The conventional double exponential smoothing is a forecasting method that troubles the forecaster with a tremendous choice of its parameter, alpha. The choice of alpha would greatly influence the accuracy of prediction. In this paper, an integrated forecasting method named Golden Exponential Smoothing (GES) was proposed to solve the problem. The conventional method was reformed and interposed with golden section search such that an optimum alpha which minimizes the errors of forecasting could be identified in the algorithm training process. Numerical simulations of four sets of times series data were employed to test the efficiency of GES model. The findings show that the GES model was self-adjusted according to the situation and converged fast in the algorithm training process. The optimum alpha, which was identified from the algorithm training stage, demonstrated good performance in the stage of Model Testing and Usage.