The combination forecasting model based on induced ordered weighted averaging (IOWA) operators, which is built according to the criterion of error sum of squares, failes to reflect the influence of errors arising from observation points in various periods on the predictive values. Moreover, this method can not be used to predict directly because future data are unknown in actual forecasting. In order to overcome the above flaws, an improved method is proposed. First, individual forecasting model that has higher forecasting accuracy is chosen as a criterion. Then, the deviation of predictive values between other models and standard model is computed. The weights are given according to the mean value size of the absolute value sum of deviation in every individual forecasting model in every period. Finally, a new forecasting model is built in accordance with the weighted error sum of squares. And genetic algorithm is used to solve the optimal weights. Verified by an example, the improved combination forecasting method is better than the original combination forecasting method based on IOWA operator. Forecasting accuracy is improved effectively.%按误差平方和准则建立的基于IOWA算子的组合预测模型并不能正确反映出各个时期观测点所引起误差对预测值的影响程度,在实际预测时预测期数据是未知的,无法直接利用该方法进行预测.针对以上缺陷,提出以单项预测模型中精度较高者的预测值为标准,计算其余模型的预测值与其偏差,再按各个时期各单项偏差绝对值和的平均值大小赋予权系数,建立按照加权误差平方和准则新的预测模型,并利用遗传算法求解最优权系数.通过实例验证,改进后的组合预测方法优于原来的基于IOWA算子的组合预测方法,有效地提高了预测精度.
展开▼
