Gaussian Mixture Filter Based on Variational Bayesian Learning in PPP/SINS

Abstract

Aiming at the problem of inaccurate estimation of parameters and noise covariance for the noise mixture model encountered when using Gaussian filter during operation of PPP/SINS integrated navigation system in the non-Gaussian noise environment, this paper proposes a Gaussian mixture adaptive filtering algorithm that is based on variational Bayesian learning. The algorithm allows accurate and efficient adaptive estimation of the parameters of Gaussian mixture filtering model based on the variational learning theory. It also further refines the Gaussian mixture filtering stochastic model, which greatly improves the filtering estimation precision; reduces the computational complexity; and effectively improves the solving efficiency. Data simulation was carried out on the PPP/SINS tightly coupled integrated navigation system. The results showed that compared to the conventional Gaussian mixture filtering algorithms, the new algorithm further improved the estimation accuracy, which was also computationally fast and less burdensome. Our results provided some theoretical support for the future application and extension of the PPP/SINS integrated filtering algorithm in the non-Gaussian noise environment.