Researchers Use AI to Optimize EV Charging Scheduling
At a parking lot designed to test charging equipment, researchers are training an AI algorithm to optimize EV charging based on employees’ arrival and departure times.
Researchers from Illinois-based Argonne National Laboratory (ANL) and the University of Chicago have partnered on a project using reinforcement learning to train an artificial intelligence (AI) algorithm to schedule and manage electric vehicle charging.
Argonne National Laboratory’s Smart Energy Plaza is a testing ground for electric vehicle charging technologies. Image used courtesy of ANL
The team is testing and refining the model at ANL’s Smart Energy Plaza, a hub for EV charging research. The former gas station was renovated to support transportation and energy projects studying grid integration, charging equipment, and interoperability testing. The plaza hosts multiple stations with 12 Level 2 alternating current chargers and four direct current (DC) fast chargers with 50 kilowatts (kW), 200 kW, and 350 kW of power. Two 350-kW DC chargers connect to a 1-megawatt battery for high-power grid integration projects. The site’s integrated communication and control system helps manage various EV charging demands, building energy use, and on-site generation.
At Smart Energy Plaza, a group of students and ANL engineers are working to optimize their AI algorithm based on data from charging sessions with vehicles owned by ANL employees. The model is fed a series of computational rewards and punishments through a reinforcement learning process designed to adapt to a given employee’s arrival and departure time to/from work. At the same time, the model should also account for peak demand on the grid.
Argonne National Laboratory’s Smart Energy Plaza offers a range of energy and transportation technologies. Image used courtesy of ANL
When timed right, this scheduling advantage allows for efficient, low-cost charging. It also helps networked charging stations allocate charges to multiple vehicles hooked up simultaneously.
The project can simplify the charging experience for America’s growing base of electric car drivers. According to the U.S. Department of Energy’s Alternative Fuels Data Center, there were 2.4 million EV registrations in 2022, up from 1.4 million in 2021. Those vehicles have access to 54,787 public charging locations hosting 141,703 EV supply equipment (EVSE) ports across the country as of July 2023. Another 3,697 stations are private locations, totaling 14,683 EVSE ports.
Reinforcement Learning for EV Chargers
The algorithm learns from feedback from positive results, such as an EV receiving the optimal amount of charge at the set departure time. It also heeds negative results, such as needing to draw power past peak thresholds.
Jason Harper, a principal electrical engineer at ANL, stated that optimization is essential in smart charge scheduling, as stations always make tradeoffs to ensure each EV is charged efficiently. Harper added that smart charging also considers all parties in the ecosystem, including charging station owners, EV drivers, and utilities–all of which have different power demands and scheduling restrictions.
Harper and fellow researchers developed a mobile app, “EVrest,” allowing ANL employees to participate in smart charge scheduling and reserve spots at the facility’s networked charging stations. The application collects data on charging behavior to inform future AI models.
Salman Yousaf, an applied data science graduate student at the University of Chicago, said the reinforcement learning techniques could also benefit overnight at-home smart charging, which supports added flexibility in how the charging load is distributed.
The researchers said they plan to simulate a larger charging network to scale the data from ANL’s chargers with a real-world operation. They’ll use EVrest data to train models on smart charge management and vehicle-to-grid integration.