allows it to find patterns without telling it any "answers."
In one study, Kim and his colleagues used a graph convolutional neural network to suggest optimal controls that would prevent the overloading of transmission lines if there were an issue with any of the lines.
They determined that this model, which used machine learning for rapid solution finding, produced far fewer errors than more traditional ones. The work leveraged Argonne's Laboratory Computing Resource Center (LCRC) and its Joint Laboratory for System Evaluation. In addition to the LCRC, Kim's work entails collaboration across Argonne.
Kim's team intends to make such models more robust, giving grid operators more effective recommendations that can inform more reliable planning and operations for unexpected events such as storms, equipment malfunctions and big fluctuations in renewable energy generation.
Another activity at Argonne is applying AI to speed up the daily calculations for regional electric system planning. One calculation is the security constrained unit commitment (SCUC), which aids grid operators in setting a daily and hourly schedules for power generation.
"In power systems, this SCUC problem is solved multiple times a day," said Feng Qiu, principal computational scientist at Argonne. "Since this problem is solved repeatedly, we can accumulate a lot of data and discover patterns that could be used to solve the next round."
Instead of replacing contemporary analytics with machine learning, said Qiu, the idea is to bolster existing ones using machine learning to suggest "hints" learned from prior solutions.
Using LCRC's Bebop cluster, a team led by Argonne postdoctoral appointee Alinson Santos Xavier devised AI that can solve SCUC about 12 times faster, on average, than traditional approaches. An early version of the technique was used successfully in tests at Midcontinent Independent System Operator (MISO), which manages electricity distribution across 15 states and one Canadian province.
"All this can lead to a more efficient market and more cost-effective electricity production," Qiu said. "For long-term planning, it could help grid operators consider more scenarios and make better expansion plans."
Programming for a Smarter Grid
Modernized grids increasingly include sensors that monitor conditions throughout the system, and these also offer potential uses for enhanced data processing. Devices installed on transmission lines and at substations, for example, can work as sentinels that alert grid operators to equipment issues when they occur.
Argonne scientists have examined an entire year of sensor data from ComEd, a utility that serves almost four million customers in the Midwestern U.S.
For this effort, the researchers employed unsupervised learning, feeding the data to the machine and asking it to find sensor output anomalies.
"It's not always known to the operator when things do not work as they should," said Mihai Anitescu, senior computational mathematician at Argonne, who worked on the project. "Our approach decides whether the current conditions of the system are expected based on past behavior, or whether something is new and different. This information can be used to alert the operator that they may have something they don't expect on the grid."
This kind of classification activity can also be applied to weather forecasting for renewable energy, correcting estimates of wind resources near bodies of water, for example, and coupling numerical models with physical measurements to enhance accuracy.
A lot of AI efforts, notes Anitescu, involve examining pure data such as recognizing speech patterns or analyzing a picture: "There aren't many physical rules," he added.
However, that's not true for large real-world systems such as weather or the electric grid. "You really have to reconcile data, even if there's a lot of it, with the physical information," he said. "This is very much in its infancy, and it's really where supercomputers are necessary."
Funding for Argonne's AI work on the grid has come from DOE's Office of Science, the Advanced Grid Research and Development Division in DOE's Office of Electricity and Argonne's Laboratory Directed Research and Development Swift program for short-term projects.