This paper proposes a reactive power and voltage optimal control method for wind farms based on data-driven power flow. Because no prior knowledge of wind farm parameters is necessary

the proposed method is model-free. Based on Koopman operator-based method

this paper constructs a power flow model of wind farms connected to the grid by using state space mapping and lift-dimension linearization. Considering reactive power devices such as wind turbines and static var generator (SVG) in wind farms

a global sensitivity-based reactive power and voltage linear optimization control model is proposed. Taking minimum reactive power adjustment of wind turbines and SVG as the objective function

combined with the sensitivity relationship between node voltage and reactive power injection

the proposed model-free voltage control method can realize optimal reactive power distribution

effectively reduce active power loss

and satisfy the requirement of rapid voltage control response of wind farms. Based on historical data of a wind farm in Ningxia

feasibility of the proposed voltage optimal control method under inaccurate parameters is verified. Compared with model-based methods

the proposed method exhibits advantages on parameter dependency and efficiency.