Evolving Elman neural networks based state-of-health estimation for satellite lithium-ion batteries

With the rapid development of aerospace industry, accurate and timely state-of-health (SOH) estimation by in -orbit measurable parameters is critical for the safe and reliable operation of satellite lithium-ion batteries. In this paper, a novel SOH estimation strategy is proposed based on the k-means clustering analysis (KMCA) and the evolving Elman neural network (EENN) by using the in-orbit discharging voltage data. The time series of equal discharging voltage difference (TSEDVD) is firstly developed as the feature to describe the capacity degradation in each charge-and-discharge cycle. Then, the KMCA is used offline to rank the capacity fading of lithium-ion battery into different health levels by virtue of the TSEDVD feature derived from the historical dataset throughout the battery lifecycle. Furthermore, a set of EENN models are constructed offline corresponding to the health levels and employed in estimating the SOH value online, where the improved evolving algorithm combining the genetic and simulated annealing algorithms is proposed to optimize the initial weights and thresholds in order to get the accurate estimation results. For the in-orbit application, the SOH value can be easily estimated online only via the monitored voltage signal exciting the corresponding EENN. The experimental results using the datasets from the NASA Ames Prognostics Center and the real GEO communication satellite demonstrate the validity of the proposed approach. It is feasible for the in-orbit SOH estimation of satellite lithium-ion batteries.