Online State of Health Estimation of Batteries under Varying Discharging Current Based on a Long Short Term Memory

Abstract

Accurate estimation of the State-of-Health (SOH) of a battery in a real-world operation is important for predicting its aging or anomaly status for the condition based maintenance as well as for the safety. Conventional approaches to the SOH estimation based on battery discharging characteristics, such as voltage and charge variations, deal mainly with the constant discharging current at individual cycles. However, it is clear that, in order to have the SOH estimation of a battery viable in real-world applications, the intra- and inter-cycle variation of discharging current should be taken into consideration. This paper shows that the battery SOH can be estimated accurately, with a sufficient generalization power, even under the varying intra- and inter-cycle discharging currents incurred by realtime payload variations. Specifically, first, we propose to represent the discharging characteristics of a battery by the four features: the entropies of the voltage and current distributions as well as the rated amounts of the total charge and the average current within a cycle. A sequence of these four feature values obtained along the progress of cycles are then input to a stacked LSTM for SOH estimation. Experiments are conducted based on CALCE datasets and the datasets collected under the intra-cycle time-varying discharging currents. The results indicate that the proposed method is able to obtain the accuracy of SOH estimation as high as, or even better than, that of the constant discharging current under varying discharging currents.