Autoregressive model-based fuzzy clustering and its application for detecting information redundancy in air pollution monitoring networks
Autoregressive model-based fuzzy clustering and its application for detecting information redundancy in air pollution monitoring networks
复制标题
DOI：
10.1007/s00500-012-0905-6
复制
发表时间：
2013-01-01
期刊：
影响因子：
4.1
通讯作者：
Maharaj, Elizabeth Ann
中科院分区：
文献类型：
--
作者：
D'Urso, Pierpaolo;Di Lallo, Dario;Maharaj, Elizabeth Ann
模糊聚类使对象同时成为两个或多个聚类中的成员。对于时间序列来说，这一点尤其重要，因为时间序列的模式常常会随着时间的推移而变化。因此，时间序列可能在不同的时间段内属于不同的聚类，在这种情况下，清晰聚类无法捕获这种多聚类成员关系。在本文中，我们采用了模糊C-中心点的方法来聚类的时间序列的自回归估计模型拟合的时间序列的基础上。我们说明了非常好的性能，这种方法在一系列的模拟研究。通过两个应用程序，我们还显示了这种聚类方法在空气污染监测中的有用性，通过考虑空气污染时间序列，即，全球和城市尺度的CO时间序列、CO2时间序列和NO时间序列监测。特别是，我们表明，通过考虑在聚类过程中，这些空气污染时间序列的自回归表示，我们能够检测到可能的信息冗余的监测网络，然后，减少监测站的数量，以降低监测成本，然后增加监测效率的网络。
Fuzzy clustering enables the simultaneous membership of objects in two or more clusters. This is particularly pertinent where time series are concerned, because very often patterns of time series change over time. Thus, a time series might belong to different clusters over different periods of time, in which case, crisp clustering is unable to capture this multi-cluster membership. In this paper, we adopt a Fuzzy C-Medoids approach to clustering time series based on autoregressive estimates of models fitted to the time series. We illustrate very good performance of this approach in a range of simulation studies. By means of two applications, we also show the usefulness of this clustering approach in the air pollution monitoring, by considering air pollution time series, i.e., CO time series, CO2 time series and NO time series monitored on world and urban scales. In particular, we show that, by considering in the clustering process, the autoregressive representation of these air pollution time series, we are able to detect possible information redundancy in the monitoring networks and then, decreasing the number of monitoring stations, to reduce the monitoring costs and then to increase the monitoring efficiency of the networks.