Abstract: With the continuous improvement of wind power penetration, how to accurately and reliably forecast wind power output is a great challenge for power system dispatching departments. At present, China has relatively mature schemes for wind power forecasting solutions, whereas extreme forecasting deviations still appear in the transitional weather period. At the same time, the transitional weather dataset is a small sample compared with the conventional prediction dataset. How to achieve accurate modeling under the small sample dataset is the key to improve the accuracy. To solve these problems, this paper proposes a transitional-weather-considered day-ahead wind power forecasting based on multi-scene sensitive meteorological factor optimization and few-shot learning. The method optimizes the meteorological sensitive features under the weather transitional process of multiple scenarios and expands the small sample set of multi-scene sensitive meteorological features by using time-series generative adversarial network. The nonlinear mapping relationship between the expanded sensitive meteorological factors and the output of wind power is modeled by using the long-short term memory neural network. The data of a wind farm in Jilin Province are used for example verification, and the results illustrate that the proposed model can partly improve the prediction accuracy of transitional-weather-considered day-ahead wind power.