Abstract: A new adaptive Kalman filtering algorithm is presented. The new algorithm assumes that the variance relationship between process noise and measurement noise is known, but both kinds of variance are unknown and varying with time. At first, let the process noise variance at the current time point be equal to that at the prior time point. Applying the method of variational Bayesian approximation, the measurement noise variance and state estimation are solved under the framework of Kalman filter, and then a new process noise variance is obtained via the function relationship. After the process above is implemented for some runs, the final state estimation and covariance are obtained. Experimental results show that the new algorithm has higher accuracy; Furthermore, the new algorithm still has strong robustness when the assumption is uncertain.