Abstract
Accurate identification of mesoscopic parameters is critical for understanding the cracking and failure mechanisms of hydraulic concrete and for improving the reliability of numerical simulations. Traditional trial-and-error methods for parameter calibration are inefficient and often lack robustness. To address this issue, this study proposes a novel inversion method combining Support Vector Regression (SVR) with a Hybrid Grey Wolf Optimization (HGWO) algorithm. First, a mesoscopic simulation dataset of three-point bending (TPB) tests was constructed using 3D numerical models with varying mesoscopic parameters. Then, an SVR-based surrogate model was trained to learn the nonlinear mapping between mesoscopic parameters and load-CMOD (Crack Mouth Opening Displacement) curves. The HGWO algorithm was employed to optimize the SVR hyperparameters (penalty factor C and kernel coefficient g) and subsequently used to invert the mesoscopic parameters by minimizing the discrepancy between experimental and predicted CMOD values. The proposed method was validated through inversion of the mortar parameters of a tertiary hydraulic concrete beam. The results demonstrate that the HGWO-SVR model achieves high prediction accuracy (R(2) = 0.944, MAE = 1.220, MAPE = 0.041) and significantly improves computational efficiency compared to traditional methods. The simulation based on the inversed parameters yields load-CMOD curves that agree well with experimental results. This approach provides a promising and efficient tool for mesoscopic parameter identification of heterogeneous materials in hydraulic structures.