In dredging engineering，determination of optimal velocity of slurry transportation usually depends on empirical formulae.Changeable working conditions lead to low prediction accuracy.This paper uses Durand model to model the resistance loss under different test conditions，and adjusts the model constants based on test data to improve prediction accuracy of Durand model.The paper uses Gaussian process regression method in the modelling，with analysis of influence of training samples numbers on prediction results.The paper proposes a velocity optimization method based on resistance-velocity model，and carries out comparative tests.The results show that prediction accuracy of pipeline resistance model established by Gaussian process regression method is higher，which can be higher than 0.97.Critical velocity can also be real-time updated in terms of changes of pipeline resistance (concentration)，providing a more effective optimization method to determine the optimal slurry velocity.