R中的带通滤波器:时间序列结束时出现怪异行为 [英] Band-pass filter in R: weird behaviour at the end of time series
问题描述
我有这个时间序列
t
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
1922 -0.25108773 -0.27732553 -0.29703807 -0.30274000 -0.30323653 -0.28441682 -0.24106527 -0.18705071 -0.17440826 -0.17291725 -0.19116734 -0.21678948
1923 -0.24487998 -0.26658925 -0.28613991 -0.29674346 -0.29335742 -0.28325761 -0.23326680 -0.18697904 -0.18443807 -0.18144226 -0.18190910 -0.21574376
1924 -0.24465806 -0.27349425 -0.29925888 -0.30386766 -0.30250722 -0.27464960 -0.23390958 -0.19300616 -0.17910621 -0.17869576 -0.19611839 -0.20447324
1925 -0.25326812 -0.27344637 -0.29352971 -0.30947682 -0.30872025 -0.27604449 -0.24065208 -0.19676031 -0.17172229 -0.18484153 -0.19542607 -0.21841577
1926 -0.25214568 -0.27450911 -0.29438956 -0.30392114 -0.30619846 -0.29089168 -0.24829621 -0.20204202 -0.18621514 -0.18808172 -0.19708748 -0.22629595
1927 -0.25107357 -0.27204514 -0.29494695 -0.30751442 -0.30800040 -0.28569694 -0.24655626 -0.19547608 -0.19018517 -0.18866641 -0.20132372 -0.22084811
1928 -0.24733214 -0.27490388 -0.28780308 -0.30407576 -0.30857301 -0.28629658 -0.23872777 -0.19590465 -0.18437917 -0.18274289 -0.19936931 -0.22368973
1929 -0.25531870 -0.27264628 -0.29418746 -0.30385231 -0.31022219 -0.27931003 -0.23404912 -0.19538227 -0.17226595 -0.18465123 -0.19072933 -0.22043396
1930 -0.24735028 -0.27386782 -0.29193707 -0.29925459 -0.30039372 -0.28014958 -0.23551136 -0.19511701 -0.18006660 -0.18282789 -0.20113355 -0.22095253
1931 -0.24903438 -0.27439043 -0.29219506 -0.30312159 -0.30557600 -0.28180333 -0.22676008 -0.19048014 -0.18982644 -0.18459638 -0.19550196 -0.22127202
1932 -0.25870503 -0.27650825 -0.28521052 -0.30685609 -0.30896898 -0.28378619 -0.23614859 -0.18945699 -0.17575919 -0.17820312 -0.19620912 -0.21774873
1933 -0.24187599 -0.25575287 -0.28325644 -0.29554461 -0.29018996 -0.27040369 -0.23514812 -0.19935749 -0.18732198 -0.18606057 -0.19327237 -0.22321366
1934 -0.24793807 -0.26986056 -0.29217378 -0.30479126 -0.30199154 -0.27574924 -0.24097380 -0.18560708 -0.18643606 -0.18501770 -0.19375478 -0.22418002
1935 -0.25587642 -0.27805131 -0.29239104 -0.30784907 -0.30459449 -0.28216514 -0.23839965 -0.20137460 -0.18619998 -0.18328896 -0.20121286 -0.22869388
1936 -0.25322320 -0.28025116 -0.29713940 -0.30800346 -0.31177201 -0.28473251 -0.23552472 -0.20313945 -0.18251793 -0.18383941 -0.20554430 -0.23061875
1937 -0.26268769 -0.28529769 -0.30230641 -0.31107806 -0.30183547 -0.28324508 -0.23840574 -0.19862786 -0.19297314 -0.19392849 -0.19603212 -0.22877177
1938 -0.25445601 -0.28160871 -0.29837676 -0.29879519 -0.30328832 -0.28288226 -0.23577573 -0.19521124 -0.18393512 -0.19039895 -0.20537533 -0.21924241
1939 -0.25180969 -0.28199995 -0.29601764 -0.30147945 -0.30372884 -0.27837795 -0.23720063 -0.19929773 -0.18770674 -0.19341142 -0.20753282 -0.22484697
1940 -0.15145157 -0.16596690 -0.17572643 -0.18225920 -0.18823836 -0.17504012 -0.16019626 -0.12920340 -0.12369614 -0.12024704 -0.12891992 -0.14234080
1941 -0.10045275 -0.11095497 -0.11585389 -0.11932455 -0.11976700 -0.11653216 -0.10259231 -0.08271703 -0.07621320 -0.07184160 -0.07284514 -0.07385666
1942 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000
1943 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000
1944 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000
1945 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000
1946 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000
1947 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000
1948 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000
1949 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000
1950 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000
1951 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000
1952 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000
1953 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000
1954 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000
1955 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000
1956 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000
1957 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000
1958 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000
1959 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000
1960 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000
1961 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000
我正在使用mFilter软件包中的bkfilter函数运行带通滤波器
I am running a band pass filter using the bkfilter function in the mFilter package
t.bk <- bkfilter(t, pl=9.7,pu=16)
第一个图显示了原始"时间序列t，而第二个图显示了t.bk的周期分量.
The first plot shows the 'raw' time series t while the second plot shows the cycle component of t.bk.
红线表示t变为零的位置，并且t:Cycle也应变为零的位置.但是，后者在红线之后仍在晃动.有什么帮助吗?
The red line indicates where t goes to zero and where also the t:Cycle should go to zero. However, the latter after the red line is still wobbling.. Any help?
推荐答案
这是预料之中的.您是否了解与吉卜赛现象有关的 Gibbs现象?它们围绕滤波器的功率传递函数振荡.
This is to be expected. Are you familiar with Gibbs Phenomenon, as it relates to BK-filters (or any finite length filter, for that matter)? They oscillate around the power transfer function of the filter.
这里有一篇文章讨论了对标准BK滤波器的修改，以减少这种振荡，尽管输入响应更为有限，当然:
Here is a paper that discusses a modification to the standard BK-filter to reduce this oscillation, though with more limited input response, of course: http://www.gla.ac.uk/media/media_219052_en.pdf
幸运的是，您可以在CRAN网站上找到mFilter软件包的源代码: https://cran.r-project.org/web/packages/mFilter/index.html 您要修改的文件是用R编写的，而不是像某些软件包那样的C编写的，并且可以在
Fortunately for you the mFilter package source code is available on the CRAN website: https://cran.r-project.org/web/packages/mFilter/index.html The file you would modify is written in R--not C like some packages--and is found in mFilter/R/bkfilter.R. The part of the function you would modify is here:
if(type=="fixed")
{
bb = matrix(0,2*nfix+1,1)
bb[(nfix+1):(2*nfix+1)] = B[1:(nfix+1)]
bb[nfix:1] = B[2:(nfix+1)]
bb = bb-sum(bb)/(2*nfix+1)
for(i in (nfix+1):(n-nfix))
AA[i,(i-nfix):(i+nfix)] = t(bb)
}
编译和安装已修改的软件包很简单.转到mFilter目录所在的目录，然后从Shell命令行输入R CMD INSTALL mFilter.下次进入R时，mFilter软件包将使用修改后的bkFilter()函数.
Compiling and installing packages you have modified is simple. Go to the directory where the mFilter directory is and enter from the shell command line: R CMD INSTALL mFilter. The next time you go into R the mFilter package will use your modified bkFilter() function.
这篇关于R中的带通滤波器:时间序列结束时出现怪异行为的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！
