Unpacking the software layer of VPP deployment

Last summer, Pacific Gas & Electric and energy services company Sunrun teamed up for a first-of-a-kind virtual power plant pilot directed at mitigating California’s summer capacity emergencies.

Over the course of three months, the pair operated a “permanent load shift” program during which more than 8,000 residential batteries throughout the state discharged back to the grid for the same period each evening. The program provided the grid with an average of 27 megawatts of power each evening, and experienced exceptionally low opt-out rates.

The less-visible but nonetheless crucial software layer of that pilot was orchestrated by Lunar Energy, a battery maker and grid services company founded by Tesla’s former head of energy, Kunal Girotra. Lunar emerged from stealth mode in 2022, with the announcement that it had acquired Honda-backed Moixa and its DER optimization software Gridshare.

Today Lunar is funded by Sunrun (as well as by the South Korean SK Group) and provides the digital controls for thousands of Sunrun batteries in VPP contracts across the United States; it’s what Lunar head of product Sam Wevers calls a “VPP SaaS service for Sunrun.” 

Juice for the grid

Lunar’s presence on the grid has expanded quickly. In 2023, the company delivered three and a half gigawatt hours of VPPs in the United States. 

“That’s not really pilot scale,” Wevers said. “That’s serious juice onto the grid.”

But, he added, VPPs exist on a spectrum, and many programs get the nametag even though they aren’t necessarily VPPs. The distinction, for Wevers, comes down to whether issues on the grid can be responded to with implicit or explicit flexibility. Implicit flexibility includes device responses to price signals, like time of use tariffs, to save customers money. Explicit flexibility is essentially contracted flexibility based on grid needs, when there isn’t another economic incentive for batteries to respond in a certain way.

All of Lunar’s global projects fall somewhere between those two approaches, Wevers said. In Japan, for example, where the company manages more than 40,000 residential batteries, all customers are on time of use tariffs. Lunar essentially connects to those devices, and discharges them when power is expensive, which has the effect of significant peak shaving — but which isn’t really a VPP, he added.

In the U.S., VPPs tend to be defined as utilities paying customers for a particular action, though the programs deployed to date aren’t particularly complex.

The PG&E pilot with Sunrun, for example, is large but otherwise falls on the simpler end of the VPP spectrum: providing power to the grid in bulk at peak times, and providing customers a flat fee. Repeated dispatch at the same time every day, like in the PG&E pilot with Sunrun, is much less complicated than a program that’s highly responsive and might dispatch at different times, Wevers said.

But this simplicity is part of the reason VPP deployments in the U.S. have generally been successful, he added. 

“The programs themselves, in terms of what you need to deliver, are not too complicated,” Wevers said, adding that the bulk of deployments in the U.S. have been geared at decongesting distribution lines, and peak shaving. “More advanced things are starting to become available in different places, like Texas for instance, that are a bit more tailored around wholesale pricing and stuff like that. But broadly speaking, that’s not been the bulk of VPP programs in the USA.”

Today, Wevers said, flexibility is integral to the VPP, given how much the country differs by region:  “If we could only deliver one model, then we’d only be working in one very specific place.”

Technical challenges

Certain hurdles remain. For instance, PG&E found that in its summer pilot,  a number of unresponsive batteries dampened output. But according to Wevers, that result shouldn’t be entirely unexpected — and was in fact mitigated by Lunar’s modeling.

“These are devices in people’s homes at scale, and they are not going to work as reliably as one hardcore massive battery with a dedicated control team,” Wevers said. “When you send an instruction to a huge fleet of devices, you get a good response overall, and yes, they’re not all going to respond, but that’s not necessarily a problem.”

The bigger the fleet, the better, he added, because unresponsive devices can be swapped out for different ones. For example, when Hurricane Hilary hit California as a tropical storm in the early weeks of the PG&E pilot, Lunar integrated National Weather Service data into its Gridshare platform to make sure that batteries in the storm’s path prioritized maintaining power for homes.The company made up the difference with the rest of the VPP fleet.

That intervention layer is key when it comes to keeping customers from dropping out of VPP programs, Wevers said. 

Lunar’s approach is based on its history with behind-the-meter energy management programs in Japan, where it receives 40 billion data points every month.

“We had to build a system that could digest a huge amount of data from every single home and then create machine learning predictions of every home’s consumption and solar generation,” Wevers said. “We take a really bespoke approach, and we’re doing it at scale.”

Translating that to the U.S. market, even for simple VPPs, means that Lunar can anticipate how a fleet will behave over the course of 24 hours, he said. Combining that predictive capability with asset management means Lunar can provide highly tailored VPP deployments, Wevers added, but those aren’t quite happening yet in the U.S.

That’s in part due to a handful of technological gaps that are holding VPPs back. 

For instance, Wevers said more standardization —  in terms of how utilities procure VPP services, how program success is measured, and whether programs are preset like the PG&E pilot or triggered by the wholesale market — would go a long way toward smoothing the road to deployment. 

“As they standardize, they’ll start to see more devices come on board,” he added.

VPP deployment could also be helped by standardization of device connectivity, he said, pointing to efforts like Open Automated Demand Response and the IEEE standards.

“It’s essentially about trying to get to a place where it’s easy to connect to people’s devices, regardless who the maker is, and easy to dispatch those devices into services,” Wevers said. “But the fact that everything is so fragmented in the U.S. makes it quite difficult.”