Build Up a Real-Time LSTM Positioning Error Prediction Model for GPS Sensors

Abstract

This paper presents a real-time long short-term memory (LSTM) recurrent neural network (RNN) to trace and predict the GPS positioning errors within next one to several seconds, offering an enhance GPS positioning. The proposed LSTM prediction model was further verified over extensive experimental data captured in cities and metropolitans, urbans and highways across several middle and eastern States of the United States. The prediction accuracy of the proposed real-time LSTM can be within less than 1-3% of its ground true values outperforms those results gained by conventional statistics and linear prediction models.