Influence of the Discretization Methods for the Model of Lithium-Ion Battery

An accurate battery model is significant for the management and safety of lithium-ion battery. Many methods have been proposed for battery modeling, and the equivalent circuit models are the most widely applied in practice. The construction of the equivalent circuit models require to be discrete, and few researches have focused the influence of discretization methods. In this work, a novel parameter identification method is proposed, which is based on the back propagation neural network (BPNN) and gradient descent method. The model parameters is employed to define the weights of the BPNN, and the gradient descent algorithm is adopted to optimize them. Compared with extended kalman filter (EKF)- based method, the proposed method can improve the model accuracy by 0.27%. Based on the BP-based method, different discretization methods are compared to explore their influence, including the backward difference method and zero-order holding method. The result shows that zero-order holding method is more reliable than the backward difference method according to the identified internal resistance.