Aiming at the problem that the effect of variational Bayesian adaptive Kalman filtering will be affected when the noise statistical characteristics in the linear Gaussian state space model are time-varying, an improved algorithm based on the principle of strong tracking is proposed.The measurement noise model is selected as the inverse Wishart distribution, the system state and time-varying measurement noise covariance are taken as the parameters to be estimated, and variational Bayesian approach is used for their recursive estimation.The optimum estimation result of the measurement noise covariance is then taken as a time-varying parameter and introduced into the sub-optimal fading factor based on the strong tracking principle, to improve the correction accuracy of the covariance of state prediction.Simulation results show that:1) The improved algorithm can track the measurement noise covariance adaptively in the linear Gaussian system with time-varying noise, and effectively overcome the influence of time-varying noise covariance;and 2) The convergence speed and accuracy of the estimated results are greatly improved.