Sequential Support Vector Machine For Structural Uncertainty Analysis And Design Optimization

The structural uncertainty analysis and design optimization is one of the most attracting research branches in recently years,especially in the field of aeronautics and astronautics.Using approximate models instead of the physical model can achieve the balance of computational accuracy and cost,and improve the efficiency.Support vector machines(SVM)have been reported that they possess the capacity of the good generalization and learning with small sample size,and resolve the highdimensional and local extreme issues.Thus,they have the advantage of dealing with non-linear and high-dimensional problems.The single-step modeling method may not often achieve a good balance of computing resources and approximate accuracy.Two sequential SVM modeling methods are proposed to address this issue in this thesis.At the same time,combining the Monte Carlo method and the improved cross-entropy method,and were applied on structural uncertainty analysis and design optimization.The primary contributions are summarized as follows:1)Based on the single-step modeling method,the performance of LCVT sampling method iscomprehensively studied for structural reliability analysis.2)Using the maximum curvature and the maximum entropy criterion as the updating criteria,two sequential SVM modeling methods are proposed,combined with the principle ofmaximizing the minimum distance,which is designed to alleviate the cluster level of newsamples.Sequential modeling overcomes the limitations of single-step modeling,improvescomputational efficiency,and reduces computational costs.3)Based on the sequential SVM method and Monte Carlo method,we investigate its applicationon structural reliability and reliability sensitivity analysis.The hybrid method can obtainfailure probability and sensitivity ranking in the same time.4)Based on SVM and improved cross-entropy method,a combined optimization algorithm forstructural design optimization is proposed,which widens the application area of cross-entropymethod.