Because the traditional static feed forward neural networks (FNN) are easy to fall into local optimum and lack of dynamic performance, the wind speed prediction model using Elman neural network (ElmanNN) is established, the principal component analysis (PCA) is used to extract the feature of wind speed data, which optimizes the inputs of ElmanNN. Furthermore, excitation function and the structures of network are improved to search for the optimum solution of function convergence rate and prediction accuracy. To solve large error and prediction accuracy fluctuations of the ElmanNN model at the peak value of wind speed, the rough set theory is proposed to compensate and correct the predicted values to further improve the forecasted results. Experimental results show that the prediction accuracy is effectively improved and the generalization ability of ElmanNN is enhanced using the proposed method. This model has precise forecasting and strong practicability, so it has promoted value.