Integration of large-scale wind-farm in power grid will impact grid planning, construction and energy quality. To improve power system dispatching and safe-stable operation of power grid containing a lot of wind power generating units, accurate short-term wind power forecasting is significance. This paper proposed a method of wind power short-term multi-step prediction research based on the improved KNN algorithm. Optimal parameters for phase space reconstruction and prediction method were studied, and the application of multiple evaluation index to evaluate the forecast results, it verified the effeteness of the model.