基于iPLS的矿井突水激光诱导荧光光谱特征波段筛选
|
摘要
矿井突水一直威胁着煤矿井下施工人员的生命安全,准确且快速识别矿井突水水源类型对于矿井的安全生产起到关键性作用。激光诱导荧光(LIF)光谱技术识别矿井突水水源,有效避免了常规的水化学法需要测定多种化学参数,水源识别时间过长的缺点。提出一种间隔偏最小二乘法(iPLS)与粒子群联合支持向量分类算法(PSO-SVC)相结合的方法, iPLS算法常应用于光谱波段优选和模型的回归分析, PSO-SVC则在机器学习领域有着重要的应用,激光诱导荧光技术具有快速的时间响应、测量精度高等特点, iPLS和PSO-SVC算法运用于光谱图和光谱数据的分析,进而可以对突水水源类型识别分类。首先,用淮南矿区采集到的7种(每种水样30组)共210组荧光光谱数据进行实验,对老空水、灰岩水、灰岩水和老空水不同体积比混合水样的激光诱导荧光光谱图的差异性进行分析。比较了留出法和Kennard-Stone样本划分方法所得到的PSO-SVC模型分类准确率,采用留出法得到的训练集水样(140组)和测试集水样(70组)作为实验样本。其次,用iPLS算法将全光谱波段依次按10～25波段区间进行等分,选取划分区间的RMSECV(交叉验证均方根误差)值小于全光谱波段RMSECV值(阈值)的波段作为特征波段,结合光谱图对比分析了划分10和14个子区间的建模结果,发现通过直接观察得到的特征波段与iPLS算法筛选出的特征波段存在误差。最后,在不进行去噪、降维等预处理条件下,根据iPLS划分不同区间数的评价指标统计数据,选取划分11个区间所筛选出具有561个波长点的410.078～478.424和545.078～674.104 nm特征波段范围数据作为PSO-SVC模型的输入,以iPLS结合PSO-SVC算法筛选出的特征波段与全光谱波段、直接观察得到波段建模准确率相比,训练集与测试集的分类准确率高达100%, PSO寻优到的最佳惩罚系数c为1.367 0,核函数参数g为0.576 2。从实验结果可以看出,利用iPLS进行荧光光谱的特征波段筛选是切实可行的,提取出的特征波段能充分反映出全光谱波段的有效信息,为激光诱导荧光光谱技术用于矿井突水水源精准在线识别的研究提供了理论依据。
Mine water inrush has been threatening the safety of underground construction personnel, so an accurate and rapid identification of mine water inrush source plays a key role in the safe production of the mine. Identification of mine water inrush source by laser induced fluorescence spectroscopy effectively avoids the shortcomings of conventional hydrochemical methods which need to determine a variety of chemical parameters and the identification time is too long. In this paper, a method of interval PLS(iPLS) and particle swarm optimization combined with support vector classification algorithm(PSO-SVC) is proposed. The iPLS algorithm is often used in spectral bands optimization and regression analysis of models, and the PSO-SVC is an important application in the field of machine learning. The laser induced fluorescence spectroscopy(LIF) technology has the characteristics of fast time response and high measurement accuracy, and the iPLS and PSO-SVC algorithms are applied to the analysis of spectral maps and spectral data, and then it can identify and classify water inrush sources. Firstly, The 210 sets of fluorescence spectrum data of 7 kinds(30 groups of each water sample) collected from Huainan mining area were used for experiment, and differences of laser-induced fluorescence spectra of mixed water samples with different volumetric ratios of old-kiln water, limestone water, limestone water and air water were analyzed. The classification accuracy of PSO-SVC model obtained by hold-out and Kennard-Stone partitioning was compared, and the training set water samples(140 groups) and test set water samples(70 groups) obtained by hold-out were used as experimental samples. Secondly, the full spectrum bands were divided into 10～25 bands by using the iPLS algorithm, and the band whose RMSECV(cross validation root mean square error) value is less than RMSECV value(threshold) of full spectrum bands was selected as the characteristic wave bands, and the results of modeling with 10 and 14 sub intervals were compared with spectrogram. It is found that there were errors in the characteristic bands selected by direct observation and the iPLS algorithm. Finally, under the condition of no pretreatment such as denoising and dimension reduction, the statistical data of evaluating indexes for dividing different interval numbers according to iPLS were obtained, and the data of 410.078～478.424 and 545.078～674.104 nm characteristic wave bands with 561 wavelength points selected from 11 regions were used as the input of PSO-SVC model. we compared with full spectrum bands and direct observation bands, and the classification accuracy of the training set and the test set was as high as 100%. The optimal penalty coefficient C of PSO is 1.367, and the kernel function parameter g is 0.576 2. It can be seen from the experimental results that it is feasible to select the characteristic wave bands of the fluorescence spectrum by using iPLS, and the extracted characteristic wave bands can fully reflect the effective information of the full spectrum bands, and it provides a theoretical basis for the application of laser induced fluorescence spectroscopy in the accurate on-line identification of mine water inrush source.
