Intelligent inversion analysis of seepage parameters for deep overburden dam foundations based on an improved grey wolf optimization algorithm

Abstract

The seepage parameters of dam foundations are essential for conducting three-dimensional seepage safety analyses. However, traditional methods often fail to achieve efficient and accurate parameter inversion because of the challenges posed by complex geological conditions and limited in situ measurements. To address this issue, a novel dual-layer collaborative intelligent inversion framework was proposed. This approach integrated an orthogonal experimental design with finite element modeling to construct a training dataset. An Improved Chaotic Grey Wolf Optimization (ICGWO) algorithm was then employed to optimize the hyperparameters of a Light Gradient Boosting Machine (LightGBM), resulting in the development of a high-precision surrogate model. Within a reasonable range of seepage parameter values, ICGWO was further applied to explore the solution space of the surrogate model, which enabled intelligent inversion of permeability parameters. A case study conducted on a deep overburden dam foundation at the Y Hydropower Station demonstrated that the ICGWO-LightGBM model accurately predicted borehole water levels. Moreover, ICGWO exhibited notable efficiency in optimizing stratified permeability parameters. Forward simulations using the permeability coefficients derived from the inversion process yielded a maximum absolute error of 7.92 m and a relative error of only 0.26 % in borehole water level predictions. The simulated seepage field displayed physically consistent distribution patterns that aligned with the typical mountain-type seepage behavior, confirming the accuracy, robustness, and engineering applicability of the method. The proposed approach can provide an efficient and innovative pathway for dam seepage safety assessment under complex geological conditions, offering considerable theoretical value and broad potential for engineering practice.