In order to overcome lower difference among weights in combination forecasting with large sample determined by traditional entropy weighting methods, a modified entropy weighting method is developed by introducing the evolutionary automatic clustering analysis. When single forecasting approaches and predictions are fixed, all absolute errors are clustered and diverse discrete distributions of absolute errors corresponding to single forecasting approaches among clusters are obtained, which are tools for further comparisons among the single forecasting methods. Then, a modified entropy weighting method is given using the obtained discrete distributions. By using a numerical study with exchange rate between dollar and yen, the feasibility and effectiveness of the developed method are illustrated.