Encrypted Dynamic RVFLN Ensemble Learning for Privacy‐Preserving State‐of‐Health Prediction in Electric Vehicle Batteries Using Fully Homomorphic Encryption CKKS

ABSTRACT This paper proposes a novel privacy‐preserving approach for accurately predicting the State‐of‐Health (SoH) of electric vehicle (EV) batteries by integrating Dynamic Random Vector Functional Link Network (RVFLN) models with Ensemble Learning under the framework of Fully Homomorphic Encryption (FHE). To ensure data confidentiality throughout the predictive process, the scheme employs the CKKS (Cheon–Kim–Kim–Song) approximate homomorphic encryption, which enables efficient computations on encrypted floating‐point data commonly encountered in SoH estimation. The proposed method, called Encrypted Dynamic RVFLN Ensemble Learning (EDRVFLN), leverages the adaptability of dynamic RVFLNs to capture complex nonlinear battery behavior. Ensemble Learning is employed to enhance robustness and accuracy, whereas CKKS encryption guarantees that all learning and inference occur on encrypted data, preserving privacy without compromising model performance. Experimental validation using the publicly available NASA battery dataset demonstrated that EDRVFLN achieved superior prediction accuracy compared with traditional unencrypted models. Importantly, the privacy‐preserving capability of this method ensures that the sensitive operational and usage data remain secure throughout the analysis pipeline. This study represents a significant advancement in privacy‐preserving predictive analytics for electric mobility. By demonstrating the feasibility and effectiveness of processing encrypted data using CKKS in a complex ensemble learning setup, this study opens new pathways for secure machine learning applications in EV systems and other domains that require strong privacy guarantees alongside analytical precision.