Electric power load forecast is the basis for the planning and safe operation of electric power, and the improvement of forecast accuracy is the important point for the research on electric power load forecast. Because of the changeable characteristics and uncertainty of load forecast, single forecast model is difficult to satisfy all forecast conditions and combined forecast is a kind of forecast method by using prediction results from each prediction to choose appropriate weight and coefficient for weighted average. Grey and time-series are used in single prediction model, then combined prediction model is set up by optimal combination for short-term load prediction of electric power system and simulation experiment shows that optimal combination prediction model has higher prediction accuracy than single prediction model and has certain advantage.