A hybrid machine learning model for landslide-oriented risk assessment of long-distance pipelines

Haijia Wen; Lei Liu; Jialan Zhang; Jiwei Hu; Xiaomei Huang

doi:10.1016/j.jenvman.2023.118177

A hybrid machine learning model for landslide-oriented risk assessment of long-distance pipelines

J Environ Manage. 2023 Sep 15:342:118177. doi: 10.1016/j.jenvman.2023.118177. Epub 2023 May 20.

Authors

Haijia Wen¹, Lei Liu¹, Jialan Zhang², Jiwei Hu¹, Xiaomei Huang¹

Affiliations

¹ Key Laboratory of New Technology for Construction of Cities in Mountain Area, Ministry of Education; National Joint Engineering Research Center of Geohazards Prevention in the Reservoir Areas; School of Civil Engineering, Chongqing University, Chongqing 400045, China.
² Key Laboratory of New Technology for Construction of Cities in Mountain Area, Ministry of Education; National Joint Engineering Research Center of Geohazards Prevention in the Reservoir Areas; School of Civil Engineering, Chongqing University, Chongqing 400045, China. Electronic address: lanaij@cqu.edu.cn.

PMID: 37210819
DOI: 10.1016/j.jenvman.2023.118177

Abstract

Preparation of pipeline risk zoning is essential for pipeline construction and safe operation. Landslides are one of the main sources of risk to the safe operations of oil and gas pipelines in mountainous areas. This work aims to propose a quantitative assessment model of landslide-induced long-distance pipeline risk by analyzing historical landslide hazard data along oil and gas pipelines. Using the Changshou-Fuling-Wulong-Nanchuan (CN) gas pipeline dataset, two independent assessments were carried out: landslide susceptibility assessment and pipeline vulnerability assessment. Firstly, the study combined the recursive feature elimination and particle swarm optimization-AdaBoost method (RFE-PSO-AdaBoost) to develop a landslide susceptibility mapping model. The RFE method was used to select the conditioning factors, while PSO was used to tune the hyper-parameters. Secondly, considering the angular relationship between the pipelines and landslides, and the segmentation of the pipelines using the fuzzy clustering (FC), the CRITIC method (FC-CRITIC) was combined to develop a pipeline vulnerability assessment model. Accordingly, a pipeline risk map was obtained based on pipeline vulnerability and landslide susceptibility assessment. The study results show that almost 35.3% of the slope units were in extremely high susceptibility zones, 6.68% of the pipelines were in extremely high vulnerability areas, the southern and eastern pipelines segmented in the study area were located in high risk areas and coincided well with the distribution of landslides. The proposed hybrid machine learning model for landslide-oriented risk assessment of long-distance pipelines can provide a scientific and reasonable risk classification for new planning or in service pipelines to avoid landslide-oriented risk and ensure their safe operation in mountainous areas.

Keywords: CRITIC; Fuzzy clustering; Landslide susceptibility; Pipeline risk assessment; RFE-PSO-AdaBoost.

MeSH terms

City Planning
Geographic Information Systems
Landslides*
Machine Learning
Risk Assessment / methods