Novel Machine Learning Approach Predicts Supercapacitor Cycle Life with High Accuracy

Strategic Insights -

A groundbreaking study has been conducted by researchers at Southeast University in Nanjing, People's Republic of China, to predict the cycle life of supercapacitors using machine learning techniques. The study, funded by the National Key R & D Program of China, proposes a novel feature-enhanced framework that accurately predicts cycle life using only limited early-cycle data. This approach has the potential to significantly reduce data requirements and deployment cost, providing an efficient solution for online assessment and second-life utilization of supercapacitors.

Key Takeaways:

  • The study proposes a novel feature-enhanced framework that integrates three complementary feature classes to predict cycle life using limited early-cycle data.
  • The framework employs a model-based recursive feature elimination process to distill the most predictive features, capturing the EDLCs' characteristic fast-to-slow degradation pattern.
  • The resulting gradient boosting regression model achieves a 4.73% MAPE in cycle life prediction, while the logistic regression-based classification model reaches a classification accuracy of 99.76%.
  • The method significantly reduces data requirements and deployment cost, providing an efficient solution for online assessment and second-life utilization of supercapacitors.
  • The study was funded by the National Key R & D Program of China and was peer-reviewed.
  • Additional authors of the study include Zhen Xu, Hongjian Tang, and Lunbo Duan.
  • The research was conducted at the School of Energy and Environment, Key Lab Energy Thermal Convers & Control, Ministry of Education, Southeast University.

Statistics:

  • The framework integrates three complementary feature classes: static cycle endpoints, instantaneous SOC and DOD point values, and trend-based descriptors.
  • The gradient boosting regression model achieves a 4.73% MAPE in cycle life prediction using only data from the first 500 cycles.
  • The logistic regression-based classification model reaches a classification accuracy of 99.76% using only the initial 8 cycles.
  • The study was funded by the National Key R & D Program of China.

Sources:

  • Feature-enhanced Machine Learning Prediction of Supercapacitor Cycle Life From Limited Early-cycle Data. Electrochimica Acta, 2025;538.
  • NewsRx. Study Data from Southeast University Provide New Insights into Machine Learning (Feature-enhanced Machine Learning Prediction of Supercapacitor Cycle Life From Limited Early-cycle Data). Information Technology Newsweekly. October 21, 2025; p 905.