时间序列数据挖掘(英文部分看不懂)

时间序列分析建模最大的优点在于不必深究信号序列的产生背景,序列本身所具有的时序性和自相关性已经为建模提供了足够的信息,只需要有限的样本序列,就可以建立起相当高精度的预测模型,但其存在低阶模型预测精度低、高阶模型参数估计难度大的不足。

 

缺点:为了捕获整个事件,需要对不同时期的数据进行观察,这样增加数据维度。

这种比较增加了数据的字相关性,这样三角不等式triangular inequality就不成立 ,限制聚类方法的选择。

为了克服以上缺点,我们提出了一种混合多维匹配方法。

这种方法通过局部改变观察规模来比较序列的相似性,这样就可以捕获长期和短期事件。

由于matching是基于相邻片段的相似性上,他能够反映数据的 趋势(上升,下降)。

同时还有一个很重要的优势,It also has a advantage that the

connectivity of subsequences is preserved in the resultant
sequence
because it checks hierarchy of inflection points.

子序列的的连通性得以保存,因为它检查了拐点层次?
While, rough clustering is a clustering method that groups up sequences based on their indiscernibility, which is defined
in the context of rough set theory [4].

粗糙聚类分析无
It does not use any distance-related features, therefore, is able to produce interpretable clusters even when similarity of sequences is defined as relative ones. In our method, multiscale matching
is used to calculate similarity of the sequences. and rough
clustering is used to cluster the sequences according to the
derived similarity. The common patterns in the clustered
sequences are then visualized using the result of multiscale
matching 10 support visual inspection of the results by expertS.

 

 一种常用的聚类逼近方法是 利用dwt离散小波分析离散傅立叶系数

posted @ 2014-05-15 15:55  puckpuck  阅读(540)  评论(0编辑  收藏  举报