Hybrid approach for online capacity estimation of lithium-lon batteries: Integrating model-driven and data-driven methods
Xin Qiao, Zhixue Wang, Xianzheng Su, Gang Shen, Enguang Hou, Yinghao Cai
Abstract
With the widespread adoption of electric vehicles (EVs) and energy storage systems (ESS), estimating lithium-ion batteries (LIBs) capacity has become crucial for academia and industry. EVs typically operate under variable current and low temperature conditions, making capacity estimation challenging compared to ESS. This paper introduces a novel method integrating model-based and data-driven approaches for battery capacity estimation. A novel feature, unit state of charge (SOC) interval capacity, is extracted during SOC estimation using the model-based method. Continuous features are then employed in a deep convolutional neural network (DCNN) to estimate capacity. Using NASA and Oxford datasets, we achieve promising results with a root mean square error (RMSE) of 0.0039 and 0.0153, respectively, under room temperature and constant current conditions. The method’s effectiveness is validated using the low temperature and variable current datasets from NASA. Additionally, with fewer DCNN layers and neurons, the method is suitable for deployment on battery management system (BMS) with limited resources. • This paper introduces an innovative battery capacity estimation method that seamlessly integrates model-driven and data-driven methods. • A novel feature, namely unit SOC interval capacity, is extracted during the EKF based SOC estimation. • This hybrid method significantly streamlines the DCNN architecture and reduces the number of neurons, enhancing its practical implementation in BMS. • This hybrid method is adaptable to different application scenarios, such as variable current and low temperature environments.