Laser Gyro Temperature Error Compensation Method Based on NARX Neural Network Embedded into Extended Kalman Filter

Laser gyro has been widely applied in strapdown inertial navigation system (SINS) with great advantages of shock resistance and high sensitivity. However, because of the temperature sensitivity of the medium, optical devices and materials inside the gyroscope, when the environment temperature changes, the bias of the laser gyro is exacerbated. It makes the precision advantages of the laser gyro cannot be fully exerted. In order to meet the requirements of high-precision and strong stability in laser gyro SINS, this paper proposes a laser gyro bias compensation method based on NARX neural network embedded into EKF (NARX-EKF). Considering the dynamic time-varying characteristics of laser gyro bias caused by external temperature variations, a non-linear dynamic model of laser gyro bias partial derivative with respect to temperature can be established by nonlinear autoregressive with external input (NARX) neural network. Then, the non-linear dynamic model is embedded into an Extended Kalman filter (EKF), thereby the real-time dynamic compensation for the laser gyro temperature error has been achieved. In order to verify the effectiveness of the proposed method, related dynamic temperature experiments are designed. The results of temperature experiments show that the compensation method can accurately predict and compensate the bias drift of laser gyro. Compared with the SINS with uncompensated temperature error under the static base condition, the compensation method proposed in this paper is more effective to reduce the attitude error and meet the high accuracy requirements.