Improved genetic programming modeling of slope stability and landslide susceptibility

Abstract

The prediction of slope stability and landslide susceptibility is crucial for ensuring the safety and reliability of high slopes and disasters prevention. This study used genetic programming (GP) to predict slope stability and landslide risks. To address the limitations of GP such as local convergence and code redundancy growth and enhance prediction accuracy, hierarchical fair competition model based on K-means clustering algorithm (K-means-HFC), niche technique of similarity based on crowding (NTSC), and self-adaptive change in probability were proposed to improve the traditional GP. Then, the improved GP was used to conduct modeling research for prediction, including slope stability, land-slide dynamic characterization, probabilistic hazard of seismic landslide, and blasting vibration parameters and hazard. The results showed that K-mean-HFC and NTSC separately increased inter- and intra-cluster population diversity and promoted the fitness, further enhancing the model prediction accuracy. In the case of multi-parameter prediction, the improved GP could realize attribute reduction on the prediction parameters, eliminate the attributes unrelated to the prediction parameters, and clearly obtain the prediction formulas. By utilizing the improved GP, the prediction model of slope stability was acquired, the mutual prediction of surface displacement rate and subsurface volumetric was established, the probabilistic prediction diagram of seismic landslide in Sichuan Province was generated, the influence of prediction parameters was analyzed, and the prediction of blasting vibration parameters and hazard of slope blasting under the influence of multiple parameters was realized. The derived prediction formulas possessed a significant reference for solving the same type of slope reliability and landslide prevention problems.