How AI chips could make smart meters smarter

The power grid of the future is going to be much more complicated than the power grid of the past, with a lot more rooftop solar systems, backup batteries, electric-vehicle chargers and smart electric appliances to manage. It’s going to be so complicated, in fact, that it might need computer chips capable of artificial intelligence to be embedded in the smart meters, grid sensors and other devices that utilities will use to keep it all running smoothly.

Last month, a consortium of utilities and Sunrun, the leading U.S. residential solar and battery installer, announced their support of one such effort. It includes two companies — grid technology vendor Utilidata and high-power chipset maker Nvidia — that have spent the past six months combining their expertise to bring this kind of computing capability to the edges of the power grid.

The work started with a Department of Energy grant-funded project launched in December to test the companies’ combined ​“software-defined smart grid chip” as a means of bringing real-time intelligence and control to distributed energy resources such as solar inverters, batteries and electric-vehicle chargers. In March, the companies demonstrated their first iteration of the technology, a device that can plug into digital smart meters that have already been deployed, with rural New York utility Lake Placid Municipal Electric.

Over the coming year, the partners intend to expand tests at utilities including Holy Cross Energy, the Colorado electric cooperative that’s been working with DOE’s National Renewable Energy Laboratory to control batteries and EVs in a residential smart-grid project, as well as Pittsburgh, Pennsylvania–based Duquesne Light, which is using the devices to map out its low-voltage distribution grid network.

Today’s smart meters don’t have the computing power to make the split-second decisions needed to optimize the flow of power between rooftop solar arrays, batteries, EV chargers and other distributed energy resources (DERs) and the grid at large, Utilidata CEO Josh Brumberger said in an interview. They also lack the programmable flexibility to adapt to emerging tasks, such as carrying out increasingly sophisticated analysis of the changing characteristics of a grid that’s seeing DERs grow to account for a majority of the electricity flowing across them, he said.

Utilidata’s technology, designed to support those kinds of split-second operations and sophisticated analytics to balance grid voltages for utilities including National Grid and AEP, could find much broader applications for managing DERs if the computing power to enable it came embedded in smart meters, he said.

Those meters are ​“the nexus between the customer, where all these new technologies are coming online, and the grid, which is ultimately responsible for making sure the lights stay on,” he said. ​“Now that it is increasingly clean and more resilient, what are the technologies required to drive that change?”

Smarter chips make smarter meters, smarter grids and smarter utilities 

Smart-meter makers including Itron and Landis+Gyr have been continually boosting the computing capabilities of their technologies, Brumberger said. But while they’re now supporting a wider array of applications to collect and provide data to utilities and customers, Utilidata’s partnership with Nvidia is ​“going a step deeper into that digital infrastructure,” he said, with a ​“chipset and modules that will fully realize these ambitions.”

That’s Nvidia’s Jetson platform, which is designed to embed high-speed and high-intensity computing that can support artificial-intelligence capabilities in distributed devices. It takes the same kind of capability now being used to train self-driving cars and enable other massive machine-learning tasks and applies it to smart meters and other grid devices, Marc Spieler, Nvidia’s head of global energy business development, said in an interview.

That could allow devices with Nvidia and Utilidata smart-grid chips to perform much more complex tasks than today’s smart meters are capable of doing, he said. He offered the example of analyzing the sub-second electricity waveform data that could reveal where grid equipment is starting to fail or is being stressed by the two-way power flows from distributed generation.

Today’s smart meters aren’t designed to sample, analyze and sort that kind of high-fidelity data on their own or in real time, he said. That means that companies like Utilidata that are analyzing data from smart meters and grid sensors for these purposes today are ​“doing it in the data center with data that’s days old.”

Real-time grid analysis is an important use case for Duquesne Light, said Elizabeth Cook, the utility’s general manager of advanced grid solutions. Over the past two years, she’s been working with Utilidata to collect and analyze grid data that can be fed into the utility’s software platforms for distribution-grid mapping and outage detection.

Duquesne Light finished deploying about 600,000 smart meters at the end of 2018, Cook said. But those meters only bring back energy-usage data on an hourly basis, and they don’t have the onboard capability to do the kind of fine-tuned data collection and analysis needed to inform the utility’s grid-mapping project, she said.

“A smart chip that can be deployed at the edge of the network that can do all that data analysis” could really help deliver the kind of visibility Duquesne is after, she said. 

