An electric vehicle power battery SOH prediction method based on CEEMDAN-LSBoostELM

As the core component of electric vehicles, the SOH prediction of power battery is of great significance to the safety of the whole vehicle operation and battery life. Aiming at the problem of low prediction accuracy of traditional methods under small sample conditions, this paper proposes a CEEMDAN-LSBoostELM prediction model that integrates fully adaptive ensemble empirical modal decomposition of noise (CEEMDNA), least squares boosting (LSBoost), and extremum learning machine (ELM) based on real-vehicle operation data. First, the aging features are extracted based on the capacity incremental analysis, and the features that are highly correlated with SOH are screened by gray correlation analysis as model inputs; second, the SOH sequences are decomposed by using CEEMDAN to extend the feature dimensions; and finally, LSBoost is introduced to improve the stability and prediction performance of the ELM model. The experimental results show that the prediction accuracy and robustness of the proposed model on the real-vehicle dataset are significantly better than those of the CEEMDAN-ELM, ELM, RF, and MLP models, and the prediction errors are no more than 1.2%, which verifies the effectiveness of the feature enhancement and model integration strategy.