基于属性频率和精度的组合预测单项模型筛选  

Individual Model Screening for Combinatorial Forecasting Based on Attribute Frequency and Accuracy

作　　者：王成宇 林名驰[1] 唐政 Wang Chengyu;Lin Mingchi;Tang Zheng(Department of Management Engineering and Equipment Economics,Naval University of Engineeing,Wuhan 430033.China;Office of Engineeing Management,No.92690 Uni,the PLA,Sanya Hainan 572000,China)

机构地区：[1]海军工程大学管理工程与装备经济系,武汉430033 [2]92690部队施工管理室,海南三亚572000

出　　处：《统计与决策》2022年第5期22-27,共6页Statistics & Decision

基　　金：国家社会科学基金资助项目(18BGL287)。

摘　　要：文章针对已有的基于粗糙集理论的组合预测单项模型筛选方法存在属性重要度可能均为0的问题,选取属性频率作为属性重要度评价标准;针对原算法未考虑单项模型预测精度可能导致组合预测精度不高的问题,结合单项模型的均方根误差构成新的属性重要度评价标准,提高了组合预测精度,同时解决了属性因重要度相同而难以选择的问题;为使算法完整,将筛选过程细化为有核与无核两种情况,并给出详细算法步骤。结合四种组合预测方法与不筛选和原算法得到的模型集进行实例对比分析,验证了筛选的必要性与该算法的有效性。In view of the problem that the attribute importance degree may all be 0 in the existing individual model screening method of combinatorial prediction based on rough set theory, this paper selects the attribute frequency as the evaluation standard of attribute importance. Aiming at the problem that the original algorithm did not consider the prediction accuracy of individual model which might lead to the low accuracy of combinatorial prediction, the paper combines the root mean square error of individual model to construct a new evaluation standard of attribute importance, which improves the accuracy of combinatorial prediction and solves the problem that it is difficult to select attributes because of the same importance. In order to complete the algorithm, the screening process is divided into two cases with and without kernel, and the algorithm steps are given in detail. The necessity of screening and the effectiveness of the proposed algorithm are verified by comparing the model sets obtained by the four combined prediction methods and the original algorithm.

关 键 词：粗糙集理论 均方根误差 属性重要度 单项预测模型 筛选 

分 类 号：O212.1[理学—概率论与数理统计]