Research On The Rapid Prediction Model Of Submarine Resistance Based On Neural Network And The Application Of Submarine Shape Optimization

In the process of submarine optimization design,how to predict the resistance performance of submarine quickly and accurately is very important.At present,the commonly used submarine resistance performance prediction method is the Computational Fluid Dynamics(CFD)method.The design testers evaluate multiple schemes through CFD method to select the best boat type scheme.However,CFD method directly predicts the viscosity resistance of the entire hull,which leads to a long calculation period and high requirements for user experience.In order to find a method that can not only guarantee the accuracy of resistance solution but also realize the rapid solution of submarine resistance,this paper proposed a rapid prediction model of submarine resistance based on multi-layer BP neural network.The neural network model was applied to submarine shape optimization,and the effectiveness and practical value of the neural network model were verified.First of all,by analyzing domestic and foreign research characteristics,Suboff naked hull model was selected as the research object.This paper used the Optimized Latin Hypercube Design(OLHD)method and the batch geometry reconstruction technology of the submarine to expand the submarine sample;Secondly,in order to solve the problem of long calculation cycle of STAR CCM+software,this paper proposed a resistance mapping idea,which is to map the resistance of the longitudinal section slice model of the Suboff naked hull to the resistance of the threedimensional naked hull.In this paper,STAR CCM+ software was used to calculate the underwater resistance of the Suboff naked hull and the slice model,then the submarine resistance performance database was obtained,which was used as the sample data of the resistance mapping BP neural network and the resistance prediction neural network model.Then,based on the principles and algorithms of the BP neural network,combined with the acquired submarine parameters and resistance data,the resistance mapping BP neural network was constructed.For the two kinds of samples,small-amount high-precision samples obtained from the bare hull and low-computational-complexity-considerable-accuracy samples obtained from the slice model,two resistance prediction neural network models were trained respectively.Then,the gradient test method is used to gradually increase the scale of low-computational-complexity-considerable-accuracy samples to improve the prediction accuracy of the neural network.Finally,this paper obtained a submarine resistance rapid prediction neural network model with high fitting accuracy and strong generalization ability.Aiming at verifying the effectiveness and practical value of the submarine resistance rapid prediction model,this paper combined the submarine resistance rapid prediction neural network model and genetic optimization algorithm to construct a submarine shape rapid optimization system.To avoid the premature phenomenon of the algorithm,improve the convergence stability and rapidity of the optimization algorithm,this paper used OLHD to initialize the population distribution,at the same time,improved the calculation method of the crossover probability and the mutation probability,then obtained an improved adaptive genetic algorithm.Finally,the optimization of Suboff submarine with the best resistance value at underwater cruising speed(1.599m/s)as the goal was completed.At the same time,this paper used CFD method to further verify the effectiveness of the neural network prediction results,analyzed the superiority of the optimization scheme.