A non-stationary analysis method in time series: hierarchical segmentation and its applications
Date Issued
2007
Date
2007
Author(s)
Chen, Shu-Chun
DOI
en-US
Abstract
There are two parts in this dissertation. First is that we propose a new method,
Hierarchical Segmentation (HS), to analyze non-stationary time series data. The
main idea of HS is to divide a series of data into several segments according to the
characteristics of data through three level coding. At the upper level, there resemble
each other if they have the same codewords. So, it is useful in identifying patterns
or special segments with appropriate coding. Here, we apply HS in three different
fields: circadian rhythm, moving trajectory and the emotional interaction of couples.
In circadian rhythm study, a phase represents each time point for the experiment
duration. It is crucial to decide robust and reliable markers as the onset or offset
points for the daily activity. We define the onset phase markers via HS; furthermore,
we discuss the variation of periods, waveform, and the phase shift inducing by the
outside stimulation and develop a new approach in constructing phase response curve
(PRC). We compare HS and Onset methods via simulation. It reveals that HS is
more robust than Onset; in other words, the phase markers inducing by HS can mark
the benchmark in waveform. In the study of oviposition behavior, we explore the
moving trajectory of bean weevil and try to find out the moving patterns via HS. We
are interested in exploring that if subject changes her moving pattern under three
different resource environments. It reveals that the moving patterns is simpler and
local search preferred in rich patch than in half and poor. With resource consumed,
the moving patterns are increased into lots of long walk. In the couple data analysis,
we divide the segments of highly emotional response. After examining with chi-square
test, we find that there is coherence situation in both positive and negative emotion.
We separate the series data into three segments, and we prove that the approach is
helpful in catching the interaction patterns. Finally, we use stochastic small-world
network (SSWN) and dynamic movie to represent the dyadic dynamic interactions.
The other part is to explore the unstable hierarchy formation in cockroach, Nauphoeta
cinerea. We develop two indices to quantify the degree of unstable and the competitive
pressure in a group. It reveals that the degree of unstableness is highly related with
group size, and there is highly positive relation between unstableness and competitive
pressure. In other words, within a group, the more unstable a hierarchy formation is,
the stronger the competition appears.
Subjects
層級切
割法
生物時鐘
行走軌跡
夫妻情緒相互影響
階層制度
hierarchical segmentation
circadian rhythm
moving trajectory
small-world dyadic dynamic interactions
hierarchy fornation
Type
thesis
File(s)![Thumbnail Image]()
Loading...
Name
ntu-96-D92621201-1.pdf
Size
23.31 KB
Format
Adobe PDF
Checksum
(MD5):0352c19f6dcbadb725a094dad0812e8b