“…In recent years, Deep Reinforcement Learning (DRL) has achieved great success in solving computationally challenging decision-making problems, such as Atari [16], Go [17], and StarCraft [18]. Due to its powerful model-free optimisation capabilities, DRL has recently been used for real-time control problems in wind farms, such as output power maximisation [19], [20] and power tracking [21]. However, these works do not take into account the fast frequency response of the wind farm, and they do not model the power grid or the mechanical structure of WTs.…”