Multi-UAV Construction Delivery Under Energy Uncertainty: Rank Sensitivity and Knowledge-Driven Scheduling

Urban construction sites require timely delivery of lightweight tools, sensors, components, and maintenance parts. We study a three-objective scheduling model that balances fleet mission duration, interval energy consumption, and soft receiving-window deviation. Nominal carried load determines flight time, while uncertain load and wind affect energy and upper-bound battery feasibility. A candidate-pool diagnostic measures changes in scalarized winners, replacement decisions, and an endpoint-separation relation. WV-KD-MOEA/D combines a midpoint decomposition backbone with knowledge-driven operators and weight-vector-aware allocation. Six methods are compared on 27 synthetic test instances with ten repetitions per instance. MOEA/D-DYTS shares the backbone and operator pool in its specification; compared with its outputs, WV-KD-MOEA/D has lower empirical-reference IGD but no significant hypervolume advantage. The midpoint–interval volume–coverage trade-off remains unresolved and depends on the metric and condition. The main comparison provides no evidence of an advantage over fixed-global allocation. A seven-value threshold sweep examines fixed evaluated records from α = 0.50 to 0.90. Mean set size increases from 74.5 to 85.0 under energy-only uncertainty and from 68.1 to 108.5 under coupled-time uncertainty. Energy-only hypervolume varies by less than 0.00038. Coupled-time rates cross the frozen 5% cutoff, and rule agreement ranges from 23 to 29 of 81 blocks. Under the shared R1 endpoint loss, the development-fitted rule improves on the original constant-hybrid default in agreement (27 of 81 versus 6 of 81 held-out condition blocks). Its mean decision regret is also lower (0.000106 versus 0.000156). Constant midpoint performs better throughout the tested α range (42 of 81; 0.000069). The cross-condition results are exploratory and do not establish the best backbone for R2 or R3. The rule describes measured disagreement, and the fixed midpoint design remains the examined default. A companion analysis reruns matched methods on the same geometries under heterogeneous load–wind, coupled travel-time uncertainty, and landed-handover conditions. These results complement the main benchmark without establishing construction-project or field-operational benefits.