The invention relates to a prediction method for short-term power of regional wind power. The method comprises the steps that wind power plants in a region are clustered, and the region is divided into multiple subregions according to the clustering result; correlation coefficients of exerted force of all the wind power plants in the region and exerted force of the regional wind power are acquired; prediction errors all the wind power plants in each subregion are acquired; one wind power plant is selected from each subregion to serve as a representative wind power plant according to the corresponding correlation coefficient and the corresponding prediction error; a weight coefficient of each representative wind power plant is calculated according to the installed capacity of the corresponding subregion; and a power prediction value of the regional wind power is obtained according to power prediction values and the weight coefficients of all the representative wind power plants. The invention furthermore relates to a prediction device for the short-term power of the regional wind power. Through the prediction method and the prediction device, dependency of a regional prediction model on data completeness and prediction precision of a single wind power plant can be relieved.