The prediction of goaf stability is a critical component of mine safety management.Accurate forecasting of goaf stability during mining operations is essential for ensuring both the safety and environmental integrity of mining activities.Due to the intricate nonlinear relationship between goaf stability and its influencing factors,traditional prediction methods often fall short in delivering precise results.Machine learning,however,is well-equipped to address this issue.To overcome the challenges of algorithms becoming trapped in local optima and exhibiting suboptimal convergence speed and accuracy,a stability prediction model for goaf,based on the SIDBO-BP algorithm,was proposed to effectively classify and predict the stability of mine goafs.In light of the internal and external factors influencing goaf stability,nine specific factors were identified based on the unique conditions of the mine:Burial depth (X1),exposed roof area (X2),high collapse ratio (X3),goaf volume (X4),goaf dip angle (X5),conditions of adjacent goafs (X6),rock compressive strength (X7),geological structure (X8),and rock mass structure (X9).These factors were utilized to classify the stability status of four goafs,which served as output levels in the construction of a goaf stability prediction model. The study utilized ninety-three datasets from a lead-zinc mine in Yunnan as the research object.Correlation and visualization analyses of the goaf data were performed using graphical correlation analysis and correlation coefficient graph methods,thereby validating the appropriateness of the selected factors.The goaf data were normalized using MATLAB.Subsequently,the SIDBO algorithm was applied for optimization,yielding the optimal position and fitness,which were then employed in a BP neural network to determine the optimal threshold weight for prediction.The macro-average method was employed to compute the metrics of accuracy,precision,specificity,recall,and F1 score for the predictive outcomes.The model’s predictions were assessed and benchmarked against those generated by other algorithmic models.The findings revealed that only a single prediction from the SIDBO-BP model was inaccurate.The SIDBO-BP model demonstrated superior accuracy,precision,specificity,recall,and F1 score compared to the other algorithmic models,aligning closely with the actual results.Additionally,its overall performance exceeded that of the other five models.These results suggest that the SIDBO-BP model provides substantial advantages in terms of prediction accuracy,convergence speed,and the avoidance of local optima.
随着矿山开采工作的不断深入,采空区的产生与变化对地面环境及周围建筑的安全具有重要影响(郑怀昌等,2005)。因此,对矿山开采过程中采空区稳定性进行精确预测,对于保证矿井的安全和环保具有重要意义。采空区稳定性预测方法主要有基于层次分析(AHP)法的模糊综合评价法、概率积分法、神经网络和数值模拟法(周逸文等,2022)。基于层次分析(AHP)法的模糊综合评价法是一种应用广泛的多准则决策方法(Liu et al.,2020),该方法对各备选方案优选指标及权重进行结构化确定,进而实现对复杂矿山开采过程中随机与模糊问题的全面、量化表征。采用数值模拟法对采空区围岩应力—应变—位移进行数值仿真(张敏等,2012;Guo et al.,2021;Jia et al.,2021),揭示采空区的变形规律,考虑空间与时间的关系,进而直观、动态地展现各影响因子对采空区风险的作用机理。矿山开采过程中存在高度的复杂和不确定因素,利用传统方法对采空区现场环境进行完全真实预测是很难实现的。
随着人工智能和机器学习等先进技术的不断发展,多变量、非线性和复杂问题的研究呈现多样化。因此,机器学习备受关注和重视,且已有部分相关研究成果用于采矿领域。利用自适应遗传算法与神经网络相结合的方法来评价采空区危险性具有一定的优越性(李孜军等,2015)。相比单模型的机器学习方法,运用强学习器进行组合的Stacking集成学习方法能够对采空区稳定性进行更精准的预测评价(王牧帆等,2020)。利用优化算法与支持向量机(SVM)、数值模拟方法相结合的方法,能够对采空区危险性进行更全面和精准的预测,为采空区危险性预测提供了一种新途径(Li et al.,2024)。
根据确定的采空区稳定性预测模型,设置种群数目为30,最大迭代次数为100;设置神经网络的输入节点数为7,隐藏节点数为5,输出节点数为4,学习率为0.01,训练目标为1E-4,最大训练次数为100次。将表2所示的93组采空区数据代入到算法模型中进行计算,得到相应的采空区稳定性预测等级。为了进一步验证SIDBO-BP模型的预测精度,将其精度与IDBO-BP、QHDBO-BP、PSO-BP和DBO-BP这4种算法模型(李斌等,2024;刘艺梦等,2024;Zhu et al.,2024)和原始BP神经网络的精度进行对比,6种采空区稳定性预测模型预测结果如图5所示。
ChenJiao, LuoZhouquan, HouZaoshui,2013.Stability evaluation of metal mine goaf based on improved catastrophe progression method[J].Journal of Safety Science and Technology,9(11):17-24.
