Synergistic regulation of strength, durability, and alkalinity in reef-analogue ecological concrete via machine learning-enabled inverse design

To enhance the multi‑objective synergistic optimization efficiency for reef‑analogue concrete, this study develops a composite binder system of cement, FA, SF, GGBS and desulfurized gypsum (DG), where FA, SF and GGBS act as supplementary cementitious materials and DG functions as a sulfate activator. A complete framework integrating experimental analysis, data‑driven modeling and inverse design is established. Orthogonal experiments suggest DG and cement govern strength development, impermeability enhancement and alkalinity regulation, identifying 6% DG as the optimal dosage within the studied orthogonal design space to balance mechanical performance and marine adaptability. Based on 2845 multi‑source data records, multi‑performance prediction models are built: XGBoost for compressive strength and slump, BPNN for chloride migration coefficient, and SVR for pH, yielding test‑set R 2 values of 0.9644, 0.8673, 0.9118, and 0.9549, respectively. SHAP and PDP analyses further suggest DG dosage, water‑to‑binder ratio and cement content are the dominant controlling factors, among which DG exerts substantial influences on slump and pH of concrete. Based on surrogate models and Latin hypercube sampling, an inverse design framework is proposed to obtain optimal mix proportions from preset performance targets. All validation cases show relative deviations within 5%, where the maximum 4.02% error specifically denotes the relative deviation for chloride migration coefficient prediction. Compared with performance‑matched OPC reference mixtures, some optimized mixtures achieve cement‑dosage reductions that can exceed 50%, and achieve 41.60%–51.00% embodied‑carbon reduction. Meanwhile, the material‑cost increment narrows from 13.45% to 7.00% at higher strength grades. The whole optimization process completes within 0.5 s. This framework offers an efficient and reliable pathway for low‑carbon and intelligent design of low‑alkalinity ecological concrete.