基于神经网络和 Kalman 滤波的陀螺阵列融合算法

Aiming at the low precision and reliability of micromechanical gyroscopes, a fusion algorithm of gyroscope array based on neural network and Kalman filter is proposed. By combining the neural network with Kalman filter, LSTM-RNN is used to calculate the confidence degree of each gyroscope. The confidence degree, measured value and angular velocity estimated by Kalman filter of each gyroscope are input to BP neural network for data fusion, so that BP network has more characteristic information about gyroscopes during training, so as to improve the angular velocity fusion accuracy. Since the confidence degree of each gyroscope is obtained first, BP network can identify the fault gyroscope more easily, thus reducing the utilization rate of the fault gyroscope measurement data. The actual system verification shows that in the case of gyroscope fault, the MAE and RMSE of gyroscope array of the proposed algorithm are reduced by 80.25%and 81.39% respectively compared with Kalman filter, and reduced by 60.33%and 63.41% respectively compared with LSTM-RNN fusion algorithm with only measurement input, which has strong fault tolerance and robustness.