Photo 132528154 © Wrightstudio | Dreamstime.com
Dreamstime M 132528154
Dreamstime M 132528154
Dreamstime M 132528154
Dreamstime M 132528154
Dreamstime M 132528154

Protecting the Future of the Electric Grid with Artificial Intelligence

July 28, 2023
As the DOE urges grantees to incorporate advanced technology into grid operation, more and more power utilities will look to solutions that employ artificial intelligence (AI) technology to take advantage of the funding.

Following major outages over the past year in Texas and California, it became clear that America's power grid system is long overdue for a refresh. Fortunately, the Department of Energy (DOE) has taken action to address these outages with the largest single direct federal investment in critical transmission and distribution infrastructure.

With $10.5 billion set aside for competitive grants under Grid Resilience and Innovation Partnership (GRIP) programs, the energy industry has the power to make substantial changes. $2.5 billion of this funding is allocated for grid resilience and industry grants, which will aid efforts to modernize the grid, with the goal of reducing the effect of extreme weather and natural disasters on grid operations. These grants will go to grid operators, power storage operators, electricity generators, distribution providers, transmission owners or operators, and fuel suppliers.

As the DOE urges grantees to incorporate advanced technology into grid operation, more and more power utilities will look to solutions that employ artificial intelligence (AI) technology to take advantage of the funding. AI technology can transform the American grid by providing accurate prediction of energy consumption and production levels days in advance, giving operators the ability to plan for and prevent power outage incidents, while increasing overall power grid reliability. AI can also help power utilities manage numerous flexible loads and distributed generation units in real time while introducing new efficiencies to lower costs, reduce waste, and cut carbon emissions.  

Anticipating Energy Demand with AI

The United States’ grid network was not designed to handle modern demand. Power is typically generated in certain parts of each city, so when every home, factory, and shop needs electricity during high-usage times, transmission lines are stretched to their capacity as they attempt to send power in every direction. When the grid fails, blackouts can leave entire cities without power for hours, even days. To make matters worse, this happens with increasing frequency as the world digitizes, and daily operations become more reliant on electricity.

To resolve this ongoing issue, AI can improve grid resiliency through accurate predictions about the amount of energy consumers will use in the coming days or months. This is essential as utilities depend on both short and long-term power demand predictions to successfully plan operations. Because demand is heavily dependent on predictable circumstances—like cyclical consumer behavior, business activities and severe weather swings—AI can analyze patterns and point associations in historical data to reliably generate demand predictions.

This is already in play, with many power utilities encoding these predictions with AI technology in the form of tree-based models and neural networks. The models are trained to spot trends that could affect the next day’s electrical demand, allowing the utility to adjust the amount of power they generate to closely meet demand.

With well-planned operations, utilities can minimize spinning reserves. This comes with the added benefit of reducing carbon emissions and cutting operational costs, as the plant is not running unnecessary generation units. As utilities put GRIP program grants to use, AI will emerge as a key tool for grid operators across the United States.

Predicting Power Generation in Real-Time

AI can also play a role in improving grid operations in real-time. Most companies currently rely on methods such as mixed-integer programming — mathematical models with switching variables that involve generating a sequence of estimates of the solution—to predict power generation and enhance power plant operation. However, as the number of renewable power sources (i.e. solar and wind) grows, these calculations become increasingly complex given the stochastic and intermittent nature of renewable power generation. The addition of extra decision variables has made mixed-integer programming a much slower process than it used to be.

To accurately predict power generation levels, the industry must move away from traditional methods. AI offers an alternative method by replacing traditional optimizers with learning-based search mechanisms such as reinforcement learning. AI-powered methods like reinforcement learning can handle far more complex calculations and stochasticity than mixed-integer programming. Reinforcement learning models can be trained on several plant operation scenarios to learn patterns and make real-time decisions. In turn, this cuts runtime significantly, providing insights to operators within a fraction of a second, as opposed to the hours that it takes to solve complex problems with numerical programming methods.

The capacity for AI solutions to help plant operators predict power generation, something that is necessary as the industry looks to conserve resources, indicates that the energy industry is sure to take advantage of this emerging technology.

An AI Power Grid

Although AI technology is appealing for an industry with the resources to upscale its operations, it’s not necessarily a silver bullet. Many AI-based technologies give rise to concerns around feasibility and safety. For example, mixed-integer programming have hard-coded constraints in the model, whereas reinforcement learning AI models only rely on training, which does not ensure that every solution it proposes will be safe.

Yet, this is by no means a permanent problem. Even today, work is being done to eliminate this issue, with some ideas involving the combination of AI with traditional numerical methods. Other options include the development of platforms that offer AI as a decision-support tool to aid human operators, rather than relying solely on AI as the decision-maker. There is also the possibility of encoding constraints into these models, such that they only choose decisions from a certified domain.

Given the promise that AI technology is already showing for the grid in these early stages, and the solutions that are currently at work to improve the grid, AI has the potential to remedy the energy industry’s biggest problems. The flood of funding and resources that the industry will soon experience marks the beginning of the AI revolution for American grids.

Mehdi Hosseini is Data Scientist at Beyond Limits.

Voice your opinion!

To join the conversation, and become an exclusive member of T&D World, create an account today!