|
Title |
Selection of Ground Subsidence Assessment Factors and Development of a Multistage Risk Assessment Model and Criteria
|
|
Authors |
허춘근(Choun-Kun Hur) ; 이창훈(ChangHoon Lee) ; 박정찬(Jung-Chan Park) ; 송수민(Su-Min Song) ; 오홍섭(HongSeob Oh) |
|
DOI |
https://doi.org/10.11112/jksmi.2026.30.5.119 |
|
Keywords |
계층분석법; 평가인자; 지반침하; 안전등급; 지하안전 Analytic hierarchy process; Evaluation indicators; Ground subsidence; Safety grade; Underground safety |
|
Abstract |
Urban underground development and aging infrastructure have increased the need for systematic assessment methods that identify ground subsidence risks and prioritize preventive investigations. This study developed a multistage assessment framework by integrating evidence from domestic studies and international practices with simulation-based indicator selection and expert judgment. An initial set of 23 candidate indicators was examined for engineering relevance, data availability, and influence on assessment results. Analyses using the analytic hierarchy process (AHP), weights of evidence (WofE), and tensor field network (TFN) incorporated records from 867 accident locations. TFN analysis and engineering interpretation distinguished inherent ground characteristics from factors associated with changing conditions. Subsequent WofE simulation achieved an accuracy of 89.3% and a recall of 84.7%, exceeding the study targets of 80% for both measures. The selected indicators were organized into nationwide primary assessment and secondary assessment for open-cut excavation, tunnel excavation, or warning histories, using 100 m by 100 m grids. AHP was reapplied to the finalized assessment tables to determine indicator weights. Weighted scores were converted into four grades for each assessment stage, and a matrix combined the primary and secondary grades into Alert, Caution, Attention, or Normal management levels. This structure combines inherent vulnerability with excavation or warning influences. Application to seven accident sites and three comparison sites without registered accidents classified six accident sites as Alert and all comparison sites as Normal. With Alert defined as the risk classification, agreement reached 90% within these purposively selected cases. The framework provides a consistent basis for prioritizing investigations and allocating safety management resources.
Future research will apply tree-based artificial intelligence models to sensitivity analysis and weight refinement, followed by independent verification using diverse accident and comparison cases.
|