基于降阶投影的初始渗流场反演方法

Inversion method of initial seepage field based on reduced order projection

为了解决传统基于降阶模型的参数反演方法的高耗时问题,提出了一种改进的降阶反演方法(IterROM),用于执行快速参数反演任务。不同于传统降阶反演方法先训练后反演的流程,该方法将参数反演与降阶模型训练紧密融合,使用训练中的降阶模型代替原始模型进行参数反演,同时将反演结果应用于降阶模型的训练。该方法在参数反演阶段使用观测资料和后验误差共同组成目标函数,并使用降阶模型解与全阶模型解的最大节点误差作为迭代终止精度。经过某工程初始渗流场反演模型的测试,发现IterROM相比传统方法精度高、计算成本低,可用于快速执行初始渗流场的反演任务。

In order to solve the time-consuming problem of traditional parameter inversion based on reduced order models,this study proposes an improved reduced order inversion method(IterROM)for performing fast parameter inversion tasks.Unlike the traditional reduced order inversion method,which trains first and then inverts,this method tightly integrates parameter inversion with reduced order model training,that is,using the reduced order model in training to replace the original model for parameter inversion,and applying the inversion results to the training of the reduced order model.This method uses observation data and a posterior error to form the objective function during the parameter inversion stage,and uses the maximum node error between the reduced order model solution and the full order model solution as the iteration termination accuracy.After testing the initial seepage field inversion model of a certain project,it was found that IterROM has higher accuracy and lower computational cost compared with traditional methods,and it can be used to quickly perform the inversion task of the initial seepage field.