Objectives Based on the interaction characteristics of various environmental factors affecting landslide deformation, a new method for multi-source heterogeneous monitoring data fusion is proposed to improve the accuracy of landslide deformation prediction.
Methods First, environmental factors are selected based on mutual information method.Then, the selected environmental factors are taken as the input varia-bles of long short-term memory(LSTM) model, and the accumulated displacement data of landslide are taken as the expected output data, and the parameters of the model are optimized through improved particle swarm optimization method, so as to further improve the prediction accuracy of the fusion model.The global navigation satellite system(GNSS) data of Fa'er landslide in Shuicheng County, Liupanshui City, Guizhou Province are analyzed.
Results Experimental results show that the improved particle swarm optimization(IPSO)-LSTM neural network data fusion algorithm, based on mutual information is suitable for landslide deformation prediction with multi-source heterogeneous monitoring data.The environmental factor variable selection method based on mutual information is better than Pearson correlation coefficient selection method. After optimizing the parameters of the improved particle swarm optimization algorithm, the prediction accuracy of the fusion model is higher.
Conclusions The proposed fusion prediction model has high prediction accuracy in landslide cumulative displacement prediction, which has important reference value for improving the reliability of landslide monitoring and early warning. It should be noted that only a few typical environmental factors are collected. In practical application, other factors such as groundwater level, soil moisture and human activities can be considered to further improve the prediction accuracy and reliability of the fusion model.