Two improved Gaussian particle filter algorithm

Two improved algorithms are proposed in this paper, aiming at overcoming the disadvantages of Gaussian particle filter (GPF) for high precision state estimation. One is that the measurement innovation is identified by using Grubbs criteria in the update phase of standard GPF. Thus the effect of bad samples, which are produced in the random sample process, is reduced for state estimation. Another is that Gauss-Hermite filter (GHF) is used to optimize the importance density function of GPF. So the current measures are integrated into process of the system state transition, which makes the prediction samples be closer to the samples of the real posterior probability. The results of the simulation show that filtering accuracy of both algorithms has been improved with standard GPF.