This study investigates the vibration characteristics of 10 MW, 8 MW, and 6.7 MW large-capacity jacket-supported offshore wind turbines (OWTs) with varying conditions and foundation types in a deep-sea wind farm. By using an improved Natural Excitation Technique combined with Eigensystem Realization Algorithm (NExT-ERA) to accurately identify modal parameters of OWTs under operational harmonic interference across six operational phases (shutdown, startup, grid-connection, transition, rated speed, cut-out) using high-precision accelerometers (50 Hz) and SCADA data. Research on vibration characteristics revealed that the improved NExT-ERA algorithm, reduced harmonic-induced errors in fundamental frequencies by 0.2-0.5 Hz during operational states. Still, due to the complex operation of the grid-connection phase and the transition phase, the modal recognition showed a positive deviation and has a large degree of discreteness. Besides, the influence of foundation type on vibration amplitude is more significant near the foundation, while the effects of installed capacity and turbine manufacturer dominate closer to the tower. Advanced aerodynamic designs and modern control optimizations significantly reduce vibration amplitudes, thereby enhancing structural durability, as demonstrated by the greater vibration amplitudes of the 6.7 MW turbine compared to larger, more optimized turbines. While a single sensor can identify overall modes, the tower top's sensitivity to complex loading necessitates multi-level monitoring to comprehensively capture vibration modes, ensuring robust structural health assessment.