Reliability-Based Design Optimization Research Of Ship Structure Based On Ensemble Learning Method

The reliability calculation of ship structure includes reliability analysis and reliability-based design optimization.Traditional ship structure optimization only considers some certainties of design variables,while ship reliability-based design optimization considers some uncertainties of design variables,such as size,material properties and load.Taking into account the economic and safety of the ship structure,the reliability research is one of the most advantageous research directions of ship technology.Aiming at the problems of multi-dimensional structural variables,high degree of nonlinearity and complicated calculation,the key is to find efficient and accurate calculation methods of solving the problem.As one of the research hotspots of machine learning,ensemble learning method is widely used in data mining,image processing and speech recognition.In this paper,ensemble learning method is applied to reliability technology research of ship structure,and ensemble learning method is improved for the problems in the application process,which improves the calculation accuracy and efficiency of ship reliability analysis and reliability-based design optimization.The research progress of ship reliability analysis,reliability-based design optimization,ensemble learning method and surrogate model technology is reviewed.Based on the previous studies,the theory of base learner combination strategy and three frameworks are summarized.The design of experiments,the common surrogate models(polynomial response surface model,Kriging model,radial basis function model and artificial neural network model)and technology route of static/dynamic surrogate model are explained,which provides technical support and theoretical guarantee for this paper.In the calculation of ship reliability analysis,the Gradient Boosting Decision Tree(GBDT)in the ensemble learning method is used as the surrogate model,and aiming at the problem that the proportion of sample points near the failure surface is small in the data set,the GBDT algorithm has been improved by using the Smooth algorithm.By oversampling a few samples,the number of sample points near the failure surface is improved.Furthermore,a high degree of fitting to the Limit State Function(LSF)is achieved.The improved GBDT method is combined with the Monte Carlo to calculate the failure probability and reliability index.Two examples are used to verify the applicability and superiority of the improved GBDT method,and the method is applied to the reliability analysis of the ultra large container ship’s lashing bridge.During the process of the reliability-based design optimization,in order to solve the problem that low computational efficiency and difficile convergence due to the high degree of nonlinearity of the ship structure,this paper takes advantage of the Xgboost algorithm in the ensemble learning method,and improves Xgboost by combining the Smote algorithm on the sample.Then put improved Xgboost algorithm and Adaptive Simulated Annealing(ASA)into the Single Loop Approach(SLA).Two mathematical examples are used to verify the effectiveness of the improved Xgboost method,and the method is applied to the reliability-based design optimization of the ultra large container ship’s lashing bridge,and the calculation time is greatly reduced under the premise of ensuring accuracy.