Landslide evolution state prediction and down-level control based on multi-task learning

In this paper, multi-task learning is introduced into the study of landslide evolution state prediction and control. Firstly, we define two landslide evolution states and propose a method of landslide evolution state level classification prediction. Specifically, we use Gaussian mixture model (GMM) to reconstruct labeled data sets and establish a landslide evolution state level prediction model based on Multi-Task Learning-Stacked Long-Short Term Memory (MTL-SLSTM), and then use task weight rules to design multi-task losses for network training. The parallel multi-step prediction of the evolution state is achieved by the above works. Secondly, considering the spatial–temporal correlation of different monitoring points on the same landslide, we analyze spatial–temporal data to construct spatial–temporal features, and design a multi-task correlation learning mechanism combined multi-task weight learning and multi-task relationship learning methods to construct spatial–temporal relation. The landslide multi-point prediction model based on Multi-Task Correlation Learning-Stacked Long-Short Term Memory (MTCL-SLSTM) achieves single-step prediction of the evolution state of multiple monitoring points with high accuracy. Finally, according to the idea of neural direct inverse model control, we propose down-level control method based on the prediction of the landslide evolution state. We build an interval prediction network based on bootstrap method and model selection strategies, and then a safe rainfall interval predictor is trained offline. Moreover, we design an online landslide down-level control process combined the landslide evolution state level predictor, which realizes single-step control of single-point of the dangerous landslide. Furthermore, the effectiveness of the proposed method is verified on Baishuihe landslide.