基于改进神经网络的板桩结构可靠性分析

Sheet-pile Structure Reliability Analysis Based on Improved Artificial Neural Network

板桩码头是港口工程的一种常用结构,计算其可靠度指标对港口工程的安全意义重大。由于板桩结构的设计计算复杂,功能函数表达难度较大。Monte-Carlo法是解决此类问题的一种方法,但需要大量的抽样与数值计算,很不经济。人工神经网络模型可以用来逼近功能函数,在此基础上可平行地建立一次二阶矩法进行可靠度分析。但传统的BP神经网络模型有着容易陷入局部极小及预测精度低等问题。针对上述问题,引入Adaboost算法来改进BP神经网络模型,提出一种基于Adaboost的BP神经网络法来计算板桩结构的可靠度。以天津某板桩码头为例,采用新方法对板桩结构进行可靠度分析,并将计算结果与传统BP神经网络法及Monte Carlo法进行比较。结果表明:新方法的计算精度高于传统BP神经网络法,且计算结果与Monte Carlo法接近。

Sheet pile is a widely used port structure. Reliability index is of great significance to measure the safety of port engineering. However, the design and calculation process of sheet pile structure are complex,and its performance functionsare implicit. Monte Carlo method can be used to solve this problem. However, vast sampling and numericalcalculation are needed, which is uneconomic. Artificial neural networkcan approach the performance function. The first-order second-moment method can be established to carry out the reliability analysis. However, traditional BP neural network is easy to fall into minimum and has lower accuracy. To solve these problems, Adaboost algorithm was introduced to improve BP neural network. This paper proposed an Adaboost-based BP neural network method to compute the reliability of sheet-pile structure. A sheet-pile wharf in Tianjin is taken as an example to verify the feasibility of the Adaboost-BP neural network method. Calculating results are compared with those by both traditional BP neural network method and Monte-Carlomethod. The results indicate that the reliability indexes calculated by new method are with higher accuracy than those by traditional BP neural network method and are similar to those by the Monte-Carlo method.