State of Charge Estimation of Lithium-Ion Batteries Based on Extended-Kalmannet

Abstract

Expanded Kalman Filtering (EKF) is a widely utilized technique in the field of lithium-ion battery charge state estimation. However, inadequate knowledge of the model is able to result in significant performance degradation of the EKF. To address this issue, this paper proposes a SOC estimation method based on Extended-KalmanNet, which provides an accurate estimation of the state of charge even in the absence of sufficient knowledge of the model. The method uses a Recurrent Neural Network (RNN) with an internal storage unit to learn the Kalman gain (KG) from the data. By learning the KG, Extended-KalmanNet circumvents the dependency of the KF on knowledge of the underlying noise statistics, thus bypassing numerically problematic matrix inversions involved in the KF equations. And the hidden state of the internal storage unit adapts to the output of the RNN as it is used. Consequently, the method is able to perform accurate SOC estimation in the presence of model mismatch. Simulation results show that Extended-KalmanNet is superior to the classical EKF algorithm and is able to perform accurate SOC estimation under conditions of insufficient knowledge of the noise distribution or mismatched measurement models.

Keywords: Extended Kalman filter, state of charge estimation, model-driven, neural network.