Construction of glass fiber reinforced concrete proportioning model based on simulated annealing algorithm

The optimization of glass fiber-reinforced concrete (GFRC) mix proportions involves nonlinear interactions among multiple constituents. This study develops an improved simulated annealing (SA) framework coupled with a gradient boosting regression surrogate for four-objective mix-design exploration. The objectives are to maximize compressive and flexural strength while minimizing unit cost and cradle-to-gate embodied carbon. Practical bounds on the water-binder ratio and superplasticizer dosage are used as proxy feasibility screens; measured slump is not predicted or optimized. Using 612 curated mix-performance records, the surrogate achieved a test-set R² of 0.917 and an RMSE of 3.56 MPa for compressive strength, with a mean single-prediction time of 0.78 ms. On a separately held external subset of 60 records, the corresponding values were R²=0.894 and RMSE = 4.05 MPa. Under the reported operating budgets, the improved SA obtained an HV of 0.823 over 1,500 objective evaluations and a mean single-run time of 9.4 s. Sensitivity analyses indicated HV variations below 6% for ± 20% perturbations of the examined algorithm parameters. Four representative Pareto candidates illustrate high-strength, low-cost, low-carbon, and balanced preferences. The low-carbon candidate has a predicted compressive strength of 47.4 MPa and an estimated embodied carbon of 248 kg CO₂e m⁻³; it remains a computational candidate that requires confirmatory laboratory testing before engineering use. The framework therefore provides rapid decision support within the training-data domain rather than a substitute for experimental mix qualification.