Utilidata CTO Marissa Hummon highlighted the value that this kind of distributed data-analysis capability can bring to managing the disruptions that can emerge as more and more customers install solar panels or EV chargers, or make the switch from gas-fueled to electric heating, dryers, stoves and other appliances.

“At the substation, I don’t really worry about you turning your dryer on or charging your electric vehicle,” she said in an interview. ​“But at the very edge of the grid, you have to worry about coincident capacity issues. Me and my neighbors all turning on our EV chargers and electric dryers at the same time may blow the local transformer.”

Giving utilities the flexibility to analyze and forecast these kinds of hyperlocal issues could help them better manage processes, from connecting new DERs and all-electric appliances to constrained grids to operating the grid-control systems that keep them in balance, she said.

Cook agreed that utilities could make use of more data to better manage system planning and interconnection studies. ​“If we’re creating these databases, we can have better forecasting and more scenario planning,” she said. That, in turn, ​“ultimately will drive affordability and drive the costs down.”

Embedding this level of intelligence in smart meters could also significantly cut down on the costs of getting lots of disparate DERs to work together smoothly, said Bryan Hannegan, CEO of Holy Cross Energy.

“If you’re able to build this meter with the capability to orchestrate all these devices the consumer is going to go out and get, you help the consumer, and you save us the headache of having to instrument every one of these devices we want to orchestrate,” he said. ​“It’s almost like a microgrid controller inside the meter.”

How to pay for smarter smart meters

The more than 100 million smart meters that have been deployed across the U.S. aren’t technically capable of carrying out these more sophisticated tasks. In fact, earlier generations of smart-meter deployments haven’t been effectively put to use to perform the simpler tasks that many utilities promised they would provide, like delivering real-time data to customers.

These problems have led to utility regulators bringing an increased level of scrutiny to smart-meter rollout plans that can cost hundreds of millions or even billions of dollars. That’s added to pressure on utilities and meter vendors to be wary of smart-meter proposals that carry additional costs that come with embedding more sophisticated computing capabilities.

Spieler said he’s gotten this feedback from utilities and regulators he’s talked to about Nvidia’s work with Utilidata. ​“The first question is always, ​‘How expensive is the chip?’” he said.

While it’s hard to predict how smart-meter vendors might adopt the new chipset or how that might be reflected in prices, ​“I don’t believe there will be a significant difference in the overall cost of the implementation,” Spieler said — ​“maybe a few percent upfront in capital costs,” which represent about a fifth of overall smart-meter deployment costs.

“Shouldn’t the question be: What is the return to the consumer from the chip?” he asked. Not all utilities will be in a position to tap into the full range of capabilities that Nvidia and Utilidata are planning to provide from this technology right away, he said. But given that most smart-meter deployments are designed to stay in place for more than a decade, it’s important for utilities and regulators to be thinking ahead to what they might be asked to do in the future, he said.

In the meantime, Utilidata and Nvidia are looking at alternative approaches to getting the technology into the field, Hummon said. That’s starting with a test in Lake Placid of meter-attached modules, a type of device commonly known as meter collar adapters, as they’re designed to plug into the round sockets where meters connect to buildings. They allow utilities to test new technologies ahead of their full-scale integration by meter vendors, or to selectively deploy them to certain sites rather than rolling them out en masse.

Utilidata and Nvidia are also talking to manufacturers of solar inverters, batteries and EV chargers to gauge their interest in integrating the chipsets, Hummon said. More and more of these DERs are being aggregated into ​“virtual power plants” by non-utility actors to tap their value in wholesale energy markets today, and potentially to help utilities better manage their distribution grids in the future.

As the number of use cases for virtual power plants grows, so does the complexity of managing how they respond to different requests for serving those use cases, and how those individual responses interact with each other at the local grid level, she said. Technology that can handle this increasing complexity could make a big difference in how well the DERs being deployed today can adapt to play a role in these expanding opportunities, she added.

“We’re definitely selecting something that from the beginning is designed to support a broad, competitive third-party environment,” including utilities and DER aggregators, and create ​“a really robust and long-term platform for operations,” Hummon said.

That work is just getting started, Spieler pointed out. ​“The first thing we’re going to do with Utilidata is put pilot systems out there and capture data,” he said. The coming year will be ​“the year of pilots — collecting and processing data, all of that.”

“By next year, we should have models ready for production” in a variety of form factors, he said. ​“And once they roll out, you can upgrade to the latest version as often as you feel like doing it.”