Research On Surrogate Model Technology For Prediction Of Mechanical Impedance Of A Typical Foundation In The Submarine Structure

As an important characterization parameter for structural vibration,mechanical impedance is widely used in the study of dynamic response and vibration transmission of foundations in the submarine structures.When designing the foundation structures,it is necessary to analyze the impedance characteristics under different conditions.Usually,the optimization process embedded in the finite element analysis is very inefficient.Surrogate model technology,as a pure mathematical model,directly fits the mapping between input and output.Compared to the finite element analysis,it is more efficient and less time-consuming.In this paper,the application of surrogate model technology in fast prediction of mechanical impedance of a typical foundation in the submarine structure is studied.The main work and conclusions are listed as follows:(1)A comparison among response surface method,support vector regression method,back propagation neural network method,radial basis function method and kriging method is conducted for their fitting accuracy of mechanical impedance of a typical foundation in the submarine structure under different quantities of sample points.The results show that the kriging method has the best applicability to the problem when the fitting accuracy reaches convergence.(2)A local error prediction surrogate model is constructed by using the cross-validation error of sample points to calculate the local error of the sub-surrogate model at prediction points as the weighting coefficient calculation standard.Based on this,an adaptive hybrid surrogate model is built to avoid the risk of the modeling method selection strategy caused by the difference of the single surrogate model.The effectiveness of the proposed method is verified by a series of test functions and its application to the prediction of impedance characteristics of the foundation.The results show that the adaptive hybrid surrogate model has excellent applicability and can provide better prediction accuracy than the single surrogate model.(3)Combined with the characteristics of mechanical impedance and practical engineering background,a targeted data preprocessing method is proposed.The envelope of impedance data is obtained by graham method and then applied to build surrogate models.The results show that the proposed method can effectively improve the accuracy of the impedance prediction surrogate model.The singular value decomposition is applied to the construction of kriging surrogate model,and a new method for determining effective rank order is proposed.The feasibility of the method is illustrated by mathematical examples.The results show that although the method lowers the prediction accuracy of the model to some extent,it can effectively reduce the cost of modeling.With the singular value decomposition and the data pre-processing method,the kriging method is used to construct the relational surrogate model to predict the mechanical impedance of the foundation.Under the premise of satisfying the engineering precision,the construction cost of the surrogate model has been significantly reduced.The research work in this paper provides a useful guidance for the design of foundations in the submarine structure.