基于SMOTE算法的船体结构极限状态代理模型研究

Research on the Agent Model Technology for Ship StructuresLimit State Based on Smote Algorithm

建立典型集装箱船舱段结构参数化有限元模型,基于SMOTE过采样算法,增加样本数据中失效面附近样本点的数量,分别采用BP神经网络、径向基神经网络以及支持向量机三种代理模型技术,结合十折交叉验证法调试模型参数以提高模型的泛化能力,完成构建船体舱段结构极限状态代理模型并对其进行测试与分析,获得了效果与泛化能力均较为优良的船体结构极限状态高精度代理模型.结果表明:经SMOTE算法处理的样本数据结合BP神经网络代理模型技术,能够在不增加有限元计算任务量的同时提高船体结构极限状态代理模型的精度.

A parameterized finite element model of typical container cabin structure was established.Based on SMOTE oversampling algorithm,the number of sample points near the failure surface in the sample data was increased.Three kinds of proxy model technologies,namely BP neural network,radial basis function neural network and support vector machine,were adopted respectively,and the model parameters were debugged by ten-fold cross validation method to improve the generalization ability of the model.Then,the hull cabin structure limit state proxy model was constructed,tested and analyzed,and a high-precision proxy model with excellent effect and generalization ability was obtained.The results show that the sample data processed by SMOTE algorithm combined with BP neural network proxy model technology can improve the accuracy of the limit state proxy model of hull structure without increasing the workload of finite element calculation.