| Landslides and other geological hazards have been seriously affecting people’s personal safety and property security.The development of science so far has made it very important to monitor landslide slopes accurately and effectively by using existing technologies,and it is of paramount importance to realize effective prediction of landslides.In this paper,an automatic GNSS data solving system is built,and a Hyperopt-LSTM model based on LSTM neural network model is realized for the fusion of multi-source heterogeneous landslide monitoring data:(1)A slope intelligent monitoring network is established in the swelling soil slope of Ningming County,Guangxi,containing Beidou/GNSS monitoring sub-network,rainfall monitoring sub-network,soil pressure monitoring sub-network and soil moisture monitoring sub-network.(2)The long-term stability of the datum was analyzed by using GAMIT/GLOBK software,and the coordinate time series information of the datum and the movement speed of the datum in the framework of ITRF were obtained.(3)The GNSS data automatic solution system was built,and the accuracy of GAMIT baseline solution for landslide monitoring was analyzed in detail,and it was concluded that the accuracy of GAMIT baseline solution can be fully used for landslide monitoring.In addition,the TRACK module was used to process the monitoring data,and the single ephemeris monitoring results of high frequency were obtained.(4)The Hyperopt-LSTM model was established to realize the fusion of multisource heterogeneous landslide monitoring data.The GNSS monitoring data,rainfall data,earth pressure data and soil moisture data were pre-processed to explore the effects of the latter three on landslide displacements,and the model input factors were screened and input to the Hyperopt-LSTM model to predict landslides with abrupt displacements in two consecutive time periods,and the final predicted RMSE value was 1.47 cm and the MAE value was 1.0 cm,and the goodness-of-fit reached 0.95. |
