ANFIS (adaptive neuro-fuzzy inference system) based online SOC (State of Charge) correction considering cell divergence for the EV (electric vehicle) traction batteries

• We propose a novel approach for online estimation and correction of the pack's SOC.
• The traditional Kalman filter based estimator is applied to get the “averaged SOC”.
• The ANFIS is used to online correct the “averaged” SOC estimation.
• Influence of cell imbalance on SOC estimation is embodied in the fuzzy rules of ANFIS.
• The method can get a more reasonable SOC value than the traditional estimators.

A common drawback of the SOC (State of Charge) estimators of EV (electric vehicle) traction batteries nowadays is that they don't consider the difference among individual cells and employ the “averaged SOC” as the state of charge of the pack. Over-charge or over-discharge may happen to the weak cells with this SOC value in vehicular applications. In this study, a novel approach for online pack SOC estimation and correction is proposed, which combines a traditional SOC estimator and an ANFIS (adaptive neuro-fuzzy inference system). The traditional KF (Kalman filtering) based estimator is applied to firstly estimate the “averaged SOC” of the battery pack, and the ANFIS is then used to online correct the “averaged” SOC estimation with the information of cell differences and loading current. The influence of cell differences on SOC estimation is embodied in the fuzzy rules of the ANFIS, which is trained offline. Validation results by experiments show that, the proposed method has the potential to overcome the drawbacks of traditional SOC estimators caused by cell-to-cell variations in a battery pack, and the corrected SOC is more reasonable than the traditional “averaged SOC”.