GNSS location error reduction method for microtremor survey system based on EMD-CNN-LSTM

Abstract

Abstract Large errors exist when the microtremor survey system uses the global navigation satellite system (GNSS) for static localization. Aiming at the problem that the existing methods cannot effectively weaken the random error and multipath error, an error weakening method based on Empirical Mode Decomposition (EMD), Convolutional Neural Networks (CNN), and Long Short-term Memory Networks (LSTM) is proposed. The model first uses EMD to decompose the high-frequency random error, then reconstructs the low-frequency component and extracts the local features using CNN, and finally learns the change rule of multipath error using LSTM and weakens it. The model can remove random errors in the early stage while reducing the interference of noise on the neural network in the later stage and then improve the accuracy of localization. The experimental results show that the model can effectively improve the localization accuracy in the case of short-time measurements so that the localization accuracy in the E, N, and U directions can be improved by 74.57%, 74.76%, and 71.86%, respectively, which is more than 10% higher than the localization accuracy improvement rate of the existing CNN-LSTM model.