CEEMDAN-iLSTM-Attention based Method for Attitude Data Prediction and Correction
To improve the prediction accuracy of attitude sensor data,the paper presents a CEEMDAN-iLSTM-Attention based method for prediction and correction of attitude data.Firstly,the CEEMDAN method is used to obtain different mode angular sequence information.Secondly,an improved whale optimization algorithm is used to find the optimal network parameters of LSTM-Attention.Then,multiple iLSTM-Attention prediction models corresponding to each mode are established,and the prediction data is obtained by superimposing the prediction results of each mode component.Subsequently,the errors between the predicted values and the real values are calculated,and the error prediction model is established by using the iLSTM-Attention network.Finally,the corrected prediction data is obtained by subtracting the error prediction model output from the data prediction model output.The experiments show that the corrected prediction errors of angle,acceleration,and angular velocity drop by 57.2％,39.2％,and 76.2％,respectively,compared to the errors of the predicted data without correction.
attitude data predictionlong and short-term memory networksattention mechanismserror correction
NETL
NSTL
万方数据
