A federated learning based approach for predicting landslide displacement considering data security

Homeland security is an important concern in contemporary society. National mega strategic engineering areas and other key regions, characterized by the presence of high mountains and valleys, are prone to various geological hazards, including landslides. Therefore, the timely geological hazard prediction and forecasting are required, and local data security protection is also crucial. To address the aforementioned problems, a landslide displacement prediction method based on federated learning which can protect data security is proposed in this paper and validated in the Three Gorges Project area. The essential idea is to employ the federated learning approach to enable the joint training of deep learning models for landslide displacement prediction without exchanging data. The proposed method (1) trains each landslide displacement prediction model locally without data exchange, ensuring geospatial data security, and (2) improves the accuracy of landslide displacement prediction in most cases, protecting people's lives and properties. The proposed method has the potential to improve the prediction and forecasting of geological hazards in other key areas, thereby protecting people's lives and properties while ensuring national homeland security.