首页 > 论文 > 光谱学与光谱分析 > 39卷 > 8期(pp:2425-2430)

LIF和CNN的矿井突水水源类型判别

Online Discrimination Model for Mine Water Inrush Source Based CNN and Fluorescence Spectrum

  • 摘要
  • 论文信息
  • 参考文献
  • 被引情况
  • PDF全文
分享:

摘要

矿井进入深部开采过程中, 突水威胁分别来自顶板老空水和底板高压岩溶水。 煤矿突水水源类型的在线识别能够预警煤矿水害, 是矿井水害防治关键环节, 对煤矿安全生产具有积极意义。 代表离子法作为传统的煤矿突水水源类型识别方法, 需要深入现场采集水样, 密封处理后在实验室检测水样中7种典型的无机离子浓度, 计算得到突水评价因子。 这种存在检测周期过长、 样品易被污染以及预警响应滞后、 无法在线判别等不利因素。 针对代表离子法方法的不足, 提出了一种基于激光诱导荧光(LIF)和卷积神经网络(CNN)的矿井突水水源判别模型。 首先, 针对淮南矿业集团新集二矿的4种水体, 2016年6月—2017年6月期间分批次取得161组水源样本, 其中采空区积水46条, 砂岩水59条, 太灰水42条和奥灰水14条。 用LIFS-405激光诱导荧光系统发射的405 nm激光检测水样, 水体受激后得到突水水样的荧光光谱。 主成分分析得到前10个主成分累计贡献率不足85%, 4种水样无法有效直接辨识。 针对该问题和水样荧光光谱中的随机高频波动干扰, 采用一阶滞后滤波方法抑制波动频率较高的周期性干扰; 针对线判别分析对数据更新率的要求, 采用递推平均方法; 在此基础上, 提出了一种改进的递推平均一阶滞后平滑滤波方法, 并对滤波处理后的荧光光谱进行自相关计算, 得到二维自相关荧光光谱特征图。 实验表明, 采用改进后的滤波法处理方法, 计算得到的4种测试水样的二维荧光光谱图较好的滤除了噪声干扰, 并表现了出了明显的差异性。 针对二维自相关荧光光谱特征图, 构建了基于卷积神经网络(CNN)的突水水源类型判别模型, 用于判别突水水源类型。 该方法采用深度学习的模型框架, 直接对二维自相关荧光光谱特征图进行识别, 有效避免了PCA降维的片面性。 理论分析和实验结果表明: 该模型对水源类型的准确识别率达到了98%, 是一种有效的矿井突水水源类型判别方法, 为在线矿井突水水源类型判别方法提供了新的思路。

Abstract

As deep mining goes, the water inrush threat is from the roof goaf water and the bottom pressure karst water. Coal mines water inrush water types on-line discrimination, serving as an effective monitoring method to predict mine water hazards, is an important step in Mine water disaster prevention and control work to ensure coal mine safety production. Representative ion method, as a traditional method to discriminate mine water inrush sources, must collect and seal water samples on-site, test samples in laboratory using 7 typical inorganic ion concentrations, and calculate water bursting evaluation factor. The method has disadvantages of too long detection time,easy contamination for samples, delayed warning response and misjudgment. Due to above reasons, the paper proposes a mine water inrush sources discrimination model based on Laser Induced Fluorescence (LIF) and Convolutional Neural Network (CNN). First, based on 4 types of water sources, 161 samples were collected from Xinji Second Mine of Huainan mining group during June 2016 to June 2017, including oaf water 46 items, Sandstone water 59 items, Limestone water 42 items and Ordovician limestone water 14 items. In the experiment, samples were stimulated by 405 nm laser using LIFS-405 Laser Induced Fluorescence System, and the fluorescence spectra of four kinds of 161 groups of water inrush samples were obtained. During principal component analysis, the cumulative contribution rate of the top ten components was less than 85%, making 4 types of water samples almost indistinguishable. Second, considering the random high frequency fluctuations in water fluorescence spectra, first-order lags filtering method should be used to reduce periodic high frequency fluctuations. Considering data update rate, recursive averaging method should be adopted. The paper proposes an improved recursive average first-order lag smoothing filtering method further to calculate autocorrelation processing to get enhanced two-dimensional autocorrelation characteristic fluorescence spectra. The experimental results show that calculated autocorrelation characteristic fluorescence spectra have excellent performance on interference elimination and discrimination. Finally, based on autocorrelation characteristic fluorescence spectra, mine water inrush sources discrimination model using CNN was constructed to discriminate water inrush types. The method adopts deep learning framework using autocorrelation characteristic fluorescence spectra to avoid selecting features in subjective ways. Theoretical analysis and experimental results show that the correct recognition rate of water source type can reach 98%. It is an effective way to discriminate the source of water inrush from mines and provides a new idea to discriminate the types of mine water inrush sources.