Mine water inrush has been threatening the safety of underground construction personnel, so an accurate and rapid identification of mine water inrush source plays a key role in the safe production of the mine. Identification of mine water inrush source by laser induced fluorescence spectroscopy effectively avoids the shortcomings of conventional hydrochemical methods which need to determine a variety of chemical parameters and the identification time is too long. In this paper, a method of interval PLS(iPLS) and particle swarm optimization combined with support vector classification algorithm(PSO-SVC) is proposed. The iPLS algorithm is often used in spectral bands optimization and regression analysis of models, and the PSO-SVC is an important application in the field of machine learning. The laser induced fluorescence spectroscopy(LIF) technology has the characteristics of fast time response and high measurement accuracy, and the iPLS and PSO-SVC algorithms are applied to the analysis of spectral maps and spectral data, and then it can identify and classify water inrush sources. Firstly, The 210 sets of fluorescence spectrum data of 7 kinds(30 groups of each water sample) collected from Huainan mining area were used for experiment, and differences of laser-induced fluorescence spectra of mixed water samples with different volumetric ratios of old-kiln water, limestone water, limestone water and air water were analyzed. The classification accuracy of PSO-SVC model obtained by hold-out and Kennard-Stone partitioning was compared, and the training set water samples(140 groups) and test set water samples(70 groups) obtained by hold-out were used as experimental samples. Secondly, the full spectrum bands were divided into 10～25 bands by using the iPLS algorithm, and the band whose RMSECV(cross validation root mean square error) value is less than RMSECV value(threshold) of full spectrum bands was selected as the characteristic wave bands, and the results of modeling with 10 and 14 sub intervals were compared with spectrogram. It is found that there were errors in the characteristic bands selected by direct observation and the iPLS algorithm. Finally, under the condition of no pretreatment such as denoising and dimension reduction, the statistical data of evaluating indexes for dividing different interval numbers according to iPLS were obtained, and the data of 410.078～478.424 and 545.078～674.104 nm characteristic wave bands with 561 wavelength points selected from 11 regions were used as the input of PSO-SVC model. we compared with full spectrum bands and direct observation bands, and the classification accuracy of the training set and the test set was as high as 100%. The optimal penalty coefficient C of PSO is 1.367, and the kernel function parameter g is 0.576 2. It can be seen from the experimental results that it is feasible to select the characteristic wave bands of the fluorescence spectrum by using iPLS, and the extracted characteristic wave bands can fully reflect the effective information of the full spectrum bands, and it provides a theoretical basis for the application of laser induced fluorescence spectroscopy in the accurate on-line identification of mine water inrush source.
引文
[1] SUN Yun-jiang,ZUO Jian-ping,LI Yu-bao,et al(孙运江,左建平,李玉宝,等).Rock and Soil Mechanics(岩土力学),2017,38(8):2335.
[2] FENG Zhi-dong,WANG Tao,GU Qing-hua,et al(冯治东,王桃,顾清华,等).Systems Engineering-Theory & Practice(系统工程理论与实践),2017,37(12):3289.
[3] ZHU Le-zhang(朱乐章).China Coal(中国煤炭),2018,44(5):100.
[4] LI Shao-dan,LIN Yuan-sheng,TAN Si-chao,et al(李少丹,林原胜,谭思超,等).Nuclear Power Engineering(核动力工程),2014,35(4):137.
