In view of the influence of long delay on information propagation between mobile sensor nodes caused by the complexity and dynamics of underwater environment in underwater wireless sensor networks, and the large node positioning error caused by this problem, this paper proposes an underwater mobile node positioning algorithm based on CNN-LSTM sound velocity prediction. First, the sound velocity data set is divided by K-fold cross-validation method, and then the CNN-LSTM hybrid model is constructed and trained by using the feature extraction capability of CNN and the sequence modeling capability of LSTM. This model can capture both spatial and temporal information of sound velocity data set, thus improving the prediction accuracy of sound velocity data set. Secondly, in the process of mobile node positioning, the sound velocity value predicted by CNN-LSTM model is used for TDOA ranging, and the TDOA ranging value is corrected. Finally, the modified ranging values are used to adaptively select different ranging positioning methods for unknown nodes under different node densities according to the number of reference nodes, so as to achieve accurate positioning of underwater mobile nodes. Experimental results show that under the same beacon node, the mean positioning errors of MCLS proposed in this paper are reduced by 46.96%, 39.93%, 27.64% and 15.24%, respectively, compared with SLMP, DMP, NDSMP and BLSM.