Accurate identification of fluctuating vibration gas-liquid two-phase flow pattern is of great significance for the safe and stable operation of nuclear power platform under floating vibration. Through comparing the differential pressure signals and the corresponding spectrums in static and fluctuating vibration pipelines, it is found that the differential pressure signals in fluctuating vibration pipelines have larger fluctuation amplitude and contain more frequency components, and both flow patterns have dominant frequency, which is the fluctuating vibration frequency. Aiming at the complexity of the pressure difference signals of gas-liquid two-phase flow in fluctuating vibration state, complete ensemble empirical mode decomposition with adaptive noise ( CEEMDAN) and ensemble empirical mode decomposition (EEMD) are used to decompose the pressure difference signals after wavelet de-noising. It is found that CEEMDAN can reduce the mode components and obtain more effective components at the same time. Through calculating the Spearman correlation coefficient, the IMF components that have symbolic meaning are selected to perform Hilbert transform, and the energy is calculated and used as the eigenvalue. Probabilistic neural network is used to identify the flow pattern. The results show that using CEEMDAN to perform mode decomposition, combining with probabilistic neural network, the accuracy of the identification method is 95. 83% , and this method can be used to identify the flow pattern of gas-liquid two-phase flow under fluctuating vibration.