摘要
基于距离的函数型聚类分析包含曲线拟合和聚类两个独立步骤,最优曲线拟合未必有利于类别信息的提取和保留。根据曲线拟合与聚类分析的计算过程,重新梳理了函数型聚类算法;基于距离度量,提出了同时考虑拟合和聚类效果的函数型聚类一步法;在交替方向乘子法(ADMM)框架下推导并给出了迭代求解算法。模拟试验结果显示,该函数型聚类算法有助于提高聚类精度;针对北京市空气质量监测站点二氧化氮(NO2)污染物小时浓度数据的实例验证分析表明,该函数型聚类算法对不同类别空气质量监测点具有更好的区分度。
Distance-based functional clustering analysis includes two independent steps:Curves fitting and clustering.Optimal curves fitting are not necessarily valid to the extraction and retention of segmentation of information.According to the calculation process of curves fitting and clustering,the functional clustering algorithm is reviewed.A distance-based functional clustering approach that considering both fitting and clustering effect is proposed.And an iterative algorithm is deduced under the framework of alternating direction multiplier method(ADMM).The simulation results show that our algorithm is aid to improve the clustering accuracy.The case study on the hourly concentration data of nitrogen dioxide(NO2) at Beijing Air Quality Monitoring Station shows that our algorithm has better discrimination for different types of air quality monitoring stations.
作者
黄恒君
高海燕
张梦瑶
HUANG Heng-jun;GAO Hai-yan;ZHANG Meng-yao(School of Statistics,Lanzhou University of Finance and Economics,Gansu Lanzhou 730020,China)
出处
《数理统计与管理》
CSSCI
北大核心
2019年第6期986-995,共10页
Journal of Applied Statistics and Management
基金
国家社科基金青年项目(14CTJ009)
全国统计科学研究重点项目(2017LZ43,2017LZ41)
甘肃省“飞天学者”特聘教授项目
兰州财经大学“青年学术英才支持计划”资助
关键词
数型数据分析
聚类
交替方向乘子法
functional data analysis
clustering
alternating direction method of multipliers
