For large cross-section tunnels constructed with roadheaders, the coordination of multiple construction procedures often relies heavily on operational experience, leading to low efficiency and difficulties in settlement control. To address these challenges, and based on the Changshui Airport Tunnel project along the Chongqing-Kunming High-Speed Railway, this study proposes an intelligent optimization framework that integrates excavation efficiency prediction, settlement early warning, and multi-procedure parameter optimization. First, multi-source monitoring data were temporally and spatially aligned to develop a LightGBM-based prediction model enhanced with Bayesian optimization for excavation efficiency, and a Multi-LSTM vault settlement prediction model incorporating geological conditions, construction procedure features, and dynamic face advancement. The results show that the two models achieve coefficients of determination of 0.77 and 0.94, respectively, accurately mapping relationships among construction parameters and geotechnical properties. Furthermore, the NSGA-Ⅱ multi-objective optimization algorithm was adopted to conduct coordinated optimization of key construction parameters, including cyclic advance length, support duration, and support strength. The results indicate that under typical karst geological conditions, the optimized strategies improve tunneling efficiency by an average of 34.6% while reducing settlement by an average of 23.1%.