Online Estimation Method of Gyro Random Drift Based on LSTM Network

In order to reduce the adverse effect of random drifts on Micro Electro Mechanical Systems (MEMS) gyroscope's measurement accuracy, a real-time Long Short-Term Memory (LSTM) network-based estimation method for random drift is proposed. The traditional modeling method Autoregressive Moving Average (ARMA) is limited by linear assumption. In this paper, the LSTM network model is trained for MEMS gyro random drift sequence to fully learn the time series information in the sequence. At the same time, the network model is incorporated into the framework of Cubature Kalman Filter (CKF) algorithm through certain strategies to realize real-time estimation and compensation of gyro random drift. The experimental results show that the prediction accuracy of the network model is higher than that of the ARMA model, which verifies the effectiveness and superiority of the CKF method based on LSTM (LSTM-CKF) in random drift estimation and compensation.