Abstract: In order to improve forecasting precision and forecasting scale of short-term wind speed of wind farm, a short-term multi-step forecasting molding method for time sequence of wide speed combining with wavelet transform, empirical mode decomposition and least square support vector machine was proposed. The method uses wavelet transform to decompose wind speed data which can be decomposed into high-frequency and low-frequency components, uses least square support vector machine to construct forecasting models relevant to the components, then adds forecasting values of the models to obtain forecasting result which is called result of model I. Secondly, the method considers forecasting result of model I as training sample, and uses empirical mode decomposition to decompose the training sample into several intrinsic mode functions and trend term. Then it uses least square support vector machine to build forecasting models for each intrinsic mode functions and trend term, meanwhile extends forecasting scale of the models, and adds forecasting values of the models to obtain forecasting result which is called result of model II. Finally, it adds forecasting results of model I and model II to obtain forecasting result. The experiment result showed that the value of RMSE is 0.153 and proved the effectiveness of the method.