APEX-CPM: Beidou Slope Deformation Prediction Based on Adaptive Path Exploration and Composite Feature Modeling

In large-scale water conservancy construction projects, the deformation of high and steep slopes not only threatens the safety of the project and the surrounding ecological environment, but may also cause catastrophic natural disasters. Although the BeiDou satellite monitoring system has been widely used, monitoring data alone cannot ensure safety. To address this challenge, we propose a high and steep slope deformation prediction method (APEX-CPM) based on BeiDou high-precision monitoring data. Based on the partitioning of associated patches, this method integrates time resolution, time distance and channel dependency to perform composite feature modeling [1]. By partitioning associated patches of different sizes, we decompose the time series into associated patches of different time resolutions, and use multiple attention mechanisms to capture the local, global and channel correlations of the time series. In addition, inspired by the Mixture of Experts (MOE) [2] model and Pathways [3], we use adaptive path exploration to further improve the generalization ability of the model under BeiDou high and steep slope monitoring data at different locations, so that it can dynamically adjust the composite feature modeling process according to the changes in the input time series, thereby ensuring the accuracy and generalization of APEX-CPM. Experiments conducted on a real dataset of high and steep slope deformation monitored by the Beidou satellite navigation system show that APEX-CPM is significantly better than the current advanced models in the field of long-term time series prediction in terms of prediction accuracy. Taking the BDS001 measurement point data as an example, the MSE of the model reached 0.82084 and the MAE was 0.70027. Compared with the best performing PatchTST [4], the MSE of APEX-CPM decreased by 0.03117, an increase of 3.66%; the MAE decreased by 0.01793, an increase of 2.49%.