This paper addresses the bottleneck of transferring 100 million tons of materials for lunar base construction. We establish a closed-loop intelligent decision-making framework that integrates discrete optimization, robustness analysis, and machine learning forecasting. To tackle the coordination challenges of heterogeneous transport modes, a bi-objective mixed-integer programming model combined with a genetic algorithm is proposed, identifying a Pareto-optimal oper - ating point that achieves comprehensive superiority over pure strategies with a completion time of 156.99 years and a cost of $23.62 trillion. Distinct from traditional deterministic studies, this research incorporates robust optimization to reveal structural risk asymmetries, demonstrating that elevator solutions are critically sensitive to schedules while rocket solutions are cost-sensitive, thereby proving the hybrid strategy's superior load-distribution efficacy under extreme sce - narios. Furthermore, a data-driven supply scheduling method is developed by coupling ridge regression with high-quantile reliability analysis, which transforms stochastic human water consumption behaviors into reliable, engineering-grade replenishment timelines. Finally, by synthesizing Life Cycle Assessment with TOPSIS, the study quantifies environmental externalities and establishes dynamic decision thresholds, providing rigorous quantitative foundations for the strategic planning of sustainable large-scale deep-space logistics systems.

mixed-integer programming, robust decision-making, life cycle assessment