The Study And Application Of Approximate Method And Stochastic Analysis For Multidimensional Models With Correlated Parameters

In the research of Semi-rigid connection, the main interest is the determination of constitutive relation between moment and rotation. However, this curve relation, whether experimental research or theoretical research, both under the assumption that the structural design variables were determined completely, that is the impact of the whole structure response changed based on the design variables changed separately, ignoring the relations of the design variables. This relation function obtained by fitting the changes of single factor would be counterintuitive in practical application. Also, parameter correlation widely existed in engineering analysis and the assessment of the decision field, large deviations and even erroneous analysis results would be obtained for without considering the correlation of parameters, so there is an important practical significance to research not limit to the semi-rigid joints by considering the correlation of parameters.In the process of the semi-rigid joint research, there are many problems for solving the correlations of parameters, including the limitations of the finite element analysis software and the improvement of the algorithm. In the process of the finite element analysis, the restriction of the ranges of parameters by software has becoming the bottleneck in the field of the moment-rotation relation of determined sizes extending to the same type of joints. In the probabilistic design analysis of finite element software, the finite element simulation would be increased by the increase of sampling number for the sampling process of Monte Carlo, so a large amount of computer time would be consumed, this method is difficult to achieve for the moment-rotation curve of the same type semi-rigid joints. It is necessary to study the improvement method for the above problems by the mathematical methods.According to the Monte Carlo correlation probability design problems and the analysis of random problems which derived from the research of the joints, the main research is divided into the following sections:In this paper, firstly, the method of correlation parameters transformed to independent parameters equally in Monte Carlo sampling would be solved and the core is making the original relation factors to the uncorrelated factors, this parameters transformation method would also play an important role in other comprehensive evaluation programs which considering the related factors. Secondly, the hybrid neural network possessing significant learning capacity and generalization capability at a small amount of information is used to construct the approximation model which has high degree matching with the meta-models. The networks would be optimized by the improved chaotic particle swarm for high efficiency and accuracy. The stochastic analysis of the structure is conducted by Monte Carlo, and a new method of sensitivity is put forward to measure the coefficients of random parameters employing the global sensitivity method. Finally, the method of stochastic analysis and the calculation of sensitivity would be applied to the study of moment-rotation model in semi-rigid joints, the initial rotational stiffness of the semi-rigid joints which fitted by the above methods would be useful for engineering applications.