基于GeoStudio建模与集成学习的土石坝渗流预测模型OA

Conventional seepage prediction models lack effective integration with machine learning meth-ods at the modeling level,resulting in incompetence in achieving both physical interpretability and non-linear representation.To address this challenge,an improved seepage prediction model integrating GeoS-tudio modeling and Stacking ensemble learning is proposed to enhance prediction accuracy.This model first utilizes the SEEP/W module of GeoStudio for numerical simulation under"saturated only"condi-tions.Subsequently,based on the Stacking ensemble learning framework,it fits the residuals between the numerical simulation outputs and the monitoring data,thereby improving prediction performance through residual correction.Compared with conventional seepage prediction models for earth-rockfill dams,the R2 on the training set increases from 0.86 to 0.96,and on the test set from 0.79 to 0.89.Other model e-valuation metrics(MAE,MAPE,RMSE)also show significant improvement.The improved model in-tegrates the strengths of software modeling and machine learning,merging the physical interpretability of conventional statistical models with the flexibility of algorithmic approaches.It provides a novel modeling approach to seepage prediction for earth-rockfill dams.