改进的模糊时间序列模型论域划分算法  

Modified fuzzy time series model interval partitioning algorithm

作　　者：帅勇[1] 宋太亮[2] 王建平[1] 詹文斌 SHUAI Yong;SONG Tai-liang;WANG Jian-ping;ZHAN Wen-bin(Department of Technical Support Engineering, Academy of Armored Forces Engineering, Beijing 100072, China;China Defense Science and Technology Information Center, Beijing 100142, China;Communication Station of Air Force Equipment Research Academy, Beijing 100085, China)

机构地区：[1]装甲兵工程学院技术保障工程系,北京100072 [2]中国国防科技信息中心,北京100142 [3]空军装备研究院通信所,北京100085

出　　处：《计算机工程与设计》2017年第2期379-383,394,共6页Computer Engineering and Design

基　　金：军队技术基础基金项目(A157167);武器装备预先研究基金项目(9140A19030314JB35275);军队维修科研基金项目(2014BZ54)

摘　　要：针对模糊时间序列模型多尺度比率算法对样本特征分类结果的贡献因素及比率值的变化考虑不全面的问题,提出变化的区间标度并构造不等距的多尺度区间,选择不同区间长度隶属度对应的数据作为比率,对论域进行划分,采用改进的加权模糊C均值聚类算法聚类数据,达到样本类内距离最小化和类间距离最大化,计算各类数据的平均相对误差并基于这些误差推算多尺度、非等间距的论域划分区间,建立改进的模糊时间序列模型,使用加权平均值法处理模糊化问题并预测结果。通过案例分析,验证了改进的模糊时间序列模型的有效性。Aiming at the problem that fuzzy time series multi-scale ratios algorithm considers the contribution factor of sample characteristics classification result and the metabolic ratios incomprehensively,variational region scale was pointed out and non isometric multi-scale region was built and data corresponding to different region length were selected to be the interval partitio-ning ratio.Improved weighted fuzzy C-means algorithm was brought in to cluster the data for minimizing the distance in the sam-ple clusters and maximizing the distance between the sample clusters.The average relative errors were calculated for different kinds of data and multi-scale,non-equidistant interval partitioning regions based on these errors were calculated to build fuzzy time series model.The weighted average method was used to fuzzify and predict the result.The example analysis shows the modified fuzzy time series model is effective.

关 键 词：模糊时间序列 多尺度比率 改进的模糊C均值聚类 论域划分 加权平均值法中图 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]