英语翻译2 Related WorkVarious methods have been proposed for tim

英语翻译
2 Related Work
Various methods have been proposed for time
series searching.In [l],an indexing scheme called
F-index is suggested to handle data sequences and
query sequences of the same length.Firstly,each
data sequence is transformed by n-point Discrete
Fourier Transform.The first fc coefficients are
kept and regarded as a fc-dimensional point.The
feature points are then indexed by an R*-tree
[16].For a range query,the query sequence is
first mapped to a point in the fc-dimensional
space similarly.Then,the R*-tree is searched
and all feature points that are within the error
distance from the query sequence are retrieved.
This method guarantees no false dismissal,but it
may cause false alarms.Thus,the original Idata
sequences corresponding to the points retrileved
have to be checked against the query sequence.
The results in [l] are further generalized in [2]
and the ST-index is proposed to handle data sequences
of different lengths.A sliding window with
length n is placed over the data sequences.The
subsequence within each window is transformed
by n-point Discrete Fourier Transform.After all
data subsequences are transformed,a trail will be
formed.The trail is divided into sub-trails,which
are then represented by minimum bounding rectangles(
MBR) o f an R*-tree.For range query,al1
MBR that intersect the query region will be retrieved.
This method also guarantees no false dismissal,
yet false alarms are still possible,and so the
original data sequences have to be checked against
the query sequence,too.The methods proposed
in [l,21 are very elegant.However,they use Euclidean
Distance for sequence similarity without
considering any transformation.As shown by the
example in Section 1,it is better to consider sequence
similarity with scaling and shifting in some
applications such as stock analysis.
The definition of similarity used in this paper is
similar to those proposed in [4].In [4],the authors
develop a general framework for similarity queries.
The framework consists of a transformation rule
language T.An object A is said to be similar to an
object B if A can be transformed to B by a series of
transformations defined in T.Each transformation
applied has a cost and the total cost is used to
measure the distance between A and B.
In [5],the authors consider the case that T contains
the transformation of moving average and
time wrapping.They first show that the definition
of sequence similarity with moving average
and time wrapping has a wide range of real applications.
Then,they illustrate by real stock data
that the transformations help to identify similar
runs of stock price.They also propose a first indexing
method that can handle moving average
and time warping.An index I is constructed as
in [l,2] first.For each query,a transformation in
T is given.
不要放入翻译词典 再拿出来交给我好不好。
囧

2 相关工作
关于时间序列搜索,大家已经提出了很多方法.在【1】中,称为F指数的指数方案被用于处理数据序列和相同长度的查询序列.首先,对每个数据序列进行了N点离散傅里叶变换.保存第一个fc系数,并把它作为fc维数的点.然后用R *-树[16]检索特征点.对于范围查询,首先查询序列类似地映射到fc维空间里的一点.然后,搜索R *-树,返回所有与范围查询在误差距离内的特征点.这种方法可以确保没有漏检,但它可能会导致虚假警报.因此,返回值所对应的原来的1数据序列要经过与查询序列的核对.把[1]的结果进一步推广为[2],就提出了ST指数.它用来处理不同长度的数据序列.一个长度为n的滑动窗口被安置在数据序列上.每个窗口内的数列进行N点离散傅里叶变换.所有的数列都经过变形以后,将形成一个轨道.这条轨道被划分成子轨道,然后由最小边界矩形一个R *-树（MBR）代表.对于范围查询,返回插入查询序列的MBR.这个方法也保证不存在漏检,但仍然是可能的误检,因此原始数据序列也要与查询序列进行核对.在1.21提出的方法非常优雅.然而,他们使用欧几里德距离序列的相似性,而不考虑任何转换.正如在第1部分的例子所示,在某些应用中,综合考虑序列的相似性的标定和转换更好一些,如股票分析.在本文使用的相似性定义和在[4]中提出的类似.[4]中,作者展开了相似查询的大体框架.该框架包含一个转换规则语言T.如果A可以通过T中定义的一系列的变形,转化到B,那么就说对象A与对象B相似.在T每次用转化有一个定义的转换成本,总成本用来衡量A和B之间的距离.[5]中,作者考虑T包含滑动平均和时间规整算法的情况.首先表明,这种对T的定义实际应用范围很广.然后,他们用真实股票数据说明该转换有助于确定类似股票价格走势.他们还提出了能够处理滑动平均和时间规整算法的第一个索引方法,在【1,2】中建立了一个索引I.每个查询都给出了相应的T.
累死了…… 这是数学的论文?