Andrew Corselli

The researchers found that streetlight charging stations, compared to traditional EV charging stations, were more cost- and time-effective, had fewer negative environmental impacts, and were more convenient and accessible. (Image: Provided by XB Hu/Penn State. All Rights Reserved.)

Electric vehicles (EVs) can have lower fuel costs and reduce emissions relative to cars that use gasoline, but they are only a practical option if drivers have convenient ways to charge them. For people who live in multi-unit dwellings or in urban areas, access to charging infrastructure may be particularly limited, which in turn limits EV adoption.

To address this issue, a team of researchers at Penn State created a scalable framework to develop, analyze, and evaluate using streetlights as a low-cost, equitable EV charging option. They then installed 23 streetlight charging units in Kansas City, MO, and tested their framework. The researchers found that streetlight charging stations, compared to traditional EV charging stations, were more cost- and time-effective, had fewer negative environmental impacts, and were more convenient and accessible.

Their results were published in the Journal of Urban Planning and Development, which is overseen by the American Society of Civil Engineers.

“The motivation for this work comes from the fact that many apartment and multi-unit dwelling residents, particularly in urban and downtown areas, lack access to dedicated home EV chargers, since they don’t have the privilege of owning a garage,” said Xianbiao “XB” Hu, Associate Professor of Civil and Environmental Engineering. “Fortunately, streetlight poles are already powered and typically owned by municipalities, making them relatively easy to work with. Their placement — often near on-street parking and in high-traffic areas — makes them well-positioned to serve both local residents and visitors.”

Here is an exclusive Tech Briefs interview, edited for length and clarity, with XB.

Tech Briefs: What was the biggest technical challenge you faced while developing these streetlight chargers?

XB: There were many challenges we had to overcome, but I think, for me, the biggest was to predict the charging demand. Our goal is to invest money at a location; to install a charger at the right location, and after we install it, we hope people will come to use it and use it in a way that fits with our expectations. I think that was the most challenging portion.

The reason we started this project was as part of a Department of Energy project funded in 2018. When we started this project, EV chargers were not as widely available as today, so we didn’t have a lot of data for us to build the machine learning models to predict which site would be more popular than others.

If you look at all the streetlights in Kansas City, MO — I think we have probably over 10,000 streetlights that we could have used, and we had a limited budget. In the end, we were able to reach no more than 30. So, probably selecting 30 out of 10,000 streetlights was the most challenging part.

Tech Briefs: Can you explain in simple terms how it works? Is it just like a traditional EV charger, just in streetlight form?

XB: Exactly. The entire process involved a lot of coordination, so that was more complicated. But, if we are only looking at the technical details, we settled on an EV charger that is common in the market.

Before we reached that conclusion, we did a lot of surveying, and then we looked into the streetlights. There were several questions that we had to ask ourselves. The first question was: What are the requirements for the streetlight posts? Like what are the voltage requirements? Also, proximity to apartments and the utility transformer, those kinds of things.

The second thing was we needed to look at the different kinds of EV chargers on the market, to decide what to select. And then the third hurdle was community engagement; some people liked the idea, and some people said, ‘Just don’t bring any strangers in front of my house.’

Tech Briefs: Do you have any set plans for further research/work/etc.? If not, what are your next steps?

XB: Research-wise, we are doing quite a few things. One of my Ph.D. students is working on the weather impacts on EV charging, specifically to determine what would happen if the temperature drops below freezing or rises above 100 degrees. How would that impact people’s behavior? Would that encourage high usage of the electric vehicle vs. a gasoline vehicle? Would they charge at a different location? And, in the end, should the state utility company install the chargers differently? So, that’s one area of research we’re currently working on.

Also, socioeconomic data. Meaning, who is more likely, when they purchase their next vehicle, to purchase an electric vehicle. And in what communities might we see a fast EV ownership increase.

Tech Briefs: The article I read says, “To determine demand, the researchers looked at different factors … and then used the data to train AI models to make demand predictions based on those factors.” My question is: Can you talk about the training-AI-with-data part?

XB: The objective as I mentioned at the beginning of this interview — demand prediction, identifying 30 out of possibly 10,000 potential site candidates — is to make sure that if we build it, people will come to charge. This was not easy; there were many factors.

To do this, we built machine learning models to predict future demand. To train a machine learning model, we needed a lot of data. We obtained a total of 10 years of charging data from over 400 public chargers in Kansas City, MO. Each time you charge, and plug in the cord, the EV chargerß records your charging data, including your vehicle ID, when you plugged inthe charging cord, when the charge started, when the charge ended, how much time you spent there, how much energy was transferred into the vehicle, how much you paid — those kinds of things.

So, on one hand we have 10 years of charging patterns, charging usage data. On the other hand, we were able to gather the values for independent variables such as traffic volume. We used that 10 years of data to train our machine learning model.