Prediction Model for Harmful Gas Risk Levels in Non-Coal Strata Tunnels Based on SSA-CatBoost
Open Access
Abstract
Harmful gas is a major hazard in underground engineering construction, and accurate prediction of its risk level is essential for tunnel safety. Existing prediction methods for harmful-gas risk in non-coal strata tunnels are limited by empirical scoring, subjective indicator assignment, and insufficient quantitative characterization of reservoir performance. To address these limitations, this study proposes an SSA-CatBoost prediction model for harmful-gas risk levels in non-coal strata tunnels. Eight influencing indicators were selected as input variables. Among them, reservoir performance was quantitatively characterized by measured porosity and permeability, while the other six indicators were quantified using engineering-based scoring criteria. A database containing 138 real harmful-gas tunnel cases was constructed, and CatBoost was used as the base classifier. The Sparrow Search Algorithm was introduced to optimize the hyperparameters of CatBoost. The proposed SSA-CatBoost model achieved an average accuracy of 93.63% in five-fold cross-validation and an accuracy of 92.86% on the independent test set. Compared with CatBoost, SSA-SVM, and SSA-XGBoost, the proposed model showed the highest cross-validation accuracy. Engineering validation further showed that all selected validation samples were correctly classified. In addition, replacing empirical reservoir-performance scoring with measured porosity and permeability improved the recognition performance of adjacent risk levels, with the F1-scores of Level III and Level IV increasing from 0.667 and 0.727 to 0.909, respectively. The novelty of this study lies in integrating measured reservoir-performance parameters into a machine-learning-based harmful-gas risk prediction framework, thereby reducing the subjectivity of conventional scoring systems and improving the quantitative characterization of non-coal strata tunnel gas hazards.Keywords
Funding Information
- Science and Technology Department of Sichuan Province (2024ZDZX0009)
This publication has 16 references indexed in Scilit:
- A prediction model for hazard levels of shallow natural gas in tunnel based on K-means clustering and tabular prior-data fitted networkResults in Engineering, 2025
- Research on gas tunnel prediction in Central Sichuan using energy valley optimizer and support vector machineBulletin of Engineering Geology and the Environment, 2025
- Coal mine gas emission prediction based on multifactor time series methodReliability Engineering & System Safety, 2024
- Quantitative evaluation of the indexes contribution to coal and gas outburst prediction based on machine learningFuel, 2023
- Prediction of coal mine gas emission based on hybrid machine learning modelEarth Science Informatics, 2022
- Coal and gas outburst prediction model based on extension theory and its applicationProcess Safety and Environmental Protection, 2021
- Predicting rockburst with database using particle swarm optimization and extreme learning machineTunnelling and Underground Space Technology, 2020
- Study on formation mechanism of gas tunnel in non-coal strataNatural Hazards, 2012
- A case study on the methane explosion in the excavation chamber of an EPB-TBM and lessons learnt including some recent accidentsTunnelling and Underground Space Technology, 2011
- The application of a coupled artificial neural network and fault tree analysis model to predict coal and gas outburstsInternational Journal of Coal Geology, 2010