State of Health Estimation of Lithium-Ion Batteries Based on Multiphysics Features and CNN-EFC-BiLSTM
Chaolong Zhang, Yujie Zhou, Ziheng Zhou, Shi Chen, Ji Wu, Liping Chen
Abstract
Electric vehicles (EVs) can reduce reliance on finite resources, such as oil, aiding in the reduction of carbon emissions, while an accurate battery state of health (SOH) helps to ensure the safety of EVs. Because EVs generally have a nonzero battery state of charge (SOC) at the start of charging, obtaining complete charging data for the power battery is challenging. Therefore, a method for estimating the SOH of lithium-ion batteries based on multiphysics features and convolutional neural network-enhanced feature combination-bidirectional long short-term memory (CNN-EFC-BiLSTM) is proposed in this article. First, various data of the battery during the constant-voltage charging phase are measured by the sensors of the battery testing system, and the analysis of battery temperature, current, time, and energy data during the phase is conducted. Multiphysics features, including the average charging temperature, length of the current trajectory, and incremental energy, which are highly correlated with the battery SOH, are extracted from the measured data. A CNN-EFC-BiLSTM model is proposed to map features to battery SOH and establish an SOH estimation model through training. Experiments were conducted with batteries at different charging currents, and the results indicated that even with significant nonlinearity during battery SOH degradation, the method can achieve a rapid and accurate estimation of SOH. The maximum mean absolute error (MAE) is only 0.3219%, the maximum root-mean-square error (RMSE) is only 0.4723%, and the average training time is reduced by approximately 32%.