How AI and IoT Are Making the Grid Smarter

IoT and AI are making the grid more user-friendly, reliable, flexible, secure and profitable.

(Image courtesy of the National Renewable Energy Laboratory (NREL).)

(Image courtesy of the National Renewable Energy Laboratory (NREL).)

Today’s sensors are small, intelligent and inexpensive. Armed with communication capabilities, some of these miniature monitors are joining the billions of devices that comprise the Internet of Things (IoT). Utilities take advantage of this by monitoring virtually every aspect of power generation, distribution and consumption using sensors that predict component failures, controllers that automatically reroute power from point to point, and smart meters that closely track customer usage and trends. Processing that much information is far beyond human computational capabilities, so utilities are relying on the IoT and artificial intelligence (AI) to make the grid more user-friendly, reliable, flexible, secure and profitable. 

Some of the ways in which AI is helping to make the smart grid smarter. (Image courtesy of the National Renewable Energy Laboratory [NREL].)

Some of the ways in which AI is helping to make the smart grid smarter. (Image courtesy of the National Renewable Energy Laboratory [NREL].)

The Customer Experience

Two years ago, our local power company installed smart meters in the area. Shortly after, they offered customers a choice of keeping their current flat electric rate or switching to a time-of-use model, where the cost of electricity varies, hour-by-hour, depending on supply and demand. During high-demand hours—usually midafternoon through the early evening—rates are higher. Under this plan, the average hourly rate is usually much less than the flat rate, but customers also pay for demand—the maximum power that a customer is drawing from the grid during a given period of time. (In our case, it’s in 15-minute increments.) Customers paying the flat rate aren’t charged separately for demand; it’s rolled into the flat rate. 

From the customer standpoint, time-of-use pricing allows us to reduce our electric bills by using less power when rates are high and reducing the number of high-power devices running simultaneously. The utility incentivizes these behaviors because it reduces the load on their generation and transmission equipment and decreases the need to build peaker plants, which typically run for just a few hours each day and rarely pay for themselves. To help the consumer decide when it’s best to use energy, the power company provides a website with real-time rates.    

Hourly rates for Time of Use pricing for two days in November. (Image courtesy of ComEd.)

Hourly rates for Time of Use pricing for two days in November. (Image courtesy of ComEd.)

Currently, this information helps me to decide when to turn on major appliances (washer, electric dryer, dishwasher, etc.) and when to postpone their use until the electric rate goes down—a “man-in-the-loop” control system. I’m also actively reducing demand by not running multiple high-power devices at once. (For example, we don’t run the bread machine at the same time as the dryer, regardless of the real-time rate.) We don’t own an electric vehicle (EV) yet, but eventually, that will become a standard “appliance” in most homes, adding yet another variable to the equation. 

EV charging at home. (Image courtesy of Tesla.)

EV charging at home. (Image courtesy of Tesla.)

It’s certainly not an overwhelming amount of information to process, nor is it a complicated decision tree. (That’s perhaps easy for me to say—I’m an engineer!) But the average consumer doesn’t want to worry about demand and time-of-use pricing; they want an automated system that frees them from mundane decision-making, so utilities will soon allow a home’s smart appliances to communicate directly with the smart meter, which can tell the appliance whether to run based on the current rate and the amount of power currently being used by the household. A user can set priorities (e.g., Do I need a clean shirt before the loaf of bread is baked?) and let the system handle the load shifting.  

Managing a home’s electricity usage doesn’t require AI or machine learning if you’re only considering home energy consumption, but the power company has to balance supply and demand from all of its customers. Now, factor in rooftop solar, behind-the-meter storage, and net-metering, which can make every home a potential power plant. Suddenly, both the homeowner and the utility are making decisions that affect the other in terms of when to charge the batteries, when to run on battery power, when to sell power to the grid, and so on. And since solar is affected by weather, every day is a new adventure in estimating production. This is where the variables become so numerous that AI is necessary. A machine learning algorithm can track customer behaviors to predict energy consumption while tapping into weather forecasts to estimate solar production.

Operations and Maintenance

In addition to balancing supply and demand, grid operators need to ensure reliability. A slew of IoT-based sensors provides a real-time snapshot of the grid, allowing operators to immediately detect fault conditions. In some cases, they may even identify impending failures. In either case, repair crews can respond more quickly, reducing downtime. In a best-case scenario, the system is intelligent enough to reroute power, limiting the disruption to a small group of customers. 

Quickly finding and correcting a fault is just the first step in improving reliability; next, grid operators need to determine the cause, and that’s where many of those sensors, and the information they produce, become valuable. By organizing billions of data points, AI can help engineers perform diagnostics and determine the root cause of grid malfunctions. Moreover, AI will improve models of the increasingly complex power grid, a task that grows more difficult with the proliferation of power producers, microgrids, and other distributed energy resources. 

Taking things a step further, machine learning can build models based on past performance and run simulations to make predictions about future behaviors. AI will help to recognize production and consumption patterns and choose the most appropriate model for each situation.  

Closing the loop altogether, diagnostics, analysis and simulation provide engineers with the tools they need to design more resilient and reliable energy networks, which expands the options for distributed energy systems, renewable energy (grid level and rooftop), energy storage, and an EV charging infrastructure. 

AI turns big data into grid optimization.

AI turns big data into grid optimization.

Grid Security

Decentralized energy production, assisted by a smart grid, increases security by isolating compromised segments, but AI has much more to offer in the grid security realm. Computer malware is designed to infect, mutate and propagate, and—just like a biological virus—the mutations make malicious code more difficult to detect and remove. Machine learning excels at pattern recognition, which is crucial to tracking code mutations and identifying malware such as denial-of-service attacks like the one that affected the Ukrainian power grid in 2016

And while electricity theft doesn’t necessarily affect grid performance, it does increase everyone’s electric rates. (You didn’t think the utility would simply absorb the losses, did you?) Using data from sensors placed throughout the transmission and distribution network, AI identifies anomalies that could be attributed to someone physically tapping into the grid and siphoning power. At the consumer level, machine learning algorithms analyze data from smart meters to identify usage changes indicative of meter tampering. 

NREL Report

As part of its “Voices of Experience” series, the National Renewable Energy Laboratory (NREL) conducted a qualitative study of utilities’ experiences with AI, interviewing 61 individuals from 40 utilities across the U.S. The report doesn’t draw a roadmap on how to integrate AI into a particular utility, as such. Instead, think of it as a series of additions to the knowledge base, noting common themes that numerous utilities have encountered while incorporating AI into their businesses.  

Qualitative research, especially when dealing with new technology, is often used as an exploratory tool that helps us learn how to create and evaluate systems that involve human behavior. Why? Because regardless of how powerful the tool, if people choose not to use it or use it incorrectly, it has no value. So, the first steps in adopting AI are to motivate and educate the team members. The case studies presented in the report show how these utilities incorporated AI into their organizations by encouraging collaboration between AI experts and all corporate stakeholders. 

Vehicle electrification, energy management and renewable generation are all converging into a perfect storm that the current grid is unprepared to navigate. Going back to shore is not an option, so engineers are harnessing the power of that storm with data, IoT and AI.