STOCHASTIC SENSITIVITY ANALYSIS METHOD BASED ON HYBRID NEURAL NETWORK OF IMPROVED CHAOTIC PARTICLE SWARM AND MONTE CARLO

In order to study the random effects of uncertain parameters on the performance of structure mechanics, the hybrid neural network possessing significant learning capacity and generalization capability at a small amount of information is used to construct the relativities of complex functions between the input and output samples. The networks would be optimized by the improved chaotic particle swarm for high efficiency and accuracy. According to the proposed method, global sensitivity coefficients of random variables can be calculated through random analysis by Monte Carlo simulation. In order to verify the feasibility of the proposed method, several examples are analyzed including mathematical examples and engineering examples. The results of these examples indicate that the proposed algorithm increases the precision and the response distributions of engineering examples could be reflected by the fitting probability density curves, meanwhile the structure response sensitivity and correlation could be better reflected based on the method of this paper.