Battery Health Prognosis Based on Sliding Window Sampling of Charging Curves and Independently Recurrent Neural Network

With the development of lithium-ion battery (LIB) technology and the increasing popularity of electric vehicles, the issue of battery safety has become increasingly urgent. The state of health (SOH), known as a critical parameter in the prognosis and health management of LIBs, has considerable attention from industry and academia. This article proposes a novel method for estimating the SOH of LIBs based on sliding window (SW) sampling of charging curves and independently recurrent neural network (IndRNN). Considering the number of battery cycles and practical applications, the SW sampling based on cycle number is utilized to determine the different partial voltages as the inputs to the SOH estimation model. To address the gradient disappearance and gradient explosion problems, in the proposed SOH estimation model, we suggest the IndRNN which introduces independent weights between inputs and outputs, trains the IndRNN with rectified linear units, and learns the long-term dependencies by stacking multiple layers of IndRNN to achieve long-term accurate aging tracking of batteries. Finally, experiments are validated on the most widely used Oxford University battery dataset, and the effectiveness of our method is also verified by comparing it against three methods on our laboratory data with different operating conditions.