Hybrid Federated Neural–Observer Predictive Control for Robust Electric Vehicle Battery Management

Abstract

The rapid growth of electric vehicles (EVs) has intensified the demand for reliable battery management systems (BMS) capable of ensuring safety, longevity, and performance under diverse operating conditions. Conventional observers such as the Extended Kalman Filter (EKF) and data-driven neural networks have shown limitations in scalability, robustness to noise, and interpretability. This paper proposes a Hybrid Federated Neural–Observer Predictive Control (HFNOPC) framework for robust EV battery management. The framework integrates three innovations: (i) a robust super-twisting sliding observer for noise-resilient state estimation of state-of-charge (SOC), state-of-health (SOH), and temperature; (ii) a physics-guided neural residual network that enhances the baseline equivalent circuit and thermal models by capturing nonlinearities due to aging and hysteresis; and (iii) a federated learning strategy that enables distributed EVs to collaboratively train the neural residual component without sharing raw data, thus ensuring scalability and privacy. The enhanced state estimates are coupled with a model predictive control (MPC) scheme, which optimizes charge–discharge trajectories subject to safety and thermal constraints. Simulation studies demonstrate that the proposed HFNOPC reduces SOC estimation error by up to 32% compared with EKF and decreases control cost by 24% compared with conventional PID–based charging strategies. Furthermore, robustness tests under ±10% sensor noise and thermal stress confirm improved stability and accuracy. These results highlight the potential of the proposed framework as a next-generation BMS solution, offering interpretability, robustness, and fleet-wide scalability, thus paving the way for safer and more efficient EV deployment.