Abstract: The medium-long-term electricity forecasting plays an important role in drawing up medium-long-term generation plan, improving new energy consumption and ensuring power balance in power system. The forecast information of future climate state is helpful to improving the forecast precision of medium-long-term quantity of electricity; however, at present the forecast information of medium-long-term quantity of electricity can not be effectively mined and utilized. Consequently, in order to improve the adaptability of the prediction model, a wind energy feature mining model is constructed by taking the wind energy resources regional climate forecast data as input, and the selection of data sets with different forecast error characteristics is realized. Moreover, combined with the actual power generation data of wind farm, the adaptive prediction models are constructed with GWO and LSTM. Compared with the current forecasting methods, the results show that the proposed forecasting method for medium-long-term quantity of electricity can be adopted to realize the total power forecasting of a coastal wind farm and a region, and the performance of the forecasting models is optimal, which proves the validity and advance of the method.