Extended LSTM Model Based Satellite Positioning Error Estimation Method

With the development and application of the global satellite navigation system (GNSS), satellite-based positioning technology has become an important source of positioning for fundamental applications. The problem of deterioration of GNSS positioning accuracy performance due to multipath and non-line-of-sight propagation still remains during the process of urban aviation traffic. In order to solve this problem, the method based on extended Long Short-Term Memory (LSTM) Neural Network model to correct initial observation position error is proposed. The method first studies the relationship between satellite observations and positioning errors in different environments. Secondly, the dropout layer, ReLU layer, fullconnect layer and regression layer are added to the LSTM network to extend the model and establish a positioning error estimate model, in order to optimize the regression performance of the model. Finally, the extended LSTM model is applied to the initial observation position correction. The experimental results show that in the static urban canyon environment, the errors in the east, north and up directions corrected by the extended LSTM model are all about 1.0 meter. In the dynamic ground reflection environment, the errors in the east, north and up directions corrected by the extended LSTM model are all within 1.5 meters, which is a significant improvement of the accuracy compared to before correction.