An adaptive Kalman filtering algorithm based on maximum likelihood estimation
Zili Wang,Jianhua Cheng,Bing Qi,Sixiang Cheng,Sicheng Chen
DOI: https://doi.org/10.1088/1361-6501/ace9ef
IF: 2.398
2023-01-01
Measurement Science and Technology
Abstract:Traditional adaptive Kalman filtering algorithms based on innovation are often used to solve the problem of reduced or even divergent filtering estimation accuracy under abnormal measurement noise. However, these algorithms are usually characterized by difficulties in selecting window width and window weight, which cannot simultaneously take into account the filtering tracking sensitivity and filtering accuracy. In this paper, an adaptive Kalman filtering algorithm based on maximum likelihood estimation is proposed, which determines the window size and window weight size under the kth moment by designing a window adaptive selection function and a weight function to change the innovation covariance at the kth moment, which in turn changes the measurement noise covariance at the kth moment, so that the measurement noise covariance is no longer a fixed single value, but can better adapt to the changes in the environment, reflecting good adaptive characteristics. The simulation results based on GPS/SINS integrated navigation system demonstrate that the new filtering algorithm of this paper reflects higher filtering accuracy and stronger robustness under the carrier in multiple motion states and accompanied by time-varying measurement noise interference. Compared with the traditional adaptive Kalman filtering algorithm based on innovation, the accuracy of attitude angle estimation error under this method is improved by 119.97%; the accuracy of velocity estimation error is improved by 264.42%; the accuracy of position estimation error is improved by 156.69%.