The phase-sensitive optical time-domain reflectometer (Φ-OTDR) has many advantages such as high sensitivity, fast response, and large dynamic range, and it has been widely used in fields such as border security, structural health, and oil and gas pipeline monitoring. Due to polarization attenuation, interference attenuation, and Rayleigh backscatter noise, the signal-to-noise ratio of vibration positioning signals is low. Obtaining vibration signals with good signal-to-noise ratio is a key to improving the positioning accuracy of the Φ-OTDR.The proposed method is a complementary set empirical mode decomposition (CEEMD) approach for noise reduction in Φ-OTDR systems. CEEMD essentially resolves the mode aliasing and end effects of EMD by adding white noise (Fig.3). In this paper, multi-scale arrangement entropy (MPE) is designed to evaluate the randomness of the IMF component after CEEMD decomposition (Fig.4), and an improved wavelet threshold function is used to accurately deal with noise (Eq.15). Finally, the multiverse optimization algorithm (MVO) is designed, and the SNR is used as the fitness function to optimize the parameters of the algorithm (Fig.6), so as to achieve the optimal effect of the algorithm. After the vibration signal is simulated by PZT and denoised by the algorithm designed in this paper, the signal-to-noise ratio is used to represent the final denoising effect (Tab.3).A Φ-OTDR system for heterodyne coherence detection was actually built (Fig.9). The 10.14 km optical fiber end was connected to PZT, and the 600 Hz vibration signal was simulated by PZT to verify the effectiveness of the algorithm. The signal collected by the acquisition card is shown in Fig.10(a), and the vibration position diagram obtained after moving the difference is shown in Fig.10(d). The parameters of each algorithm were optimized by MVO algorithm, and the signal was decomposed by CEEMD algorithm (Fig.10(d)). The decomposed IMF component was shown in Fig.12. The randomness of each component was calculated by MPE algorithm (Tab.2). The value of IMF₁₅ is low, which can be determined as the signal component, and the other components are denoised signals. The denoised signals are denoised by the improved wavelet threshold algorithm, and finally the signals are superimposed to obtain the final denoised signals (Fig.13). In order to further illustrate the effectiveness of the algorithm, the same set of experimental data were used in the Φ-OTDR system to carry out denoising experiments and comparisons with the proposed algorithm respectively by EMD-PCC, VMD-NWT and CEEMDAN. In the experiment, PZT was used to simulate low-frequency signals of 10, 20, 30, 40 Hz, medium-frequency signals of 200 Hz and 500 Hz, and high-frequency signals of 1.2 kHz and 1.4 kHz. Figure 14-16 shows the vibration location curves of the 10 Hz, 200 Hz, and 1.2 kHz sinusoidal drive signals after noise removal. Figure 14(a)-Fig.16(a), Fig.14(b)-Fig.16(b), Fig.14(c)-Fig.16(c), Fig.14(d)-Fig.16(d) are denoising results of EMD-PCC, VMD-NWT, CEEMDAN and the algorithm in this paper, respectively. The SNR of vibration signal with a frequency of 10 Hz is 6.28, 4.06, 5.92, 8.88 dB, respectively, while that of vibration signal with a frequency of 200 Hz is 20.52, 23.12, 29.62, 30.26 dB, respectively. The SNR of vibration signal with frequency of 1 200 Hz is 6.86, 6.56, 7.52, 11.9 dB respectively. The results show that compared with other algorithms, the proposed algorithm is effective and slightly improved in the localization and denoising process of low frequency, medium frequency and high frequency signals, which provides a new optimization direction for signal denoising.A noise reduction method based on CEEMD decomposition is proposed to improve the signal-to-noise ratio of Φ-OTDR system. The signal is decomposed by CEEMD algorithm, each component is screened by MPE algorithm, the filtered signal is denoised by improved wavelet threshold algorithm, and the parameter design of the three algorithms is optimized by MVO algorithm. The SNR of low frequency 10 Hz, medium frequency 200 Hz and high frequency 1.2 kHz vibration events is improved to 8.88, 30.26, 11.9 dB respectively, which is superior to EMD-PCC, VMD-NWT and CEEMDAN algorithms. In particular, the positioning accuracy of vibration signals with a frequency of 200 Hz is analyzed, and the positioning accuracy of CEEMD algorithm is higher and the phase information demodulation is more accurate. In summary, the CEEMD denoising algorithm proposed in this paper has effective denoising ability for vibration signals in different frequency segments, and can optimize signal quality to a certain extent. It can not only locate the vibration position well, but also accurately obtain its frequency information, providing a new idea for the denoising of Φ-OTDR system.