Abstract: Wind farm wake deflection control is an important tool to reduce the wake effect and improve the total power generation. The wind prediction is an important input to the wind farm wake deflection control, and its error has a huge impact on the actual control effect, even leading to a "decrease instead of an increase" in the overall power generation, which greatly limits the engineering application of wind farm wake deflection control technology. Therefore, this paper explores the impact of minute-level wind speed and wind direction prediction errors on the wind farm wake deflection control effect of an offshore wind farm, and proposes an offshore wind farm wake deflection control based on adaptive wind condition prediction error and a control model based on deep neural network. The results show that the total power generation of the proposed method is improved by 1.77% compared with the conventional wind farm wake deflection control method without wind prediction error adaption.