Completion of Global Navigation Satellite System Missing Data Based on Bidirectional Recurrent Neural Network

In fight test, Global Navigation Satellite System (GNSS) data missing inevitably in GNSS data recording attribute to such factors as satellite anomalies, follow-up reject of gross errors and so on. It has serious effects on the data correlation, principle component analysis and spectral analysis. Thus it is significant to fill missing values by interpolation method in the GNSS time series of aircraft. The research result of time series interpolation problem has already rich but time serious methods frequently-used such as traditional interpolation method, empirical Orthogonal Function and Singular Spectrum Analysis still have some disadvantages: For example, local feature fitting of the time series is not very good or the conditions of appliance is so harsh that it is hard to application and extension and It is easy to cause artificial distortion of the reconstructed time series. The article has presented a way of filling missing GNSS values of aircraft based on bidirectional recurrent neural network. Experiments in the article are carried out with samples from the complete time series of a plane throughout the year. Then we trained the bidirectional recurrent neural network and used the interpolation method to complete some groups missing GNSS values that missing time is three minutes, six minutes, nine minutes, twelve minutes, fifteen minutes respectively. Finally, the accuracy and validity of the experimental model used for completing the time seriously by interpolation method are verified.