Abstract: With the use of large eddy simulation (LES) model and overset mesh, the vortex-induced vibration (VIV) mechanism of wavy cylinder, with suction control at azimuthal angles of 90°/270° and 120°/240°, has been numerically investigated for reduced velocities U_r from 2.4 to 13.2. In addition, to suppress the vibration by an adaptive suction control, velocity information is monitored in real-time and serves as feedback in reinforcement learning framework. With the optimization goals of reducing the amplitude of VIV and minimizing energy input, the proximal policy optimization (PPO) algorithm is employed to adjust the momentum coefficient of suction. The result indicates that lock-in range narrows significantly with suction control, but threshold wind speed at which cylinder begins to vibrate advances. Under suction control, the shear layer is extended in the streamwise direction, and the vortex shedding mode is transformed into a parallel shedding mode characterized by elongated tail vortices. These changes reduce the negative pressure on the leeward side of the cylinder and suppress the pulsating excitation of the Vortex-Induced Vibration. The vibration suppression effect is better when suction control is applied at 120° and 240°. The amplitude ratio at the peak is reduced by about 73.17% compared to the uncontrolled condition. The adaptive control process shows that higher momentum jet than open-loop control is taken firstly to quickly decrease the vibration amplitude, and then a lower momentum jet is utilized to maintain the low amplitude state. Compared to uncontrolled condition, the amplitude ratio is reduced by approximately 96.9% with a lower momentum jet than open-loop control. The research result has achieved adaptive closed-loop vibration suppression control of the wavy cylinder, providing new insights and approaches for the adaptive active control of flow around bluff bodies.