Li-ion battery is a strongly coupled nonlinear time-varying device and is difficult to establish its deterministic mathematical model for online parameter identification and state estimation for the performance management of battery. This paper proposes an AI model based on back-propagation neural network (BPNN). The training process is designed and the parameters of the network are determined referring to the electrochemical mechanism of the battery. The performances of BPNN model are compared with the mathematical model using the voltage response of step-discharge current of battery. The experiments verify that BPNN model not only predicts the V-I characteristics accurately, but also shows associative and migratory abilities.