Annual energy output of a candidate site in its life span is an important reference criterion of wind farm macro siting. A regional information fusion method， which allows the use of multiple reference wheather stations with a long history of wind speed and wind direction measurements, was proposed to improve the annual energy output prediction accuracy. Firstly, the correlation model was established between the short-term wind data of a single reference wheather station and the candidate wind farm, and the multiple long-term wind speeds of candidate site based on different reference stations were predicted by using the model. Then, the multiple prediction results were integrated by neural network to obtain the final long-term hourly wind speed data, and the annual energy output was subsequently determined on the basis of the knowledge of these wind speeds. The simulation results show that, by using the proposed method, the error reduction up to 11.32% has been achieved in the relative error of the average annual power output, with respect to the case of using a single reference wheather station method.