A team from Georgia Tech has developed an AI that can analyze user reviews of electric vehicle charging stations, allowing it accurately to identify places where there are insufficient or out-of-service stations. An open-access paper on the work is published in the journal Patterns.
We’re spending billions of both public and private dollars on electric vehicle infrastructure. But we really don’t have a good understanding of how well these investments are serving the public and public interest.—Omar Asensio, principal investigator and corresponding author
Electric vehicle drivers have started to solve the problem of uncertain charging infrastructure by forming communities on charge station locator apps, leaving reviews. The researchers sought to analyze these reviews to better understand the problems facing users.
With the aid of their AI, Asensio and colleagues were able to predict whether a specific station was functional on a particular day. They also found that micropolitan areas, where the population is between 10,000 and 50,000 people, may be underserved, with more frequent reports of station availability issues. These communities are mostly located in states in the West and Midwest, such as Oregon, Utah, South Dakota, and Nebraska, along with Hawaii.
Predicted discussion frequency of station availability for US metropolitan and micropolitan statistical areas. The map reveals areas with high and low discussion frequency for predicted Availability issues in all metropolitan statistical areas (e.g., population greater than 50,000). Micropolitan statistical areas (e.g., population 10,000–49,999) have higher Availability discussions in some states in the West and Midwest regions. The algorithms predict that many micropolitan statistical areas could be underserved with regard to station availability. Ha et al.
Compared to analyzing data tables, texts can be challenging for computers to process. A review could be as short as three words or as long as 25 or 30 words with misspellings and multiple topics, noted co-author Sameer Dharur. Users sometimes even throw smiley faces or emojis into the texts.
To address the problem, Asensio and his team tailored their algorithm to electric vehicle transportation lingo. They trained it with reviews from 12,720 US charging stations to classify reviews into eight different categories: functionality, availability, cost, location, dealership, user interaction, service time, and range anxiety.
The AI achieved a 91% accuracy and high learning efficiency in parsing the reviews in minutes.
As opposed to previous charging infrastructure performance evaluation studies that rely on costly and infrequent self-reported surveys, AI can reduce research costs while providing real-time standardized data.
The electric vehicle charging market is expected to grow to $27.6 billion by 2027. The new method can give insight into consumers’ behavior, enabling rapid policy analysis and making infrastructure management easier for the government and companies. For example, the team’s findings suggest that it may be more effective to subsidize infrastructure development as opposed to the sale of an electric car.
While the technology still faces some limitations—such as the need to reduce requirements for computer processing power—before rolling out large-scale implementation to the electric vehicle charging market, Asensio and his team hope that as the science progresses, their research can open doors to more in-depth studies about social equity on top of meeting consumer needs.
This work was supported by the National Science Foundation, Microsoft Azure Sponsorship, and the Ivan Allen College Dean’s SGR-C Award.
Sooji Ha, Daniel J. Marchetto, Sameer Dharur, Omar I. Asensio (2021) “Topic classification of electric vehicle consumer experiences with transformer-based deep learning,” Patterns doi: 10.1016/j.patter.2020.100195