Building a better lithium-ion battery management system

Lithium-ion batteries serve as a cornerstone of renewable energy and electric vehicle markets worldwide, increasing the need for effective battery management systems. Conventional models use mostly voltage, current, and temperature measurements that struggle to efficiently and accurately measure key quality indicators, especially under high current rates and varying conditions. This can compromise battery performance and presents significant safety issues.

Olugbade and Park introduced a new system that enhances predictive accuracy by improving on a traditional single particle (SP) model while combining it with Extreme Gradient Boosting (XGBoost) machine learning (ML).

“We specifically improved the SP model, a simpler battery model that is commonly used because it is computationally efficient,” said author Emmanuel Olugbade. “And then we integrated a ML technique known as XGBoost, which makes this model much more accurate.”

After integrating the ML technique into the SP model framework, the researchers compared their results with those from a full-order electrochemical model and a conventional SP model. Also, using simulations software, they conducted dynamic stress tests and galvanostatic constant discharge tests under demanding conditions.

“Our model significantly improves predictive accuracy, particularly in these challenging scenarios,” said Olugbade. “At the same time, it remains computationally efficient, meaning it can run quickly and handle real-time predictions without requiring extensive computational resources. Additionally, because the model can capture complex behaviors of the electrolyte within the battery, its predictions align more closely with real-world data.”

The researchers are hopeful the new model will help pave the way for further advancements in battery modeling, control and management, and help serve the growing demands around EVs and grid-scale energy storage.

Source: “Boosting predictive accuracy of single particle models for lithium-ion batteries using machine learning,” by Emmanuel Olugbade and Jonghyun Park, Applied Physics Letters (2024). The article can be accessed at https://doi.org/10.1063/5.0230376 .