RBF神经网络可靠度分析方法在边坡稳定性研究中的应用

Application of RBF neural network reliability analysis method in slope stability research

针对边坡稳定性可靠度分析,当状态函数无法显式表达且传统计算方法求解复杂问题困难时,提出一种基于ABAQUS和粒子群优化径向基函数神经网络的可靠度分析方法。基于ABAQUS的强度折减方法计算所选随机变量对应的安全系数,利用径向基函数神经网络的数据拟合功能,建立模型并映射出安全系数和随机变量之间的关系,构造响应面功能函数;利用蒙特卡罗生成的大量随机样本代入功能函数得到相应的安全系数,进而计算边坡的失稳概率和可靠度指标来反映边坡稳定性。研究结果表明:相对于传统方法,本文方法计算效率更高、误差更小,适合实际工程应用。

For the reliability analysis of slope stability,when the state function cannot be explicitly expressed and the traditional calculation method is difficult to solve complex problems,a reliability analysis method based on ABAQUS and particle swarm optimization radial basis function(RBF)neural network was proposed.The strength reduction method based on ABAQUS was used to calculate the safety coefficient corresponding to the selected random variable.The data fitting function of the RBF neural network was used to establish the model and map the relationship between the safety coefficient and the random variable,so as to construct the response performance function.The large number of random samples generated by Monte Carlo were substituted into the performance function to obtain the corresponding safety coefficient,and then the instability probability and reliability index of the slope were calculated to reflect the slope stability.The results showed that compared with the traditional methods,this method was more efficient and less error-oriented,which is suitable for the practical engineering application.