Crunching data to keep EVs from overloading California’s grid

Aram Shumavon, CEO of Kevala, has some bad news and some good news for California utilities and regulators trying to prepare for the millions of electric vehicles the state wants to have on the road by 2035.

The bad news? According to a newly released study from his grid-data-analytics startup, commissioned by California regulators, California’s big three investor-owned utilities will need to invest up to $50 billion by 2035 to beef up their distribution grids to handle the massive growth in electricity demand those EVs will create.

That’s way more than the utilities, Pacific Gas & Electric, San Diego Gas & Electric and Southern California Edison, spend today on those distribution grids — the networks that connect high-voltage, long-distance transmission lines to the low-voltage lines and transformers that power homes and businesses. And ​“the number could be much larger,” Shumavon said, depending on how quickly EVs are adopted.

Because EV uptake is challenging to predict, utilities are going to find it ​“incredibly difficult to thread the needle on all this,” he said. If California’s utilities fail to build ahead of growing EV power demand, they’ll risk overloading more than a third of their thousands of miles of distribution lines, according to Kevala’s analysis. Since no utility can tolerate widespread grid overloads, it’s more likely that California’s utilities would be forced to stall new EVs from getting plugged in until they can catch up on grid buildouts to serve them — an expensive delay to the state’s progress on vehicle electrification.

But utilities can’t just start spending billions of dollars to expand the grid everywhere. After all, those future grid costs, which utilities will recover by increasing customers’ bills, are going to be piled atop California electricity rates that are already higher than they’ve ever been, leading politicians and regulators to demand cost-saving measures.

It might be possible for utilities to manage the cost of this grid buildout through the revenue that can be brought in from selling electricity to all these new EVs. But that will take planning ahead to invest in the right amount of grid upgrades to serve EV loads just as they’re emerging. Without that planning, utilities risk either overbuilding in places where the EVs don’t show up and wasting ratepayer money, or failing to build ahead of where they do show up and then wasting precious time trying to catch up.

That’s where the good news comes in. The same software that Kevala used to forecast the grid load to come from electrifying California’s vehicles can also be used to forecast where and when the grid will need to be upgraded to support that load growth, Shumavon said.

“Kevala’s platform allows us to look simultaneously at the econometrics of adoption and behavior, and power flow, in a massively parallelized, scalable cloud-computing environment,” he said.

That’s a fancy way of saying that Kevala’s software uses modern distributed computing methods to create a model of the grid that measures the interplay of electricity flowing from substations all the way to every customer address. That model is built on more than 100 terabytes of data on those customers, much of it collected by utilities through smart meters and other systems. The California Public Utilities Commission, which commissioned the study, required the utilities to provide the data to Kevala.

That data includes electricity consumption patterns that show when customers are using the most power on an hourly and seasonal basis. It also includes demographic and purchasing-history data collected by Kevala from public and private sources, which it uses to predict at what point in the future customers will be most likely to invest in solar panels, backup batteries, electric heat pumps — and EVs.

This ​“bottom-up” approach to load forecasting and grid planning is radically different from the ​“top-down” approaches in common use today, which rely on averages and estimates to project where these kinds of new distributed energy resources are going to pop up on the grid.

Kevala is one of a small but growing number of companies building software meant to tap the latest advances in cloud computing and data science to bring utilities up to speed so they can deal with the challenges of managing a grid awash in these distributed energy assets.

The standard technologies that utilities have historically used ​“are not adequate to solve the challenges we’re facing,” Shumavon said. ​“Our approach was to digitize everything and solve problems on the planning horizon as fast as they’re happening — instantaneously.”

Those are bold claims, to be sure. Kevala’s new report — the Electrification Impacts Study Part 1 — is just the first installment in a broader scope of work the company is doing for the CPUC, and it’s not yet being considered for use to actively plan grid investments.

But Kevala is building a roster of customers and partners interested in putting its hyper-granular modeling and forecasting to use, such as utilities National Grid and Exelon and the U.S. Department of Energy’s Interconnection Innovation e-Xchange (i2X).

Meanwhile, its work for the CPUC is part of a yearslong effort, led by consultancy Verdant Associates in partnership with Gridworks and Xanthas Consulting, to help the state integrate rooftop solar, EVs, batteries, electric appliances and other distributed energy resources into its aggressive decarbonization plans.

The CPUC has ordered the state’s big three utilities to respond to the findings in Kevala’s report by early June. Shumavon highlighted some key recommendations for utilities, including getting far more granular in their grid analyses and looking further into the future to plan their distribution-grid investments.

