Sustainable dynamic multi-objective optimization framework of fiber-reinforced geopolymer for soils stabilization based on machine learning and NSGA-Ⅱ algorithm

Fiber-reinforced geopolymer-stabilized soils (FRGSS) are promising sustainable materials with enhanced mechanical performance and reduced environmental impact. Although the mix proportion optimization design for fiber geopolymer reinforcement systems has been extensively studied, its current methods are limited to single-objective optimization of strength or multi-objective optimization targeting a specific type of material. This research aims at developing a sustainable multi-objective optimization framework with dynamic input capability based on machine learning and Bayesian optimization algorithms. A unique dataset compiling 23 feature variables was established from experiments and existing literature. Using this dataset, an integrated model that combines multi-source parameters and machine learning was developed to predict the unconfined compressive strength (UCS) of FRGSS. To enable the comprehensive evaluation of sustainability performance, objective functions were formulated to describe the interrelationships among UCS, carbon emissions, and economic cost. Based on these functions and the developed machine learning model, a sustainable dynamic multi-objective optimization framework was proposed using the non-dominated sorting genetic algorithm II (NSGA-II), and finally, a graphical user interface (GUI) system was developed. The results indicated that among the four evaluated models, the Bayesian-optimized artificial neural network (ANN) model achieved the highest prediction accuracy for UCS in both the training and testing datasets. Moreover, the developed framework demonstrated excellent dynamic adaptability under laboratory data with fixed strength requirements and accurately produced optimal material mix proportions that satisfied multi-objective constraints of strength, environmental impact, and cost-effectiveness, confirming the reliability and practical applicability of the proposed framework. This study established a versatile intelligent framework, rather than a specific model, capable of dynamically adapting to material parameters to achieve multi-objective collaborative design of fiber-reinforced geopolymer-stabilized soils. This advancement contributes provide effective and environmentally-friendly, cost-effective solutions for the soil stabilization project.