The invention provides an optimized power data short-term prediction method, and the method comprises the steps: firstly modeling power data trough employing a gray scale model; secondly modeling a residual error between an actual value and a prediction value through employing an artificial neural network; and finally overlapping the above two models. The method effectively improves the robustness and accuracy of short-term prediction of power data, improves the reliability of power dispatching, and effectively improves the power utilization efficiency of an urban power grid.