Tracking the information about your manuscript
Communicate with the editorial office
Query manuscript payment status Reviewer LoginOnline Review
Online Communication with the Editorial Department Author LoginCollecting, editing, reviewing and other affairs offices
Managing manuscripts
Managing author information and external review Expert Information
Downloads
Page Views
Page visits today: 0
About the Journal
Superintended by：China North Industries Group Corporation Limited
Sponsoredy：Beijing Institute of Technology, Chinese Scoiety for Environmental Sciences, China Occupational Safety and Health Association
Edited & Published by: Editorial Department of Journal of Safety and Environment
Issues per year: 12
ISSN 1009-6094
CN 11-4537/X
Construction and application of a stability evaluation model for regenerated roof in ultra-close distance coal seam mining
HUO Bingjie;KANG Shangbo;LI Qiping;XIA Pingchuan;LIU Can;GUO Dongxu;ZHANG Chengbo;This study addresses the critical issue of safe and efficient mining of lower coal seams while considering the stability of regenerated roofs during the downward extraction of ultra-close coal seams. The research investigates theories and methodologies for evaluating regenerated roof stability to enhance mine safety. Methodologically, we construct an innovative comprehensive evaluation index system that includes three criterion layers: factors inherent to the regenerated roof itself, geological environment factors, and mining disturbance factors. This framework comprises eight specific indicators, including lithological composition and compaction-cementation duration. We propose an evaluation model that integrates the Analytic Hierarchy Process(AHP) with the Entropy Weight method and the Technique for Order Preference by Similarity to Ideal Solution(TOPSIS). This combined weighting approach balances subjective and objective influences to scientifically assess the significance of each indicator, while the TOPSIS method quantifies relative closeness to determine stability. Furthermore, utilizing the ArcGIS platform, we employed the Inverse Distance Weighting(IDW) method to visualize the spatial distribution of regenerated roof stability, leading to the formulation of targeted management strategies. We applied this methodology at Shengli Mine, focusing on the No. 14 and No. 15 coal seams, which are characterized by an exceptionally small interlayer spacing averaging 1.45 m, thereby serving as an engineering case study to evaluate the method's applicability and scientific validity. The results indicate that:(1) Lithological composition is the predominant factor affecting regenerated roof stability, with a weight of 0.543;(2) Regenerated roof stability can be categorized into four grades. In the study area, stable(Grade I) and relatively stable(Grade Ⅱ) zones of the No. 15 coal seam account for 80% of the total area, reflecting generally favorable stability conditions. Conversely, unstable(Grade Ⅲ) and extremely unstable(Grade Ⅳ) zones are primarily located in areas with complex hydrogeological conditions, minor faults, unfavorable lithological combinations, and short cementation durations;(3) A validation against a similar mining face demonstrated an 80% consistency between the model evaluation outcomes and actual field conditions. The evaluation model and methods developed in this study effectively assess and spatially predict the stability of regenerated roofs in extremely close coal seams. The findings provide a theoretical foundation and technical support for the safe mining of coal seams under comparable conditions.
