A porgram of active vibration control with unknown excitations was studied here.According to the knowledge of Kalman filtering kand linear quadratic Gaussian(LQG) methd,the control algorithm based on random walk and input estimation was proposed.First,the excitations were described as discrete recursive forms by using random walk model and they were introduced into state functions of the control system as auxiliary state variables.And then,the Kalman filtering principle was used to estimate the unknown excitation combining the auxiliary the auxiliary state functions.So,the system tarber functions could be created according to the estimation values of the estimation values of the unknown excitations,state variables and the known measured outputs.Finally,using the LQG algorithm to get the optimal values of current control inputs.Good control effects were proved by numeric simulation with sdof and 2 dof models.The results were better than those based on the conventional LQG method .