In order to study the feasibility of combination forecasting model based on least squares support vector machine for wind speed short-term forecasting, the forecasting data coming from Back Propagation (BP) neural network, Radial Basis Function (RBF) neural network, and Particle Swarm Optimization neural network (PSOBP) were used as inputs and the actual wind speed was used as output in this model. The least squares support vector regression algorithm is used for constructing the nonlinear relationship in order to achieve multi-step forecasting for wind speed. The forecasting performance of the model is compared with BP combination forecasting model and linear combination forecasting model, and it was evaluated by mean absolute error, sum of squared error, and average relative error. The results indicate that the average relative error for least squares support vector machine prediction model is less than 6%; and the rest of error indicators for it are significantly lower than other models. Therefore its forecasting accuracy is not only better than any other single forecasting model, but also better than the traditional linear combination forecasting model. It validates the feasibility of least squares support vector machine combined forecasting model for wind speed forecasting.