[5] WANG Ya,ZHOU Meng-ran,CHEN Rui-yun,et al(王亚,周孟然,陈瑞云,等).Acta Optica Sinica(光学学报),2018,38(7):367.
[6] GUO Hui-xian,ZHU Si-qi,LI Yuan-peng,et al(郭慧娴,朱思祁,黎远鹏,等).Journal of Optoelectronics·Laser(光电子·激光),2016,27(10):1136.
[7] YANG Peng-cheng,DU Jun-lan,CHENG Chang-kuo(杨鹏程,杜军兰,程长阔).Marine Environmental Science(海洋环境科学),2016,35(6):943.
[8] ZHAO Fang,PENG Yan-kun(赵芳,彭彦昆).Chinese Journal of Lasers(中国激光),2017,44(11):249.
[9] Giraudo A,Calvini R,Orlandi G,et al.Food Control,2018,94:233.
[10] XU Long,LU Jian-gang,YANG Qin-min,et al(徐龙,卢建刚,杨秦敏,等).CIESC Journal(化工学报),2013,64(12):4410.
[11] SHI Long-qing,TAN Xi-peng,WANG Juan,et al(施龙青,谭希鹏,王娟,等).Journal of China Coal Society(煤炭学报),2015,40(1):167.
[12] QI Shao-hua,LIU Qi-yuan,CHEN Jiu-hui,et al(齐少华,刘启元,陈九辉,等).Chinese Journal of Geophysics(地球物理学报),2016,59(12):4544.
[13] CHEN Yi-yun,ZHAO Rui-ying,QI Tian-ci,et al(陈奕云,赵瑞瑛,齐天赐,等).Spectroscopy and Spectral Analysis(光谱学与光谱分析),2017,37(7):2133.
[2] FENG Zhi-dong,WANG Tao,GU Qing-hua,et al(冯治东,王桃,顾清华,等).Systems Engineering-Theory & Practice(系统工程理论与实践),2017,37(12):3289.
[3] ZHU Le-zhang(朱乐章).China Coal(中国煤炭),2018,44(5):100.
[4] LI Shao-dan,LIN Yuan-sheng,TAN Si-chao,et al(李少丹,林原胜,谭思超,等).Nuclear Power Engineering(核动力工程),2014,35(4):137.
[5] WANG Ya,ZHOU Meng-ran,CHEN Rui-yun,et al(王亚,周孟然,陈瑞云,等).Acta Optica Sinica(光学学报),2018,38(7):367.
[6] GUO Hui-xian,ZHU Si-qi,LI Yuan-peng,et al(郭慧娴,朱思祁,黎远鹏,等).Journal of Optoelectronics·Laser(光电子·激光),2016,27(10):1136.
[7] YANG Peng-cheng,DU Jun-lan,CHENG Chang-kuo(杨鹏程,杜军兰,程长阔).Marine Environmental Science(海洋环境科学),2016,35(6):943.
[8] ZHAO Fang,PENG Yan-kun(赵芳,彭彦昆).Chinese Journal of Lasers(中国激光),2017,44(11):249.
[9] Giraudo A,Calvini R,Orlandi G,et al.Food Control,2018,94:233.
[10] XU Long,LU Jian-gang,YANG Qin-min,et al(徐龙,卢建刚,杨秦敏,等).CIESC Journal(化工学报),2013,64(12):4410.
[11] SHI Long-qing,TAN Xi-peng,WANG Juan,et al(施龙青,谭希鹏,王娟,等).Journal of China Coal Society(煤炭学报),2015,40(1):167.
[12] QI Shao-hua,LIU Qi-yuan,CHEN Jiu-hui,et al(齐少华,刘启元,陈九辉,等).Chinese Journal of Geophysics(地球物理学报),2016,59(12):4544.
[13] CHEN Yi-yun,ZHAO Rui-ying,QI Tian-ci,et al(陈奕云,赵瑞瑛,齐天赐,等).Spectroscopy and Spectral Analysis(光谱学与光谱分析),2017,37(7):2133.
© 2004-2018 中国地质图书馆版权所有 京ICP备05064691号 京公网安备11010802017129号 地址：北京市海淀区学院路29号 邮编：100083 电话：办公室：(+86 10)66554848；文献借阅、咨询服务、科技查新：66554700 |