关键词: 岩土工程, 可靠度分析, 隐式功能函数, BP神经网络, 四阶矩, 最大熵

Abstract: Strong nonlinear and implicit performance functions are normally encountered when geotechnical engineering is complicated. An artificial neural network (ANN)-based four order moments method, allowing to make full use of statistical information of basic random variables, is proposed to address this issue. In this method, the relation between the basic random variables and responses is established to approximate the implicit performance function; then the values of the implicit performance function in the mean point as well as partial derivatives can be obtained for the four order moments method. The first four order moments are approximated by expanding the implicit performance function into a second-order Taylor series; the corresponding higher-order moments can be estimated via Pearson curves. In restraint of the assigned moments of the implicit performance function, the probability density function of the performance function can be expressed by applying the maximum entropy theory. Subsequently, the reliability results are easily worked out. From the comparison between the results of numerical example and practical engineering calculated by different methods, it is shown that the proposed method is valuable and applicable to reliability analysis of complex geotechnical engineering involving implicit performance functions.

Key words: geotechnical engineering, reliability analysis, implicit performance functions, BP neural network, four order moments, maximum entropy