The present invention includes (1) inputting, into a model expression that defines relation between various operating values of a facility operating on a material to be rolled and energy consumption of the facility, various actual operating values as the various operating values, to calculate an actual calculation value of the energy consumption; (2) dividing the actual value of the energy consumption by the actual calculation value to calculate a reference learning value of the energy consumption; (3) inputting the set operating value defined by the setting calculation unit only in one operating value, among various operating values of the model expression, while inputting the actual operating values collected by the actual value unit in other operating values to calculate a pseudo-actual calculation value of the energy consumption; (4) dividing the actual calculation value by the pseudo-actual calculation value to calculate a correction learning value; and (5) inputting the various set operating values as the various operating values of the model expression to calculate a prediction value of the energy consumption for the material to be rolled, which is scheduled to be conveyed to the rolling line next time or later, and multiplies the prediction value by the reference learning value and the correction learning value to calculate a corrected prediction value of the energy consumption.