利用经验模态分解,将北京房山和十三陵跟踪站2000-2006年高程方向时间序列,分解成有限个频率逐渐降低的本征模态分量和表征时间序列线性变化趋势的残差,得到两个站的线性变化速率分别为-0.533 mm/a与-2.083 mm/a.对各分量进行Hilbert变换,得到二维Hilbert边际谱和时间-频率-能量的Hilbert频谱图.结果表明较大振幅的周期项对应的频率均在3 cycle/a以内,两测站均存在半周年信号与周期约为351天的周年信号,以及一个非线性变化趋势项.对时间序列进行EMD低通滤波,并与sym6、db8小波滤波结果进行比较,结果表明EMD去噪结果相对较好,光滑且无毛刺,说明EMD滤波方法具有更强的剔除瞬时强噪声的能力.%Taking the GPS fiducial stations, BJFS and BJSH as the examples, the time series in elevation di rection from 2000 to 2006 were decomposed into a finite number of intrinsic mode function, whose frequency was de creasing, and the residual representing the trend of the time series. The linear rates of the two stations were -0.533 mm/a, -2.083 mm/a respectively. Furthermore every component is transformed with Hilbert algorithm, though the procedure two-dimensional marginal spectrum and three-dimensional Hilbert spectrum in time-frequency energy space were obtained and analyzed. The marginal spectrum has perfect time-frequency concentration; and then the frequency of high amplitudes is no more than 3 cycle/a. The main components in time series of the both stations are a secular nonlinear trend, and semiannual signals, and annual signals which is obvious and about 351 days. Moreover, Hilbert spectrum has high resolution and the ability of outlier detection. Afterwards the time series was processed by EMD low-pass filtering and the result is compared to that of sym6, db8 wavelet. De-noising re sults by EMD was much superior, slick and no glitch, which indicates that this method is more efficient in elimina ting instantaneous strong noise.
展开▼
