Abstract
To address the significant random errors in gyroscopes within Inertial Measurement Units (IMU), this paper presents an error calibration and compensation method for IMU gyroscope arrays based on the Long Short-Term Memory (LSTM) neural network algorithm. Experiments were conducted using 16 Inertial Measurement Units to form a MEMS tri-axial gyroscope array, with real-time data collection achieved through embedded low-level development, and static and dynamic tests were performed on a high-precision dual-axis turntable. Experimental results indicate that, compared to averaging data from multiple IMU gyroscopes, employing the Long Short-Term Memory Neural Network algorithm for error compensation in the IMU gyroscope array can reduce bias instability and angle random walk by over 50%, and angular velocity random walk by more than 35%.