摘要
为了弥补现有方法判识结果缺少实际水循环的支撑与验证,以及对实际采矿过程中涌(突)水现象与矿井立体水文地质模型等结合不足的问题,提出一种基于水化学场机器学习分析与水动力场反向示踪模拟耦合的矿井涌(突)水水源综合判识技术。该技术首先利用水文地球化学的原理揭示矿井涌(突)水及其可能来源含水层(水体)的水化学特征,利用特征的相似性对涌(突)水来源进行定性分析;随后利用机器学习算法对涌(突)水来源进行定量判识;最后建立渗流场数值模型,实现涌(突)水来源的再验证与涌水路径的可视化输出。以曹家滩煤矿为工程实例,运用该方法对122108和122109两个工作面的涌水来源进行判识,研究结果表明:随着深度的增加,研究区地下水中阴离子始终以HCO_(3)^(-)为主导,而阳离子则呈现由Ca^(2+)为主导过渡到Na^(+)+K^(+)为主导的趋势;支持向量机(SVM)需要额外利用遗传算法(GA)等方法对惩罚系数c和核函数参数g进行优选,而随机森林(RF)无需复杂的参数设置和优化便能得到较为满意的性能,且具有更高的准确性;矿井涌(突)水渗流场可视化模型反向示踪显示122109工作面在红土隔水层“天窗”附近,存在第四系含水层地下水通过导水裂隙带涌入工作面的情况。该技术判识出122108工作面涌水来源于直罗组和延安组含水层地下水,122109工作面涌水来源于第四系含水层地下水,判识结果与工程实际情况相吻合。
Accurate source discrimination of mine water inflow or inrush is of great significance to ensure the sustainable and safe production of coal mines. A comprehensive source identification technique of mine water inflow or inrush based on the hydrochemical field machine learning analysis and hydrodynamic field reverse tracer simulation is proposed, in order to make up for the lack of support and verification of the actual water cycle in the identification results of the present methods, as well as the insufficient combination of mine water inflow or inrush phenomenon and mine three-dimensional hydrogeological model in the actual mining process. Firstly, the principle of hydrogeochemistry is used to reveal the hydrochemical characteristics of mine water inflow or inrush and its possible source aquifer(water body),and the similarity of characteristics is used to qualitatively analyze the source of water inrush.Then, the machine learning algorithm is used to quantitatively identify the source of water inflow or inrush. Finally, the numerical model of the seepage field is established to realize the re-verification of water source and the visual output of water path. Taking the Caojiatan Coal Mine as an engineering example, this method is used to identify the water inflow sources of No.122108 and No.122109 working faces. Research results show that the anions in groundwater in the study area are always dominated by HCO_(3)^(-),while the cations show a trend of transitioning from the dominance of Ca^(2+)to the dominance of Na^(+)+K^(+) with the increase of depth. Support Vector Machine(SVM)requires an extra Genetic Algorithm(GA)to optimize penalty coefficient c and kernel function parameter g. Random Forest(RF)can obtain satisfactory performance without complicated parameter setting and optimization, and has higher accuracy. Visualization model of mine water inflow or inrush seepage field reverse tracing shows that the NO.122109 working face is located nearby in the skylight of laterite aquifuge, and there is a situation that groundwater in the Quaternary aquifer flows into working face through water-conducting fractured zone. The result of the NO.122108 working face water inflow identified by the method is the groundwater of the Zhiluo Formation and the Yan’an Formation aquifers, and the NO.122109 working face is the groundwater of the Quaternary aquifer. The identification results are consistent with the actual situation of the coal mine.
作者
曾一凡
梅傲霜
武强
华照来
赵頔
杜鑫
王路
吕扬
潘旭
ZENG Yifan;MEI Aoshuang;WU Qiang;HUA Zhaolai;ZHAO Di;DU Xin;WANG Lu;LÜYang;PAN Xu(National Engineering Research Center of Coal Mine Water Hazard Controlling,China University of Mining and Technology-Beijing,Beijing 100083,Chi-na;Shaanxi Shanmei Caojiatan Mining Co.,Ltd.,Yulin 719000,China;China University of Mining and Technology-Beijing(Nanjing)New Energy and Environmental Protection Technology Research Institute,Nanjing 210005,China)
出处
《煤炭学报》
EI
CAS
CSCD
北大核心
2022年第12期4482-4494,共13页
Journal of China Coal Society
基金
国家自然科学基金资助项目(42072284,42027801,41877186)。
关键词
水源判识
水化学特征
机器学习算法
粒子反向示踪
顶板水害
water source identification
hydrochemical characteristics
machine learning algorithm
reverse tracer particle
roof water hazards