基于水位、水温、突水量和水质的充水水源识别神经网络模型  

Neural network model of water filling source identification based on water level,water temperature,water intrusion,and water quality

作　　者：桑向阳 林云[2] 刘保民[2] 潘国营[2] SANG Xiangyang;LIN Yun;LIU Baomin;PAN Guoying(Geological Survey Department,China Pingmei Shenma Group,Pingdingshan 467000,Henan,China;School of Resources and Environment,Henan Polytechnic University,Jiaozuo 454000,Henan,China)

机构地区：[1]中国平煤神马集团地质测量处,河南平顶山467000 [2]河南理工大学资源环境学院,河南焦作454000

出　　处：《河南理工大学学报（自然科学版）》2024年第5期36-42,共7页Journal of Henan Polytechnic University（Natural Science）

基　　金：国家自然科学基金资助项目(42271041);河南省高等学校青年骨干教师培养计划项目(2021GGJS055)。

摘　　要：华北石炭-二叠系岩溶型煤田主采煤层底板的太原组薄层灰岩岩溶水和奥陶系或寒武系厚层灰岩岩溶水的水化学特征存在着天然相似性,单纯以若干项水化学指标辨识这些水源存在着误判甚至错判的风险。目的为解决寒武系灰岩水、太原组下段灰岩水和部分太原组上段灰岩水的水质指标相似而难以完全正确识别的问题,方法提出并构建基于水位、水温、突水量和水质识别充水水源的神经网络模型。以平顶山矿区充水水源识别为例,以阴阳离子毫克当量百分比[γ(Na)%,γ(Ca+Mg)%,γ(HCO_(3)^(-))%,γ(SO_(4)+Cl)%]、钠钙比、碱硬比、ρ(CO_(3)^(2−))、ρ(SO_(4)^(2−))、TDS、ρ(Na+K)、水位、水位动态变化、水温、突水量、衰减天数共15项指标作为识别因子,构建结构为15-10-6的神经网络模型。结果结果表明,所有训练样本对自身水源的拟合均值均超过0.98,比单纯以水质指标为识别因子的建模方法识别正确率高,能够有效消除因水质指标相似但水源不同而出现的误判或错判情况。结论建模方法已经嵌入平顶山矿区识别充水水源计算机软件和手机APP软件中,经过检验,识别正确率达到91.3%。The water chemical characteristics of thin limestone karst water in the Taiyuan Formation and thick Ordovician or Cambrian thick limestone karst water in the main coal seam floor of Carboniferous-Permian karst coal field in North China are naturally similar.This similarity poses a risk of misjudgment or even miscalculation when relying solely on certain hydrochemical indexes.Objectives The water quality in-dexes of limestone water,L_(2) limestone water,and some L_(7) limestone water are similar,making accurate iden-tification challenging.To address this issue,Methods a neural network model for identifying water sources based on water level,temperature,quantity,and quality was proposed.Taking the filling water source identifi-cation of the Pingdingshan mining area as an example,a 15-10-6 neural network model was constructed with 15 indexes as identification factors,including the anion and cation percentages in milligram equiva-lents,the ratio of sodium to calcium,the ratio of alkali to hardness,ρ(CO_(3)^(2−)),ρ(SO_(4)^(2−)),TDS,ρ(Na+K),water level,dynamic change,water temperature,water intrusion,and attenuation days.Results The experimental re-sults showed that the mean value of all training samples’fitting to their own water sources exceeded 0.98,which significantly improved the recognition accuracy compared with the modeling method that simply took water quality index as the recognition factor,and could completely and effectively eliminate the misjudgment caused by similar water quality indexes but different water sources.Conclusions The proposed modeling method had been incorporated into the computer software and mobile app software for identifying water sources in the Pingdingshan mining area.After testing,the recognition accuracy reached 91.3%.

关 键 词：煤矿水源识别 水位 水温 突水量 水质 神经网络 

分 类 号：TD12[矿业工程—矿山地质测量]