Hierarchical energy optimization of flywheel energy storage array systems for wind farms based on deep reinforcement learning.

This article focuses on the hierarchical energy optimization and intelligent dispatching of a flywheel energy storage array system (FESAS) integrated with wind farms to smooth power output and enhance grid stability. The energy dispatching problem is modeled as a Markov decision process (MDP) and solved using a deep reinforcement learning (DRL) approach, specifically the soft actor-critic (SAC) algorithm combined with prioritized experience replay (PER) to improve learning efficiency and stability. A case study involving a 49.5 MW wind farm in Inner Mongolia with a 10 MW FESAS demonstrates that the SAC-PER algorithm outperforms other DRL methods in achieving precise energy dispatching aligned with grid dispatch instructions. The proposed hierarchical control framework integrates physical, control, and network layers to coordinate source-grid-storage operations effectively. Simulation results based on real power data confirm that SAC-PER can optimize the charging and discharging of the FESAS, thereby supporting stable and reliable grid operation amid renewable energy fluctuations.