Wake model-free active power control for wind-storage coordination using a state-decoupled MATD3 algorithm

Active wake control (AWC) is a key research direction for enhancing the energy efficiency of existing wind farms. However, existing research has significant limitations, particularly its reliance on predefined wind turbine layouts and wake models, as well as the potential for substantial fluctuations in active power. This study proposes a wind-energy storage-coordinated AWC strategy based on a novel control strategy of wind turbines. This approach, informed by mechanistic characteristics, constructs a wind turbine-specific AWC optimization framework that does not require the turbine layout and wake modeling. The solution is based on the multi-agent double-delayed deep deterministic policy gradient (MATD3) algorithm, which treats wind turbines and energy storage batteries as separate agents. Taking into account their physical properties, a state-decoupled offline training strategy and online deployment strategy are designed. Experimental results show that the proposed strategy is effective at low to medium wind speeds, improving and smoothing active power while also reducing the charging and discharging of the energy storage system, and exhibiting good robustness against wind farm wake mismatch and noise.