Prediction Of Flexible Cylinder Vortex-induced Vibration Based On Machine Learning Method

Marine risers and submarine pipelines are important components of oil and gas production systems.As oil and gas resources are mined from inland to deep sea,the strength of these marine structures is gradually receiving attention.Vortex-induced vibration phenomena are an important cause of fatigue damage to such elongated flexible cylindrical structures.At present,the research on vortex-induced vibration of flexible cylinders is not thorough enough,especially for the complex conditions prevalent in engineering practice.Experts and scholars have tried to predict their vibration response in a variety of ways,but have not yet obtained an accurate and effective method.In this paper,a new vortex-induced vibration prediction method is tried.The machine learning method is applied to it.Based on the experimental data,the vortexinduced vibration response prediction model of flexible cylinder is constructed.The main research contents are as follows:(1)The BP neural network method,support vector machine method and Gaussian process regression method are used to construct the prediction model of the vortexinduced vibration response of the flexible cylinder under vertical flow.The input data of the model is the flow velocity,axial force and spatial position of the pipe length direction.This model can predict the displacement and frequency response in both inline direction and cross-flow direction.(2)The above three machine learning methods are used to construct the prediction model of the vortex-induced vibration response of the flexible cylinder under inclined flow.The input data of the model increases the inflow dip angle compared with the vertical working condition.The prediction of in-line and cross-flow displacement and frequency response is realized.The machine learning prediction model was tested by test set and found that the prediction results of each model can reach a high degree of accuracy.Comparing the advantages and disadvantages of the three machine learning methods and the threedimensional map of the forecast results,it is considered that within the working conditions and variables discussed in this paper,the support vector machine is the most suitable for the vortex-induced vibration prediction of flexible cylinders among the three methods.