The invention provides a GNSS-RTK (Global Navigation
Satellite System-Real-Time Kinematic) multi-model
cascade robust method, which is used for carrying out filtering
processing on multipath errors and random errors by utilizing an intelligent optimization
wavelet packet
decomposition threshold
noise reduction technology improved by self-adaptive
noise complete ensemble empirical mode
decomposition (CEEMDAN)
cascade, and comprises the following steps of: 1, importing observed GNSS-RTK monitoring result data x (t); 2, CEEMDAN parameter conditions are determined, CEEMDAN
decomposition is carried out on a monitoring result, and a series of IMF components are obtained; and 3, calculating a
cross correlation coefficient rho between each IMF component and x (t), and dividing the IMF according to a division criterion condition. The low-frequency components smaller than the threshold value are directly removed; the rest part carries out
wavelet packet
noise reduction on the noise dominant part according to a cross-correlation extreme value judgment condition, and the
signal dominant part directly participates in reconstruction and is not processed; 4, carrying out
wavelet packet decomposition on the noise dominant IMF component to obtain a wavelet packet coefficient, preprocessing the decomposed coefficient through
wavelet packet decomposition coefficient
energy distribution, and determining a level to be subjected to
noise reduction; and 5, according to the decomposed coefficients, determining the threshold value of each wavelet packet, optimizing the parameters of the improved wavelet
threshold function with the parameter through the marine predator intelligent optimization
algorithm, determining the optimal
threshold function parameter, and carrying out
thresholding processing on the wavelet packet coefficients. Carrying out
inverse operation on the wavelet packet coefficient after
noise reduction to obtain an IMF component after
noise reduction; and 6, reconstructing the
signal dominant IMF component and the noise dominant IMF component which is denoised by the wavelet packet to obtain a denoised monitoring
signal.