一种考虑偏移误差的组合预测权重调整优化策略

A Weight Adjustment Optimization Strategy of Combination Forecasting Considering Migration Accuracy

针对传统组合预测赋权方法忽视时点预测效果差异性和近期历史数据重要性的问题,提出一种考虑偏移误差的权重调整优化策略。首先,用传统组合预测方法确定初始权系数;其次,定义偏移误差来衡量各单模型点预测效果与整体预测效果的差异;最后,基于偏移误差构建权重调整优化策略,从而实现时点权重随预测效果的自适应变化,取近期40%时间跨度下的调整权重均值为组合权重获得预测值。结果表明:在各步长上,权重调整优化策略可将传统组合预测方法的预测精度提高约0.3%。

In order to solve the problem that the traditional method of combination forecasting ignores the difference of prediction effect at different times and the importance of recent historical data,a strategy that took into account the time-point prediction offset error and adaptively adjusted the combination weights was proposed.Firstly,the traditional combination forecasting method was used to determine the initial weight.Secondly,the offset error was defined to measure the difference between the time-point prediction effect of single model and the overall prediction effect of single model.Finally,the model of weight adjustment was constructed based on the offset error,so that the weights could vary with the prediction effect at each time point adaptively.The average of adjusted weight in the recent 40%time span were taken as the combined weight to obtain the predicted value.The results show that in each step size the weight adjustment optimization strategy can improve the prediction accuracy of the traditional combination prediction method by about 0.3%.