This post was originally published by Allison Proffitt at AI Trends
By John P. Desmond, AI Trends Editor
Long-time US government investments to build a “smart grid” for delivering electric power are employing AI techniques to improve resiliency.
The US Department of Energy (DOE) has made supporting the smart grid a national policy goal, which entails a “fully automated power delivery network that monitors and controls every consumer and node, ensuring a two-way flow of electricity and information,” according to a report issued by Harvard University in 2017. Since 2010, the DOE had invested $4.5 billion in smart grid infrastructure and installed over 15 million smart meters that monitor energy usage per device and alert utilities of local blackouts. Estimates were that US energy demand was expected to increase 25% by 2050; the program was meant to limit the rise in peak electricity load to no more than one percent.
“AI will be the brain of this future smart grid,” the report stated. “The technology will continuously collect and synthesize overwhelming amounts of data from millions of smart sensors nationwide to make timely decisions on how to best allocate energy resources. Additionally, the advances made from ‘deep learning’ algorithms, a system where machines learn on their own from spotting patterns and anomalies in large data sets, will revolutionize both the demand and supply side of the energy economy.”
And, “As a result, large regional grids will be replaced by specialized microgrids that manage local energy needs with finer resolution. These can be paired with new battery technologies that allow power to continually flow to and between local communities even when severe weather or other outages afflict the broader power system.”
The report was prescient; the work is ongoing and progress is being made.
In April 2019, the DOE announced it would fund eight projects with $7 million to explore the use of big data, AI, and machine learning for better grid operation and management, according to an account in T&D World. The funding was part of a $20 million allocation the DOE is to disburse to promote innovative research and development in AI and machine learning. The DOE’s Office of Science announced that the remaining $13 million would support new research aimed at improving AI as a tool of scientific investigation and prediction.
(Here is a list of the Projects Awarded Funding referenced above.)
In July 2019, the DOE released the North American Energy Resiliency Model (NAERM) for national-scale energy planning including “a comprehensive resilience modeling system.” Dynamic Line Rating (DLR), to maximize load in the smart grid, was noted as an innovative technology to pursue. The Report stated the NAERM can, “Utilize artificial intelligence and machine-to-machine capabilities (M2M) to optimize the utilization and security of the energy sector.”
Scientists at DOE’s Argonne National Laboratory, based in Lemont, Ill., recently reported finding a novel approach to help system operators understand how to enhance the resilience of power systems with the help of AI. The new approach allows operators to make decisions considering both static and dynamic features of a power system in a single decision-making model with better accuracy — a historically tough challenge, according to a release from the lab.
“The decision to turn a generator off or on and determine its power output level is an example of a static decision, an action that does not change within a certain amount of time. Electrical frequency, though — which is related to the speed of a generator — is an example of a dynamic feature, because it could fluctuate over time in case of a disruption (e.g., a load tripped) or an operation (e.g., a switch closed),” stated Argonne computational scientist Feng Qiu, who co-authored the study. “If you put dynamic and static formulations together in the same model, it’s essentially impossible to solve.”
Operators of power systems must hold electrical frequency within a certain range of values to meet safety limits. Static conditions, such as the number of systems online, are usually calculated separately than dynamic features, such as ability to adjust load based on demand.
Attempts have been made to develop simple models that can bridge both types of calculations, but the models have been limited in their scalability and accuracy, particularly as systems become more complex. Qiu and his colleagues are using an AI neural network to help the static and dynamic formulas work together.
“A neural network can create a map between a specific input and a specific output,” stated Yichen Zhang, an Argonne postdoctoral appointee and lead author of the study. “If I know the conditions we start with and those we end with, I can use neural networks to figure out how those conditions map to each other.”
The team has tested its approach on a microgrid system, a controllable network of distributed energy resources, such as diesel generators and solar photovoltaic panels. Results have been encouraging. “The neural network transformed the complex dynamic equations that we typically cannot combine with static equations into a new form that we can solve together,” Qiu said.
The Argonne researchers hope power grid operators can use their model as a starting point. “We’re excited by the potential for this type of analytical approach,” stated Mark Petri, Argonne’s Electric Power Grid Program director. “For instance, this could provide a better way for operators to quickly and safely restore power after an outage, a problem challenged by complex operational decisions entangled with system dynamics, making the electric grid more resilient to external hazards.”
Startup PingThings Working on Data Platform to Support AI for Grid Resilience
Among the winners in the April 2019 DOE awards was PingThings, a startup that received seed funding from GE Ventures, which is working on incorporating AI to improve power grid resiliency.
Here is a description of the project: “The project will train and evaluate event and anomaly detection, classification, and real-time inferencing neural models and report on their performance when employed on live PMU [phasor “measurement unit] data streams.”
On its website, the company describes the PredictiveGrid platform as an “advanced sensor AI platform that is capable of ingesting, storing, accessing, visualizing, analyzing, and learning from high definition sensor data at grid scale.” The company has been awarded a total of over $8 million to fund its work, and is considering a Series A fundraising round in 2020.
NI4AI Project Aims to Create a Dataset to Help Apply AI to the Grid
Dr. Laurel Dunn, Grid Sensor and Data Research Coordinator for NI4AI, for National Infrastructure for AI on the Grid. The project is a partnership between PingThings and UC Berkeley, which was awarded $5 million in 2018, which will run through 2022. The award came from the Advanced Research Projects Agency (ARPA), a branch of the DOE created in 2007 and modeled after the Defense Advance Research Projects Agency (DARPA).
The announcement of the award by ARPA identified a critical need for the nation’s electric grid: “An increasingly complex grid with diverse resources alongside fast, automated disturbance responses requires a new level of insight, without which the system can become uneconomical, unreliable, and unstable.”
Furthermore, “The utility industry’s pace of innovation has been and continues to be hampered by a lack of data accessibility; ineffective data quality solutions; skill and tool mismatch; a lack of artificial intelligence (AI) and machine learning (ML) experts; and difficulty in transitioning research into production.”
Dunn focused her PhD studies at UC Berkeley on electric power systems and decision analysis at utility companies. “It’s compelling to me,” she said in an interview with AI Trends. “A lot of promising work is happening in the space of AI for utilities, but there is very little data going around. [Most of] the data is sensitive and protected.”
However, not all utility data is protected. “A lot of work has gone into figuring out what information is sensitive and what information is not,” Dr. Dunn stated. “It is possible to take “sensitive” datasets and strip them of any sensitive information that prevents them from being shared. NI4AI is working with utilities to create anonymized datasets that could more readily be exchanged.”
The NI4AI project aims to deploy sensors to collect data that is anonymized in a way to allow it to be open sourced. “We are hosting it on the [PingThings] platform and doing community engagements to get more people working with AI on the grid,” she said.
She hopes to build a community of interested people in the utility industry and in the analytics community to work on developing new AI applications that leverage the data. “We are trying to create a data set that is transparent and can be readily analyzed. Much of the data the utilities have is protected, so a student in school or an entrepreneur does not have access to data they could use to apply AI on the grid,” Dunn said. “We are trying to create the ecosystem that supports that, to show we have the appropriate data to train algorithms to address problems in the industry.”
Read the source articles and material from Harvard University, in T&D World and referenced Projects Awarded Funding, the North American Energy Resiliency Model (NAERM), from Argonne National Laboratory and from ARPA on NI4AI.
This post was originally published by Allison Proffitt at AI Trends