Abstract:

Since energy storage systems（ESSs） are widely used in electric vehicles，distributed power generation and other fields，the reliability of batteries has become an issue of concern for researchers. The state of charge（SOC）is a crucial parameter reflecting the battery endurance. This study proposes a new method for lithium-ion battery SOC estimation to keep the stable working of an ESS. Recurrent cerebellar model neural network（RCMNN）and Kalman filter（KF）are both applied in the estimation method. Integrating recurrent units in associative memory space and weight memory space can improve the RCMNN in dynamic feature capture. The RCMNN and KF with inputs of voltage，current and temperature can simulate the charging and discharging of the ESS. Considering the complexity of discharging process of the battery， the measured and simulated values of the SOC under different charging and discharging conditions with varied SOC initial values are compared. The results show that the proposed method has decent accuracy and robustness under different conditions.

Key words: state of charge, lithium-ion battery, energy storage system, recurrent cerebellar model neural network, Kalman filter