Predicting peak vibration velocity of blasting particles using interpretable GA-VR integrated learning
YANG Yi;WANG Shuxian;SHEN Yaxi;SHI Yulian;DING Qiuyue;Accurately predicting Peak Particle Velocity(PPV) is essential for ensuring operational safety and minimizing environmental impact. While machine learning models present advantages over traditional empirical formulas, their “black-box” nature and limited interpretability can hinder practical applications in optimizing blast parameters. This study introduces a novel, interpretable prediction framework that combines Genetic Algorithm(GA) optimization with a Voting Regressor(GA-VR) ensemble. The model integrates three distinct base learners: the Gradient Boosting Regressor(GBR), Random Forest Regressor(RFR), and Decision Tree Regressor(DTR). The GA is utilized to globally optimize critical hyperparameters for each base learner: for GBR, nestimators(range: 75-105), maxdepth(3-10), learningrate(0.020-0.030), and minsamplessplit(5-12); for RFR, maxdepth(5-20), nestimators(150-300), and minsamplessplit(2-10); and for DTR, minsamplessplit(2-20), maxleafnodes(10-50), and maxdepth(2-30). The optimized base models are then integrated through a weighted average voting mechanism within the VR framework. Additionally, SHapley Additive exPlanations(SHAP) analysis is employed to provide both global and local interpretability of the model's predictions. Experimental results based on field data from the Daye Nonferrous Metals Group Tonglvshan open-pit mine demonstrate the superior performance of the GA-VR model, which achieved a coefficient of determination(R2) of 0.948 on the independent test set. This significantly outperformed six individual GA-optimized base models: GA-DTR(R2 = 0.843), GA-ETR(R2 = 0.797), GA-GBR(R2 = 0.881), GA-KNN(R2 = 0.648), GA-RFR(R2 = 0.878), and GA-SVR(R2 = 0.696). Furthermore, GA-VR surpassed VR models optimized by five alternative algorithms: Gradient Descent(GD-VR, R2 = 0.903), Particle Swarm Optimization(PSO-VR, R2 = 0.893), K-Nearest Neighbors Regression(KNN-VR, R2 = 0.912), Firefly Algorithm(FA-VR, R2 = 0.909), and Ant Colony Optimization(ACO-VR, R2 = 0.890). SHAP analysis revealed that the horizontal distance between the measurement point and the blast location(X3) was the most influential factor in PPV predictions, followed by the maximum charge per delay(X1) and the angle between the measurement point and the direction of the minimum resistance line(X11). Local interpretability analysis for specific samples(e.g., sample 93#) further illustrated the contribution of each feature to individual predictions. Overall, the GA-VR model not only delivers high-precision PPV forecasting but also provides a transparent and interpretable rationale for its predictions, making it a powerful tool for the intelligent design of blast parameters and enhanced safety control in mining and construction engineering.
Interpretable prediction of rockburst intensity in deep high-stress zones with imbalanced data
CHENG Lianhua;XUE Kailong;QI Yun;DUAN Hongfei;YAO Rui;DONG Xinyue;FU Zhibo;To tackle the challenges of imbalanced sample data and low prediction accuracy in rockburst intensity prediction models for deep high-stress areas, this study proposes a rockburst intensity prediction model that combines an Improved Dung Beetle Optimizer(IDBO) with the Light Gradient Boosting Machine(LightGBM). First, drawing on the genetic characteristics of rock bursts, four feature factors are selected to establish the prediction dataset: Maximum Tangential Stress(MTS) of surrounding rock, Unconfined Compressive Strength(UCS), Unconfined Tensile Strength(UTS), and elastic energy index(Wet). Next, the Local Outlier Factor(LOF) algorithm is employed to eliminate outlier samples from the dataset, while the Synthetic Minority Oversampling Technique(SMOTE) is utilized to increase the number of minority samples, addressing the issue of sample imbalance among different rockburst intensities. Subsequently, three strategies—Bernoulli chaotic map, golden sine, and adaptive Levy flight—are integrated to enhance the original Dung Beetle Optimizer(DBO) algorithm, resulting in the IDBO. This IDBO is then used to optimize the hyperparameters of LightGBM, including num＿leaves(maximum number of leaves), learning＿rate(learning rate), colsample＿bytree(feature sampling ratio), max＿depth(maximum depth), n＿estimators(number of trees), reg＿alpha(L1 regularization coefficient), and reg＿lambda(L2 regularization coefficient), with the goal of improving prediction accuracy. The prediction results of the proposed model are compared with those of IDBO-SVM(Support Vector Machine), IDBO-XGBoost(eXtreme Gradient Boosting), IDBO-RF(Random Forest), IDBO-BPNN(Back Propagation Neural Network), and IDBO-GBDT(Gradient Boosting Decision Tree) models to demonstrate its superiority. Additionally, an interpretable analysis of the IDBO-LightGBM model's prediction results is conducted based on SHAP theory, revealing the specific contributions of each feature factor to the model's predictions. Finally, the proposed model is applied to practical rockburst prediction in the Maluping mining area. The results indicate that the combination of LOF and SMOTE significantly enhances the model's prediction accuracy. Among all models assessed, the IDBO-LightGBM model demonstrates the best performance in predicting rockburst intensity, achieving an accuracy of 97.75%. SHAP analysis reveals that the elastic energy index(Wet) has the largest SHAP value, indicating it contributes most significantly to rockburst intensity predictions, while the UCS has the smallest SHAP value, contributing the least. Furthermore, the practical prediction outcomes of the IDBO-LightGBM model in engineering align perfectly with actual results, confirming that this model can effectively predict rockburst intensity. The research findings provide a new theoretical foundation and technical reference for the accurate prediction of rockburst intensity in underground engineering.
