Enhancing Landslide Risk Prediction with Dynamic Fusion Neural Networks for Economic Impact Assessment

Because of their stepped deformation properties, rockslides that occur across stepped failure surfaces are notoriously difficult to predict in terms of failure period. The most prevalent type of rock slope failure in areas with complicated geological conditions is step-path rockslides, which occur as a result of excavation. Both steep and moderate fractures contribute to these failures, which in turn affect deformation, the progression of failures, and the stability of the slope. Data preprocessing is essential for accurate landslip risk prediction since it predicts deformation and failure characteristics. This work utilises dynamic fusion to improve model performance by merging numerous characteristics and presents four methods for data pretreatment that adhere to data normalisation standards. In order to make static deep learning models more efficient, it also investigates GRA feature selection methods. To anticipate mechanical situations, a new prediction system called ErrorDFHNN is suggested, which combines CNNs with error fusion from several SAEs. The results demonstrate that GRA-based feature selection improves predictive abilities; the suggested model outperforms current models with an accuracy of 99.74%. A strong early warning system for slope stability assessments in geologically complex locations can be achieved by data preprocessing and dynamic fusion, which the study highlights as crucial for enhancing landslip displacement forecasts.