Robust design of self-starting drains using Random Forest

Abstract

Groundwater lowering is one of the most important countermeasures to avoid the risk of rainfall-triggered landslides. However, the long-term reliability of many drainage methods is often a matter of concern since the drains may easily get clogged. A new hydraulic-driven self-starting drainage method is presented in this paper. In the proposed Random Forest (RF) based robust design approach for the self-starting drains, the datasets are generated using an automatically controlled numerical modeling technology. The deterministic analysis is carried out based on uncertain soil parameters and the specific designs selected using Uniform Design (UD). The ensemble of RF models is applied in the design process to improve computing efficiency. Safety requirements, design robustness, and cost efficiency are simultaneously considered utilizing multi-objective optimization. A straightforward and efficient framework that focuses on difficulties caused by an enormous design space is established for the robust design of the self-starting drains, and improved computation efficiency is achieved. The effectiveness of the proposed approach is illustrated with a case study, the Qili landslide in Zhejiang Province, China.