State of Charge Estimation for Lithium-ion Battery using Recurrent Neural Network

In this paper combines the artificial neural network (ANN) method and the internal resistance measuring method, which is designed for the lithium-ion battery state of charge (SOC) estimation. It is different from the general neural network research that only uses voltage and current as parameters. The input layer adds important parameters: the battery voltage and current, and the internal resistance of the battery as external inputs. We design a recurrent neural networks (RNN) model with non-linear autoregressive with exogenous input (NARX). The network compares the difference between the simulation results under the same benchmark with the back-propagation neural networks (BPNN). Experiments show that this architecture not only improves the convergence speed of the neural network, but also shortens its average execution time, and the mean-square error is improved. It is a good indicator of the accuracy of the measurement. This paper also discusses the application of this architecture to the difference between DC internal resistance and AC internal resistance.