State of Health Estimation of Lithium-ion Batteries Based on Data-Driven Techniques

Abstract

Accurate prediction of the state of health of lithium-ion batteries is essential to ensure safe and reliable operation with minimum maintenance cost. However, estimating the state of health of lithium-ion batteries is a challenge due to the complex and nonlinear characteristics of battery degradation during its lifetime. To this end, data-driven techniques are becoming a more attractive promising approach to provide accurate state of health estimation of the lithium-ion battery because of their simplicity and ability to cope with the nonlinearity dynamic behaviour of the lithium-ion battery. Therefore, this paper compares the performance of three data-driven algorithms: the nonlinear autoregressive neural network, convolutional neural network, and long short-term memory network in the state of health estimation. The performance of the proposed algorithms is analysed by using 16 lithium-ion cells cycled under various operating conditions from the NASA Ames Research center. The comparison results demonstrate that the long short-term memory network outperforms the other two methods with a maximum root mean square error of 0.50% and mean absolute error of 0.36%.