As battery prices have fallen and the pressures of meeting reduced-emission standards have influenced the production decisions of auto makers (17 auto manufacturers now make electric vehicles), plug-in electric vehicle adoption has steadily increased. In fact, so much so that The Edison Foundation’s Institute for Electric Innovation (IEI) projects plug-in electric vehicle sales to exceed 1.2 million per year by 2025, or more than 7% of total annual vehicle sales.
One of the big questions surrounding EV adoption is how electric utilities can enable and help accelerate the adoption. From ensuring the grid can handle the demand, to developing programs and pricing that allow customers to better manage their electric bills, to reducing range anxiety, utilities are positioned to be a key enabler of plug-in EV adoption.
Interestingly enough, the central challenge won’t be meeting demand at the macro level. Generating more electricity for the relatively modest needs of EVs isn’t going to be a problem. Instead the devil will be in the details of managing demand at the local level – and by local, I mean really local, as in at the transformer level.
Getting to that level of granularity will require a much deeper understanding of the propensities and behaviors of individual customers. And for electric utilities, as it is for many other industries, this is a big problem, because predicting individual customer behavior is not something they are good at.
At least not yet.
A Bit of a “Cluster”: Transformer-level Forecasting
Why is the transformer so important to EV-related forecasting? It’s simple. A transformer, which typically serves the power needs of 5-8 houses, wasn’t designed for the demands of EV-charging. While a single EV is manageable, EV adoption tends to happen in clumps. The Electric Power Research Institute (EPRI) refers to this phenomenon as “birds of a feather flock together,” while the National Renewable Energy Laboratory (NREL) calls it a “clustering effect.” Simply put, adoption rates will be higher in that aforementioned band of 5-8 houses if one of the neighbors there has an EV or a house has multiple vehicles.
You can see the consequences. When four or five vehicles suddenly need to charge at the same time, there is a potential for problems. In order to get ahead of all this, electric utilities are looking for ways to improve their micro-forecasting abilities, i.e., the ability to plan and manage the distribution grid at a very granular level, and predictive data science is stepping forward to help.
Yet we’re finding from first-hand experience that the answer to the EV-forecasting challenge isn’t one dimensional. Electric utilities not only need better predictive models, they need better data to feed them – especially when the goal is to predict the behaviors of individual customers.
The Power of One
So, electric utilities have the unenviable task not only of forecasting overall EV uptake, but of predicting with precision who is going to be buying an EV, when they will be buying one, and where they will be charging it.
TROVE was presented with exactly this Proof of Concept challenge by a Western Utility. Without revealing too many secrets, the Proof of Concept illustrated two particular strengths that are putting TROVE at the epicenter of the electric-utility industry’s growing interest in EV forecasting: the ability to develop precision forecasts at the individual level, and the complementary ability to aggregate those forecasts up into a meaningful macro-forecast continually honed by machine learning.
One key reason TROVE is so adept at micro-forecasting is that we have a lot of experience processing years of individual-customer hourly AMI data and are able to combine that with the best consumer and commercial data in the industry. In this case, we enriched the utility’s data with an additional 659 attributes – filling in gaps essential to making accurate predictions. We further supplemented this data with data on commuting, and other industry data, to start identifying and understanding the defining characteristics of consumers who buy EVs and, just as importantly, of those who don’t.
We call this approach The Power of One, the ability to predict individual behavior with increasing precision, and have leveraged it successfully at other electric utilities. For example, we currently run micro-forecasts on 500,000 individual demand response customers daily for one of the largest electric utilities in the country, rolling these up into macro-event forecasts that have boosted accuracy by up to 40 percent.
In the EV forecasting Proof of Concept, we fed the enriched data into a variety of propensity models and simulations, an act our data science team, the Science Squad, characterized as “a multi-component analytic approach,” but which I prefer out of deference to call “true data science.” Drawing on many different modeling techniques, our team worked diligently and creatively to match the right methods to the right set of problems, identifying types of buyers, geographic hot spots, and best-and-worst case scenarios over a 1-5-year time horizon. In so doing the team essentially created both the playbook for addressing this potential issue and a powerful new planning tool for managing it over time.
Today, as a result of this Proof of Concept, we can now provide “intuitively realistic projections that can be used to guide strategic business decisions” related to EVs and the distribution grid, including decisions related to grid maintenance and investment; non-wires alternatives to managing EV demand that can help customers manage their bill; and the placement of Level 3 charging stations based on commuting routes and distances. And because the models learn as they go, the forecasts are continually sharpened by the results they help generate.
Put another way, the Proof of Concept has proven to be a big success, changing the way the utility is thinking about its ability to enable EV adoption.