The model mismatch and parameter uncertainty of flywheel battery system are the key factors that affect the stable operation of magnetic suspension flywheel battery system under the external random interference of vehicle. The existing control methods usually ignore the uncertain effects of the system. In this study, a nominal particle robust model predictive control (NP-RMPC) method for nonlinear systems considering the uncertainty of model parameters is proposed. First, the particle characteristics are used to model the system uncertainty and observable system dynamics. Then, a robust observer is constructed to estimate the system uncertainty caused by external random perturbations. Then, based on the nominal particle prediction model, the robust model predictive control (RMPC) is optimized, and an off-line fast solution method is designed for efficient output control law. Finally, various experiments are carried out on the flywheel rotor trajectory tracking control. The results show that the response time is reduced by 60% and the tracking error is reduced by 66.94%. Consequently, the effectiveness of this method is proved.

The flywheel battery system is extremely sensitive to its own time-varying nonlinear characteristics and random disturbances in actual operating conditions. The traditional model predictive control (MPC) for flywheel batteries is usually studied based on an offline vector mechanical model with a large dynamic accuracy limit, which leads to the inability of traditional MPC research to solve the problems of online optimization of the model and quantitative compensation for uncertainty. In response to the above drawbacks, this article proposes a learning model predictive control (LMPC) method based on local sparse optimization Gaussian process. First, the system residual part and the observed dynamics of the vector mechanics model are respectively modeled as the Gaussian process regression process (GPR). Then, a sparsification Gaussian process based on screening local magnetic density induction points is introduced and a dynamic diffusion optimization mechanism is designed to reduce the computational burden. Furthermore, a learning model prediction theory based on Kalman filtering Gaussian process (GP-EFK) is proposed to ensure the practical application of the proposed method. Finally, experiments on the simulated excitation signal experimental platform show that this method has strong capability and reliability. Compared with the common learning-based model predictive control method, it reduces overshoot by approximately 10% and effectively improves the control accuracy.

As an emerging physical energy storage technology, the magnetic suspension flywheel battery boasts prominent advantages such as high working efficiency, long service life, and short charging time. However, improving the energy conversion efficiency of magnetic suspension flywheel battery systems and reducing their overall energy loss have long been critical bottleneck technologies that urgently need to be addressed for practical applications. To promote China’s green and low-carbon energy transition and accelerate the achievement of the “double carbon” goals, this paper summarizes two core components of flywheel battery systems—magnetic bearings and flywheel motors—along with two key technologies: topological structure and control strategy, based on numerous cutting-edge studies. Subsequently, focusing on further reducing the energy consumption of flywheel energy storage systems, technical prospects are extended from aspects including system material selection and intelligent integrated control, aiming to provide research directions for the low-energy-consumption operation of flywheel battery systems.