Wind power, as one of the renewable energy sources, plays an increasingly important role in today’s society where energy structure transformation and climate change issues are becoming increasingly prominent. However, the power fluctuations and instability of wind farms pose challenges to the operation of the power system, making accurate prediction of wind power particularly important. This article proposes a wind power prediction method based on improved deep transfer learning to address the problems in wind power prediction. This article first studies the correlation between the output power of multi region wind power and environmental factors, and provides the basic principle of wind power prediction. A single region wind power prediction model was established using long short-term memory algorithm; Subsequently, the deep reinforcement transfer learning algorithm was used to extend the wind power single region model to the multi region model. Finally, establish a simulation model to validate the proposed method.