An improved genetic algorithm combined with niche algorithm and immune algorithm is proposed to solve the standard genetic algorithm for prematurity and slow convergence rate. On one hand, vaccine is introduced to guide the generation of initial population, so the individual has some excellent gens which will reduce the optimal time, and an adaptive updating mechanism of vaccine and parameter scope is proposed as the wind speed data renewed. On the other hand, niche genetic algorithm is adopted to keep the diversity of evolution population. Examples show that it is feasible to apply the proposed method in short-term wind speed prediction, with higher forecasting accuracy and convergence speed.