对于功能函数不能显式表达的边坡类等复杂结构,不方便采用传统的可靠度理论计算其失效概率,因而采用BP神经网络模型代替边坡失效的隐式函数,通过神经网络强大的非线性拟合能力进而构造边坡抗滑稳定可靠度分析的功能函数,并结合蒙特卡洛方法计算边坡抗滑稳定的失效概率。采用正交试验选取网络训练所需的样本点,利用有限元强度折减法计算样本点的抗滑稳定安全系数。通过算例研究了以内摩擦角和黏聚力为随机变量的堤防边坡抗滑稳定的失效概率,验证了该方法操作的简便性与结果的合理性。

It is not convenient to calculate failure probability using traditional reliability theory for complex structures such as levee slope whose performance function cannot be explicitly expressed. In this paper, neural network model is used to replace the performance function of slope and the Monte-Carlo method is used to calculate the failure probability. The samples for training network are generated through orthogonal trial, and the safety factors of the samples are calculated by the strength reduction method of finite element method (FEM). The failure probability is calculated by constructing performance function by using the strong non-linear fitting ability of neural network. Through an example calculation, the failure probability of levee slope is studied with internal friction angle, cohesion force and elastic modulus as random variables. This method is proved to be simple and reasonable.