An inertia effect is proportional to acceleration and mass as a negative force, keeps own situation of a body during movement under inertia law. It is obvious that the inertia mass effect is available to decrease own natural frequency and arise anti-resonance which can be shut down vibration in a special case. In previous paper the authors updated a vibration control device which was having variable moment of inertia by Magneto-Rheological fluid inside a flywheel. Ferrite particles of the MR fluid are clustered when magnetic field is applied by 8 electromagnets to the flywheel. It was clear that the inertia mass effect was varied as higher current, and vibration control effect under the earthquake was confirmed numerically and experimentally. If inertia mass effect can be switched depending on anti-resonance, vibration may be able to be suppressed under the earthquake. Recently some researchers focus reinforcement learning to decrease vibration. Therefore, to switch mass inertia effect, reinforcement learning is adopted. In this paper firstly the vibration control device with thin iron plate attached inside the flywheel is updated from prototype in order to get more inertia effect. Secondly, Deep Deterministic Policy Gradient (DDPG) which is a part of reinforcement algorithm is adopted in order to decrease vibration by inertia mass effect, and vibration tests of one degree-of-freedom were carried out.