苏州科技大学学报（自然科学版）2025，Vol.42Issue(3)：20-29,36,11.DOI:10.12084/j.issn.2096-3289.2025.03.003
加权调和平均组合预测的倒数均方误差分解与模型筛选
Reciprocal mean square error decomposition and model selection for weighted harmonic mean combination forecasting
摘要
Abstract
The weighted harmonic mean combination forecasting method is a nonlinear forecasting model.For this model,the concepts of reciprocal mean square error(RMSE),reciprocal mean deviation,model matching coefficient and forecast deviation measure are proposed.The mathematical properties are discussed.On this basis,the decomposition expression of the reciprocal mean square error is proved from two perspectives.On the one hand,the reciprocal mean square error is the sum of three parts,which are the reciprocal mean deviation,the reciprocal variance of the model mismatch,and the reciprocal bias sequence variance.On the other hand,from the perspective of the prediction deviation measure,the reciprocal mean square error is obtained from the difference between two terms,which are the weighted average of the reciprocal mean square error of the sub-forecasting model and the weighted average of the forecast deviation measure.Thus,the basic criterion for selecting the weighted harmonic mean combination forecasting method is given,and the sub-models with higher accuracy and greater difference are selected as far as possible to participate in the combination forecasting.The results of the decomposition of the two RMSE sources provide a theoretical basis and an effective way for model screening and reducing the error of the weighted harmonic mean combination prediction.关键词
加权调和平均/组合预测/倒数均方误差/偏离性测度/模型筛选Key words
weighted harmonic mean/combination forecasting/reciprocal mean square error/deviation measure/model screening分类
数理科学引用本文复制引用
陈华友,叶知秋,徐雪涛..加权调和平均组合预测的倒数均方误差分解与模型筛选[J].苏州科技大学学报（自然科学版）,2025,42(3):20-29,36,11.基金项目
国家自然科学基金项目(72371001) （72371001）
安徽大学大学生创新训练计划项目(X20241699) （X20241699）
