Full-process electric vehicles battery state of health estimation based on Informer novel model

Abstract

Accurately estimating state of health (SOH) of the power battery, as the core component of electric vehicles (EVs), is of great significance to the safety of EVs and the sustainable development of energy. Given the difficulty in measuring the capacity of lithium-ion batteries (LIBs) during vehicle operation and the linearization of prediction caused by not considering the phenomenon of capacity regeneration, a novel SOH estimation framework based on Informer and the entire charging and discharging process is proposed, which can be applied to the entire life cycle of power batteries in actual complex scenarios. In this study, the charge-discharge capacity is reanalyzed based on complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) to obtain a nonlinear representation of the capacity. User behavior characteristics based on frequency and hourly meteorological data were extracted as health indicators (HIs) of LIBs. To integrate charge-discharge and meteorological data at different time scales, multivariate statistical data were resampled. The global dependency between HIs and SOH of the model input was learned by the Informer network. The results show that Informer networks outperform the cyclic structure-based model in prediction accuracy and robustness across different data distributions, and have great promise for state estimation.