“Typical distribution planning needs to change now, not because of what we need in three to five years” — today’s standard forecast horizon for distribution-grid planning — ​“but because in 10 years we’ll need all these [distributed energy resources] because the system will be at its maximum point of stress,” he said. If utilities and regulators wait for that moment of stress to arrive before taking action, he said, it will be too late.

The big picture on the grid and EVs in California

California policy requires all new cars and light trucks sold in the state to be electric or plug-in hybrid electric models by 2035, at which point the state expects to have 13.5 million EVs on the road, compared to about 1.5 million today. Its recently passed Advanced Clean Fleets rule sets a 2036 deadline for all new commercial trucks sold in the state to be zero-emissions and a 2042 deadline for all commercial trucks on the road to be emissions-free.

These EV mandates, the country’s most aggressive, will add significant strains to the power grid. Just where and when those strains will manifest is hard to predict, however. Some sites, like truck stops or fleet depots, are obvious. A single EV fast-charging site capable of charging four vehicles can require about $150,000 in grid upgrades. But much of the EV demand will be emerging in more random patterns.

Estimates of how much utilities will have to spend to support this growth range widely. A 2019 report from Boston Consulting Group forecasts that a typical utility will need to spend an average of $1,700 to $5,800 on grid upgrades for every EV in its territory, which for California’s 13.5 million EVs would equate to between $23 billion and $78 billion. California regulators clearly need a more precise set of figures to work with.

California’s three big utilities are planning billions of dollars of grid investment over the coming years, including a combined $2 billion to make the grid ready for EV charging. But until recently, the state’s Integrated Energy Resource Plan process, which provides the load forecasts that go into grid planning, was using outdated forecasts of no more than 4.4 million EVs on the road by 2030, compared to the nearly 8 million EVs that the California Energy Commission now expects to be on the road by then.

Even when more up-to-date EV forecasts are brought into account, standard grid-planning approaches can miss wide swaths of costs, Shumavon noted. For example, the Grid Needs Assessment studies conducted by California’s big three utilities analyze upgrade needs for the thousands of distribution substations, transformer banks and feeders they operate. But they don’t include the roughly 1.5 million service transformers — the can-shaped devices attached to power poles that convert power to voltages suitable for delivery to individual homes and businesses — because they’re too small and numerous to be easily accounted for, he said.

Nor do traditional grid-planning methods forecast load growth down to the individual customers, Shumavon said. That’s despite the fact that since the early 2010s, California’s three big utilities have had smart meters at customers’ homes and businesses that collect power-usage data in hourly or 15-minute intervals. That’s the data Kevala used to develop its address-level load forecasts.

But most utilities haven’t updated their grid-planning methods to make use of the latest advances in computing technology, said Stephan Barsun, partner at Verdant Associates, the analysis firm leading the CPUC project that Kevala is working on. That means utilities aren’t as good at analyzing their own data as they could be.

“The traditional way to do this is to do your forecasting at the feeder and substation level,” and limit the number of scenarios to best-case and worst-case, he said. Most load-forecasting software was developed before cloud computing made it possible to distribute enormous amounts of data processing across hundreds or thousands of servers.

It’s certainly faster and easier to run forecasts that limit the scope of data they’re crunching, Barsun said, and that’s an appropriate way to do some types of planning: ​“There’s only so much cloud computing you can throw at a problem.” But for more complex challenges, like how to plan ahead for EVs cropping up in unpredictable places, Kevala’s ​“bottom-up approach has a lot of really attractive nuances and detail,” he said.

What utilities can get from granular and timely data 

Kevala Vice President Trina Horner, a former regulatory and rate-design executive at PG&E, highlighted the value that Kevala provides by analyzing the data for each of the roughly 12 million customers of California’s three big utilities.

“We’ve individually modeled baseline load growth — a load shape for every single one of those premises,” she said. ​“And we’ve identified all of the [distributed energy resource] adoption and behavior for each of those premises, and we’ve layered that…adoption and behavior on top of each of those forecasts” to predict where EVs will be purchased and how they’ll be charged.

Kevala is also accessing much more up-to-date data than utilities typically use in their distribution-planning processes, she noted. Traditionally, those processes use data that’s relatively stale — ​“literally years old,” she said.

