Duke Energy Collaboration (Fall 2024, Spring 2025)
Students in the MS in Applied Data Science program explored methods to predict the reference discharge capacity of lithium-ion batteries without relying on their cycle history. Battery technology plays a critical role in modern energy storage solutions and powers everything from electric vehicles to energy grids. A crucial performance metric for battery cells is their capacity—the amount of energy they can store and deliver over time. Yet, as batteries undergo continuous cycles of charging and discharging, their capacity declines. This degradation is influenced by various factors including the number of cycles a battery experiences, the temperature conditions during operation, and specific charging protocols. The accurate prediction of battery capacity, especially when dealing with missing data and inconsistent charge/discharge cycles, is essential for optimizing battery performance and predicting lifespan. In many real-world datasets, critical measurements such as capacity are often missing, particularly for cells that were partially charged or discharged. This missing data poses a significant challenge when attempting to model battery behavior over time. In this project, students aimed to address two key objectives: (1) predict the capacity of fully charged or discharged battery cells by utilizing multiple battery sources and (2) explore the potential for predicting capacity degradation in cells that were only partially charged or discharged.
The research in this project utilized open-source data and the final project presentation can be found below.
