Wind turbines working close to other turbines experience interactions that affect the power production. These interactions arise as a consequence of wakes caused by upstream wind turbines. In order to achieve a more effective and precise control of the power generated by wind farms, the control strategy must consider these interactions. However, the phenomena involved in wake effects are complex especially in cases of large number of turbines. This paper presents the implementation of a gradient estimation-based algorithm as a model-free control for two different control schemes aimed to maximize the energy capture of a wind farm. One control is centralized, leaving to a supervisor the task of command computation and the other topology is decentralized, distributing the performing generation among wind turbines. This latter scheme aims to increase the reliability of the wind farm operation by reducing the communications needed to fulfill the objective of maximizing energy capture. Both control schemes are evaluated by simulation in the case of three-turbine wind farm.