广告组1 - 空间光调制器+DMD
补充资料

中图分类号:O657.3

DOI:10.3964/j.issn.1000-0593(2019)08-2425-06

基金项目:国家“十二五”科技支撑计划项目(2013BAK06B01)和国家自然科学基金项目(41674133)资助

收稿日期:2018-05-22

修改稿日期:2018-10-20

网络出版日期:--

作者单位    点击查看

杨 勇:中国矿业大学资源与地球科学学院, 江苏 徐州 221008徐州工业职业技术学院, 江苏 徐州 221140
岳建华:中国矿业大学资源与地球科学学院, 江苏 徐州 221008
李 晶:南京审计大学信息工程学院, 江苏 南京 210029
张河瑞:中国矿业大学资源与地球科学学院, 江苏 徐州 221008

联系人作者:杨勇(yongyang@cumt.edu.cn)

备注:杨 勇, 1981年生, 中国矿业大学资源与地球科学学院博士研究生

【1】WU Qiang, CUI Fang-peng, ZHAO Su-qi, et al(武 强, 崔芳鹏, 赵苏启, 等). Journal of China Coal Society(煤炭学报), 2013, 38(4): 561.

【2】XU Xing, TIAN Kun-yun, WANG Gong-zhong, et al(徐 星, 田坤云, 王公忠, 等). Journal of Safety & Environment(安全与环境学报), 2017.

【3】LIU Jian-min, WANG Ji-ren, LIU Yin-peng, et al(刘剑民, 王继仁, 刘银朋, 等). Journal of Safety & Environment(安全与环境学报), 2015, (1): 38.

【4】WANG Ya, ZHOU Meng-ran, YAN Peng-cheng, et al(王 亚, 周孟然, 闫鹏程, 等). Spectroscopy and Spectral Analysis(光谱学与光谱分析), 2017, 37(3): 978.

【5】GONG Feng-qiang, LU Jin-tao(宫凤强, 鲁金涛). Journal of Mining & Safety Engineering(采矿与安全工程学报), 2014, 31(2): 236.

【6】Sun L H. Arabian Journal of Geosciences, 2014, 7(9): 3417.

【7】WANG Ya, ZHOU Meng-ran, YAN Peng-cheng, et al(王 亚, 周孟然, 闫鹏程, 等). Journal of the China Coal Society(煤炭学报), 2017, 42(9): 2427.

【8】Tirumala S S, Narayanan A. Hierarchical Data Classification Using Deep Neural Networks. International Conference on Neural Information Processing, 2015.

【9】Ali J B, Saidi L, Mouelhi A, et al. Engineering Applications of Artificial Intelligence, 2015, 42(C): 67.

【10】Poswar F D O, Farias L C, Fraga C A D C, et al. Journal of Endodontics, 2015, 41(6): 877.

【11】Dabove P, Manzino A M. GPS Solutions, 2017, 21(3): 1213.

引用该论文

YANG Yong,YUE Jian-hua,LI Jing,ZHANG He-rui. Online Discrimination Model for Mine Water Inrush Source Based CNN and Fluorescence Spectrum[J]. Spectroscopy and Spectral Analysis, 2019, 39(8): 2425-2430

杨 勇,岳建华,李 晶,张河瑞. LIF和CNN的矿井突水水源类型判别[J]. 光谱学与光谱分析, 2019, 39(8): 2425-2430

您的浏览器不支持PDF插件,请使用最新的(Chrome/Fire Fox等)浏览器.或者您还可以点击此处下载该论文PDF