An explainable multi-algorithm optimization framework with fuzzy duration preprocessing for sustainable infrastructure project scheduling

Multi-objective construction scheduling studies often rely on one metaheuristic and do not explain why a schedule is selected. This study develops an auditable decision-support pipeline for multi-mode infrastructure scheduling in which trapezoidal fuzzy durations are defuzzified before a deterministic search. Five evolutionary solvers (NSGA-II, NSGA-III, MOEA/D, RVEA and UNSGA-III) run under identical settings, and their outputs are pooled into one de-duplicated non-dominated front that attributes every schedule to its solver. A CRITIC-MEREC weighted consensus of TOPSIS, EDAS and WASPAS selects a schedule, and a Shapley-value surrogate attributes the selection to activity-mode commitments. On a 20-activity network, duration fell by 26.12%, cost by 12.04% and resource fluctuation by 35.86%, but emissions by only 1.96%, because faster modes raised working emissions almost as much as they reduced idle emissions. MOEA/D collapsed to one distinct schedule, and three activities carried most of the attribution. Across 30 replications the pooled front exceeded every solver in hypervolume, although five NSGA-II restarts at equal budget scored 0.86% higher. On PSPLIB networks of 30 to 120 activities, emissions fell by 11.87% to 54.78%. The pipeline shows which solver suits a problem class, which commitments need managerial attention and whether scheduling alone can reduce carbon meaningfully.