[2]
ChengAibao, GuDesheng, LiuHongqiang,2011.Weights analysis of factors affecting the stability of mined-out areas based on analytic hierarchy process and rough sets theory[J].Journal of Safety Science and Technology,7(9):50-55.
[3]
GuoQ B, MengX R, LiY M,et al,2021.A prediction model for the surface residual subsidence in an abandoned goaf for sustainable development of resource-exhausted cities[J].Journal of Cleaner Production,279:123803.
[4]
HuangYinghua, WenLei, HuangMin,et al,2019.Study on fuzzy evaluation system of the complex goaf stability in metal mine[J].Mining Research and Development,39(12):63-67.
[5]
JiaH H, XueJ Z,2021.The stability study of goaf based on C-ALS data point cloud and FLAC3D coupled modeling[J].E3S Web of Conferences,261:03053.
[6]
LiBin, GaoPeng, GuoZiqiang,2024.Improved dung beetle optimizer to optimize LSTM for photovoltaic array fault diagnosis[J].Proceedings of the CSU-EPSA,36(8):70-78.
[7]
LiM L, LiK G, LiuY D,et al,2024.Goaf risk prediction based on IAOA-SVM and numerical simulation:A case study[J].Underground Space,15(2):153-175.
[8]
LiYingshun, YuAng, LiMao,et al,2025.Armored vebicle engine fault diagnosis based on KLDA-IDBO-BP[J].Acta Armamentarii,46(3):107-115.
[9]
LiZijun, LinWuqing, ChenYang,2015.Evaluation on risk of goaf based on AGA-BP neural network[J].Journal of Safety Science and Technology,11(7):135-141.
[10]
LiuY, EckertM C, EarlC,2020.A review of fuzzy AHP methods for decision-making with subjective judgements[J].Expert Systems With Applications,161:113738.
[11]
LiuYimeng, DingXiaoming, WangHuiqiang,et al,2024.Prediction model for winter and summer lettuce root zone temperature based on dung beetle algorithm to optimize BP[J].Transactions of the Chinese Society of Agricultural Engineering,40(5):231-238.
[12]
MengHui, ZhangJiahong, LiMin,et al,2020.Research on prediction and classification of coronary heart disease based on IPSO-BP neural network and BCG signal[J].Chinese Journal of Sensors and Actuators,33(10):1379-1385.
[13]
NingJianguo, WangQi, LiJianqiao,2025.Artificial neural network-based prediction model for the initial velocity of fragments in a triangular prism directional charging structure[J].Acta Armamentarii,46(3):202-215.
[14]
WangChao, GuoJinping, WangLiguan,2015.Recognition of goaf risk based on support vector machines method[J].Journal of Chongqing University,38(4):85-90,127.
[15]
WangMufan, LuoZhouquan, YuQi,2020.Stability prediction of goaf based on stacking model[J].Gold Science and Te-chnology,28(6):894-901.
[16]
WangXiaodong, YangYijie, YuqiLü,et al,2024.Prediction of open air water inflow based on improved multivariate time series model[J].Journal of Safety and Environment,24(8):2994-3004.
[17]
XueJ K, ShenB,2022.Dung beetle optimizer:A new meta-heuristic algorithm for global optimization[J].The Journal of Supercomputing,79(7):7305-7336.
[18]
YuanDongliang, HeLin, RenLianwei,et al,2023.Study on influencing factors of goaf stability based on DEMATEL-ISM[J].Coal Technology,42(2):5-8.
[19]
ZhangMin, NieLei, GuoYuan,et al,2012.Simulation on influence of fault on goaf surface subsidence[J].Journal of Jilin University(Earth Science Edition),42(6):1834-1838.
[20]
ZhangXiaojun,2006.Sensitivity analysis of factors influencing the stability of mined-out area[J].Mining Research and Development,(1):14-16.
[21]
ZhaoChao, MaJunchao, WanLiming,et al,2016.Risk evaluation of underground goaf collapse based on the hierarchy extension analysis[J].Safety and Environmental Engineering,23(6):35-40.
[22]
ZhengHuaichang, LiMing,2005.Hazard and analysis of underground goaf[J].Journal of Mining and Safety Engineering,(4):127-129.
[23]
ZhouYiwen, ZhangTao, DuanLongchen,et al,2022.Summary of research on comprehensive treatment of mine goaf in China[J].Safety and Environmental Engineering,29(4):220-230.
[24]
ZhuF, LiG S, TangH,et al,2024.Dung beetle optimization algorithm based on quantum computing and multi-strategy fusion for solving engineering problems[J].Expert Syste-ms with Applications,236:121219.