Kevala, by contrast, is able to process data as quickly as utilities can make it available from their smart-meter data-management systems and their grid-operations and mapping systems, she said. That data informs maps like this one, which shows where discrete points on the grid are experiencing undervoltage or overvoltage conditions that can result in power-quality disruptions or damage to sensitive equipment.

This combination of timely and granular data has yielded the aforementioned bad news about the huge costs awaiting California’s grid, which are ​“essentially an order of magnitude larger” than previously predicted, Horner said. But by determining just where and when individual grid circuits and transformers are most likely to face potential overloads from EVs, Kevala’s software has also given utilities and regulators tools to do something about it.

“Yes, this is an eye-popping set of numbers we need to pay attention to,” she said. ​“But the real takeaway from the study is that we can refine and update our assumptions, and look at mitigation measures” at the precise grid circuits where overloads are predicted to happen first.

“You really need to think about the localized requirements that will come up, which are going to be different, based on who’s there, and what the weather is like, and what those [EV] adoption propensities are,” she said. ​“You have to make the right distribution capacity investments at the right time.”

Those decisions need to start happening in advance of customers installing EV-charging equipment, Shumavon added. ​“We hear right now from EV developers” across the country ​“that they’re not getting information from utilities fast enough to know where to site infrastructure,” he said. ​“Sometimes they’re choosing sites and being told, ​‘Come back in three to five years.’”

Suncheth Bhat, chief business officer at startup EV Realty, agreed that finding locations with enough grid capacity is a challenge. EV Realty uses proprietary software and data sets to develop EV-charging hubs that can avoid the need for utility grid upgrades that can add months or years to wait times and add hundreds of thousands or millions of dollars to costs.

“The grid was never designed to be the fuel source for transportation,” said Bhat, who previously spent 16 years at PG&E, most recently as head of the utility’s clean transportation division. ​“Infrastructure takes a lot of time — engineering, designing, permitting, construction,” and ​“the utility has to take each request in the order in which it was received.”

Coordinating how utilities plan ahead for upgrading the grid with how their customers are predicted to adopt EVs could spell the difference between a grid-investment program that keeps costs within bounds and one that breaks the bank. A recent study in New York indicates that the money utilities can bring in from selling electricity to charge EVs can counterbalance the cost of making the grid ready for that EV charging.

But utilities and regulators that can’t accurately predict where EV growth will create the need for preemptive grid buildouts face a conundrum, Bhat said. Overbuilding the grid creates what’s known in the industry as ​“a stranded asset,” he said, and protecting utility customers at large from paying for stranded assets is a big part of regulators’ jobs.

“That said, to hit our goals” for EV adoption, ​“we can’t take a business-as-usual approach,” Bhat said. ​“We need to look at the entirety of the process, all the dimensions.” 

A 21^st-century computing and planning regime for utilities and regulators

One of the first tasks for utilities and regulators trying to tackle this problem, Shumavon suggested, could be to engage on the IT challenge of making everyday use of the data that Kevala spent nearly two years collecting, cleaning, integrating and analyzing for this report.

The broader scope of work for which the CPUC has hired Verdant, Kevala and their other partners will include a review of how utilities could use the kind of disaggregated load forecasts in Kevala’s new report to ​“develop and implement a framework for estimating specific, highly granular grid investment,” the report notes.

There are significant risks involved in moving to a new method of planning utility investments, starting with questions about confidence in the data the new method is based on. At the same time, ​“there is some low-hanging fruit” in terms of changes that California can make to prepare for the rising grid demands to come from EV adoption, Horner said. In the next phase of the Electrification Impacts Study process, Kevala proposes to use its system to evaluate much more up-to-date data in more dynamic scenarios, to keep up with changes in customer behavior and technologies as they electrify, she said.

Utilities and regulators can also use the granular customer-level data Kevala has collected to fine-tune the ​“managed EV charging” programs and rate structures seen as vital for encouraging EV owners to shift when they charge to times when the grid at large is under less stress.

Understanding how customers respond to different prices and incentives can play a big role in determining how much grid-upgrade investment is required to support their charging loads, since utility grid costs are closely tied to the maximum demands set during peak hours.

Being able to tie that data to individual locations on the grid could make it even more useful, Verdant’s Barsun noted. Kevala’s new report ​“really does a good job of setting the groundwork and the baseline,” he said. ​“If we as a state do not start putting in some mitigation measures, things are going to get really expensive. Even with mitigation measures, there are some investments that need to be made, and those are going to need to start to be planned for sooner rather than later.”