基于GPR-ARIMA-GA模型的高粘土心墙堆石坝参数反演分析

Parameter Inversion Analysis of Clay Core Wall Rockfill Dam Based on GPR-ARIMA-GA Model

以某粘土心墙坝为例,提出了基于GPR-ARIMA-GA模型结合有限单元法反演高粘土心墙堆石坝材料参数的方法,利用高斯过程回归非线性模型(GPR)来表征材料参数与坝体沉降量之间非线性关系,从而在搜寻真实坝体材料参数时减少了有限元程序样本计算次数,提高了反演效率;为进一步提高GPR拟合精度,提出应用ARIMA模型对拟合误差进行修正;在构建GPR-ARIMA模型过程中,采用遗传算法(GA)来优化GPR模型超参数;在建立GPR-ARIMA模型后,二次应用GA全局搜索坝体材料参数的真实值。实例应用结果表明,该方法极大地减少了有限元程序计算量,且能够准确、快速搜索得到坝体真实材料参数,将最优材料参数代入有限元程序计算得到沉降量与实测值较为接近,验证了该方法可行、有效。

The Gauss process regression-ARIMA-genetic algorithm model and finite element method were proposed to inverse the material parameters of clay core rockfill dam.In this method,Gaussian process regression model(GPR)was used to characterize the nonlinear relationship between material parameters and dam settlement,which reduces the calculation times of finite element program and improves the inversion efficiency.In order to further improve the fitting accuracy of GPR,ARIMA model was proposed to correct the fitting error.In the process of constructing GPR-ARIMA model,genetic algorithm(GA)was used to optimize the hyper-parameters of GPR model.After GPR-ARIMA model was established,GA was applied to search the real value of dam material parameters.The example results show that the method greatly reduces the calculation amount of the finite element method,and can accurately and quickly search for the real material parameters of the dam body.The calculated settlement and change trend are close to the measured value,which shows the effectiveness of the method.