State estimation of Li-ion batteries via improved second-order RC model
Yibo Qi, Wei Zhao, Xiaofeng Sun, Lei Qi, Dong-Hui Wang, Yuliang Zhang
Abstract
This study introduces a controlled voltage source trained by a backpropagation (BP) neural network into the conventional second-order RC model. This design enables real-time compensation for terminal voltage errors caused by temperature variations and different discharge rates, significantly enhancing the model's adaptability under diverse operating conditions. Based on the improved circuit model, the Extended Kalman Filter (EKF) is employed for State of Charge (SOC) estimation. Subsequently, State of Power (SOP) estimation is jointly performed using the improved terminal voltage, the estimated SOC, and the battery's maximum discharge current. Simulation results demonstrate that, compared with traditional models, the proposed method improves the accuracy of terminal voltage prediction, thereby significantly reducing state estimation errors and more accurately reflecting the actual battery behavior across a wide temperature range and under dynamic load conditions.