Aiming at the problem of interference noise in the sampled ultra-weak fiber grating reflection spectrum, a hybrid network denoising algorithm for large-capacity ultra-weak sensor networks is proposed. The algorithm combines the complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN) and the wavelet packet noise reduction algorithm. After the measurement signal is processed by the wavelet packet threshold denoising, the preliminarily processed signal is subjected to CEEMDAN. The eigenmode function(IMF) component and the residual component obtained by the decomposition are superimposed and reconstructed according to the order of the correlation coefficient, so as to obtain the denoised signal. Under the conditions of four different signal-to-noise ratios and 20 consecutive repeatability tests, the experimental results show that the signal-to-noise ratio of the optimization algorithm combined with CEEMDAN and wavelet packet noise reduction can be increased by 3 dB to the greatest extent, the root mean square error can be reduced by 0.245 dB and the correlation coefficient can be increased by 0.0353 at most. The three indexes have been significantly improved.