A compensation method for gyroscope random drift based on unscented Kalman filter and support vector regression optimized by adaptive beetle antennae search algorithm

The random error of fiber optic gyroscope (FOG) is an important factor affecting its performance. In this paper, a novel and efficient compensation scheme for random drift is presented. It is a new model based on fusing unscented Kalman filter (UKF) with support vector regression (SVR) optimized by the adaptive beetle antennae search (ABAS) algorithm. At first, to make up for the shortcomings of the basic beetle antennae search algorithm, an adaptive decay factor is proposed to dynamically adjust the update of the search step size. The proposed ABAS algorithm exhibits more superior global optimization capability in hyperparameter optimization of SVR. And then to better characterize the nonlinearity and randomness of random drift, the optimized SVR is presented for the modeling of random drift data. Considering the improvement of modeling accuracy, this study also presents to preprocess the raw data by using the variational mode decomposition (VMD) algorithm and sliding window method. Furthermore, as an online processing method, UKF is introduced and fused with optimized SVR modeling, and a hybrid model is constructed by designing state space equations. Finally, experiments are conducted on the measured data of FOG to verify the superiority of the proposed model. The experimental results show that compared with the conventional method, in terms of the compensation accuracy for random drift data, noise intensity (NI) and Durbin-Watson (DW) value of the proposed scheme are reduced and improved by 28.57% and 9.06%, respectively.