A dynamic artificial neural network approach to estimate thermal behaviors of li-ion batteries

Abstract

Li-ion batteries as secondary cells have important role in transfer energy in the technology world. These storage batteries have nonlinear behaviors against increasing their temperature. Therefore, thermal effect as one of important parameters is considered for modeling these accumulators. An artificial neural network is subset of computational intelligence approaches method that can identify a wide range of complex systems with the highest accuracy. In addition, this approach does not require much knowledge from the systems and is used as a black box model. In this paper, a dynamic artificial neural network approach based on Nonlinear Autoregressive eXogenous (NARX) model to analyze thermal behaviors of a li-ion battery is presented. The proposed network is trained by Levenberg-Marquardt backpropagation algorithm. In this method, discharge-load voltage and discharge-measured temperature of a li-ion is considered as input and output of the network respectively. The input and output data are provided from an experimental test that was carried out by one of the research centers of NASA. Simulation results show that the proposed approach is high accuracy to model the battery with MSE 2.032×10−2, 5.438×10−1 and 2.155×10−1 in order for training, validation, and test data. Also, regression R is measured the correlation between all outputs and targets about 9.674×10−1.