It can win through traditional genetic algorithm's shortcomins by applying accelerating genetic algorithm in solving combination forecasting problems. These shortcomings include poor adapt ability in search space (i.e. optimizing variable space), large measure quantity, premature convergence, no definitude instruct rule for setting technique of control parameter, etc. The new approach does not need canonicity in forecasting error information matrix, the objective function scale may extend widely,and the forecasting precision is high.