Résumé : Larger and more expensive offshore wind turbines, subject to more complex loads, operating in larger wind farms, could substantially benefit from more advanced control strategies. Nonetheless, the wind industry is reluctant to adopt such advanced, more efficient solutions, since this is perceived linked to a lower reliability. Here, a relatively simple self-optimizing control strategy, capable to “learn” (data-driven) which is the optimum control strategy depending on the objective defined, is presented. It is proved that it “re-discovers”, model-free, the optimum strategy adopted by commercial wind turbine in region 2. This methodology has the potential to achieve advanced control performance without compromising its simplicity and reliability.