Cloud-Integrated Hybrid Battery Management System with Hybrid State of Charge Estimation in On-Board Electric Vehicle Battery Packs

It is difficult for existing methods to solve the real-time accuracy problem of battery module-level state of charge (SoC) and the impact of single-battery inconsistency at the same time under dynamic operating conditions. The integration of data-driven technology and traditional algorithms is insufficient, leading to limited error compensation. In view of the accuracy of the SoC estimation of electric vehicle (EV) battery packs under dynamic driving conditions, this paper proposes a hybrid SoC estimation method for battery management system (BMS) based on a cloud master-slave architecture. The hybrid framework combines direct measurement methods (Coulomb counting method, open-circuit voltage method), state estimation algorithms (extended Kalman filtering, traceless Kalman filtering), and data-driven technologies (neural networks, Nonlinear Auto-Regressive Moving Average (NARMA-L2) models), and verifies its effectiveness through hardware-in-the-loop experiments. The research results show that under dynamic operating conditions, the hybrid Coulomb counting and neural network (CC + NN) method achieves the fastest error convergence rate and outperforms other methods. In addition, the proposed cloud master-slave BMS architecture significantly improves system reliability by enabling real-time cross-verification of the SoC data from the advanced algorithms of the on-board BMS (slave device) and the master device. The experiment is based on the Federal Test Procedure (FTP)-75 driving cycle and verifies the high efficiency of this method in practical applications. The final analysis shows that the CC + NN combination exhibits optimal error-suppression performance in complex scenarios and provides a high-precision solution for electric vehicle battery management.