Research on high-dose asphalt performance prediction and formula design based on machine learning and multi-objective optimization

Abstract With the increasing demand for green transportation infrastructure, High-Content Rubber Modified Asphalt (HCRMA) has garnered significant attention due to its superior pavement performance and potential for solid waste resource utilization. However, the “viscosity paradox” inherent in multiphase complex systems, combined with non-linear coupling effects among components, renders traditional “trial-and-error” formula design methods ineffective in balancing high-temperature stability, construction workability, and economic costs. To address these challenges, this study proposes a data-driven integrated research framework aimed at achieving precise performance prediction, mechanism elucidation, and multi-objective optimization for HCRMA. A standardized three-stage preparation process was established to construct an enhanced dataset comprising 600 experimental groups. By comparing four ensemble learning algorithms—Random Forest, Gradient Boosting Decision Tree (GBDT), XGBoost, and LightGBM—Gradient Boosting was identified as the optimal prediction model. This model achieved a coefficient of determination (R 2 ) of 0.9629 and a Root Mean Square Error as low as 0.3922 on the test set, effectively overcoming the accuracy bottlenecks of traditional linear models in handling multivariate non-linear mappings. The study introduces the Shapley Additive exPlanations (SHAP) game-theoretic framework to interpret the “black box” of machine learning. Mechanism analysis reveals that Crumb Rubber (CR) and Styrene-Butadiene-Styrene (SBS) are the core drivers for enhancing the softening point. CR exhibits a significant linear enhancement characteristic, while SBS demonstrates a non-linear mutation based on the “percolation threshold” when the dosage exceeds 0.60%, marking the formation of a continuous polymer network. Furthermore, interaction analysis quantifies the synergistic contribution of the T7 compatibilizer in improving interfacial compatibility and the geometric advantages of fine-mesh rubber powder in constructing a dense skeleton. Addressing the trade-off between performance and cost, a multi-objective optimization system was constructed based on the GBDT surrogate model and Genetic Algorithms to solve for the Pareto frontier. The study identified a performance return inflection point in the softening point range of 59–60 °C and recommended a "Balanced" formula (CR ≈ 29.6%, SBS ≈ 0.76%) that harmonizes technical indicators with economic viability. This scheme maximizes high-temperature performance benefits while effectively controlling costs. This research provides a theoretical basis for the intelligent design of HCRMA and offers new insights for the paradigm shift in the research and development of complex pavement materials.