Inhibition effect of dodecyl surfactant on gas diffusion in tectonic coal
MA Hongyu;WANG Long;WAN Wen;ZHANG Zongxiang;XIAO Yao;ZHAO Pengtao;Injecting active solutions into coal seams not only enhances the wettability of coal but also influences gas emission characteristics. To investigate the inhibition effects of dodecyl surfactants on gas diffusion in tectonic coal, we conducted experiments using tectonic coal treated with Sodium Dodecyl Benzene Sulfonate(SDBS) and Sodium Dodecyl Sulfate(SDS) solutions, both individually and in combination, at varying mass fractions. The effects of these active solutions on gas diffusion volume, diffusion velocity, diffusion coefficient, and inhibition ratio were systematically analyzed. This study aims to elucidate the synergistic inhibition mechanism of the SDBS-SDS blended system on gas diffusion and quantitatively assess its suppression effectiveness. The results indicate that as the mass fraction of the solution increases, the contact angle between the active solution and coal exhibits a “V”-shaped trend. Overall, SDBS solutions demonstrate superior wetting performance on coal compared to SDS solutions. For individual surfactant solutions, optimal inhibition of gas diffusion is observed at a mass fraction of 3% for both SDBS and SDS. At this concentration, the measured gas diffusion coefficients are 2.059×10-6 cm2/s and 2.465×10-6 cm2/s, respectively, corresponding to reductions of 62.9% and 55.6% relative to dry coal. However, when the mass fraction of SDS is increased to 5%, the gas inhibition performance deteriorates, which can be attributed to hydrophobic reversal effects that compromise wetting efficiency. In blended surfactant systems, a notable synergistic inhibition effect on gas diffusion is achieved when SDBS and SDS are combined at a volume ratio of 1:1. This optimal blend results in a diffusion coefficient of only 2.241×10-6 cm2/s for wetted coal, with an inhibition rate exceeding that of the SDS solution alone by more than 10 percentage points. Conversely, deviations from the 1:1 volume ratio lead to decreased micelle stability in the solution, weakening gas diffusion efficiency. This study provides a formulation optimization strategy for gas control in coal seams through liquid injection.
Division of spontaneous combustion “three zones” in goaf based on fuzzy C-means clustering algorithm
PAN Jingtao;QIU Chen;ZHAO Dan;LIU Liren;YUAN Qiang;To address the challenges posed by fuzzy boundaries and the limitations of traditional methods in managing uncertainty when delineating the spontaneous combustion “three zones” in goafs, this study proposes an enhanced division method based on the Fuzzy C-Means(FCM) clustering algorithm. A multi-dimensional evaluation index system that integrates temperature, oxygen concentration, carbon monoxide levels, and burial depth was established to capture the complex environmental characteristics. The study develops a comprehensive mathematical model encompassing data standardization and calibration using Manhattan distance, the construction of fuzzy similarity matrices, and their transformation into equivalence matrices via transitive closure. Utilizing MATLAB, a dynamic clustering mechanism based on λ-cut sets was implemented. Notably, to determine the optimal classification granularity, an adaptive threshold selection method for λ was developed by maximizing the geometric mean of F-statistics across multiple spatial monitoring points. Results indicate that clustering sensitivity exhibits spatial variation: monitoring points near the working face respond sensitively to changes in λ, reflecting high environmental heterogeneity caused by air leakage, while deeper goaf areas demonstrate strong stability. Using the optimal λ, the goaf was distinctly categorized into heat dissipation, oxidation, and suffocation zones. The identified ranges for the oxidation zone are 72-96 m on the intake side and 36-72 m on the return side. In comparison with the on-site “Disaster Comprehensive Intelligent Management Platform”, which is based on single oxygen thresholds(intake: 72-103 m; return: 36-66 m), this method effectively identified a 7 m “sub-oxidation zone” on the intake side where oxidation remains weak despite high oxygen levels, and it extended the risk boundary by 6 m on the return side, revealing a “latent risk area” attributed to thermal inertia. The relative deviations of the boundary endpoints are 6.8% and 9.1%, respectively. This data-driven approach uncovers risk details overlooked by single-threshold methods, providing a more scientific foundation for precise fire prevention.
