Multi-Objective Performance Prediction and Proportion Optimization of Composite-Modified Cement-Based Grouting Materials Using a Bayesian-Optimized Gradient Boosting Tree Model

Earth-rock dams are prone to leakage hazards during long-term service, and conventional grouting materials used for seepage remediation are susceptible to dispersion and loss under flowing-water conditions. In addition, it remains difficult to simultaneously optimize fluidity, anti-dispersion stability, and mechanical strength. To address these issues, a composite-modified grouting material system composed of superfine cement (SC), graphene oxide (GO), polyacrylamide (PAM), and expansive agent (EA) was developed in this study. Orthogonal experiments were first conducted to obtain multi-performance datasets, including fluidity, anti-dispersion stability, and compressive strength. Based on these datasets, a data-driven optimization framework integrating a Bayesian optimization-based least squares boosting model (BO-LSB), Criteria Importance Through Intercriteria Correlation (CRITIC), and Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) was established for multi-performance prediction and optimal mix proportion design. The results indicated that the recommended mix proportion was W/C = 0.8, GO = 0.025%, PAM = 1.2%, and EA = 8%. The relative prediction errors between the BO-LSB model predictions and the mean values of three parallel validation tests ranged from 1.10% to 4.00%, with an average relative error of 2.41%, indicating good agreement between the predicted and measured values. Compared with the control group, the optimized mix exhibited enhanced anti-dispersion stability and mechanical performance while maintaining acceptable fluidity. Scanning electron microscopy (SEM), energy-dispersive spectroscopy (EDS), and X-ray diffraction (XRD) analyses showed that the composite system promoted the formation of hydration products and developed a denser and more continuous cementitious network structure, which was consistent with the macroscopic performance improvement. These findings provide an effective data-driven approach for the multi-objective design and optimization of grouting materials for seepage remediation in earth-rock dams under flowing-water conditions.