LIF技术与PLS-DA算法联合辨识矿井涌水水源类型的研究

摘要

Rapid source identification of mine water inrush has great significance for early warning and rescuing after the mine water inrush.Conventional method taking the concentration of ions as the discriminant factor takes such a long time that a meth-od of rapid source identification of mine water inrush is in urgent need.This method is combined with Laser induced fluorescence (LIF)technology and Partial Least Squares-Discriminant Analysis (PLS-DA)algorithm.In the experiment,405 nm laser was used to excite the water and 100 groups of fluorescence spectrum from 5 different aquifer of the mine were obtained.According to the spectra curve features,the data was compressed to obtain proper spectral data.1 5 groups of spectrum of each water inrush samples were applied,with a total of 75 groups of spectrum as the prediction set while the rest of 25 groups of spectrum as the test set.To verify the experimental result,an experimental model was built with soft independent modeling of class analogy (SIMCA)to compare with PLS-DA.The result shows that the fluorescence spectra of different aquifer water samples is of great difference,without any pre-treatment,the PLS-DA algorithm based on the PLS model has higher modeling accuracy compared with SIMCA algorithm,which reaches to 100%,the validation results and the correlation of separation of variables are both more than 0.951,the RMSECV and RMSEP are both less than 0.123,using the model to identify the 5 water samples of test set,the accuracy are up to 100%.%快速的矿井涌水水源辨识对于矿井的水灾预警及灾后救援意义重大。常规方法使用离子浓度做为判别因子,耗时过长,因此提出一种激光诱导荧光光谱(LIF)技术与偏最小二乘判别分析(PLS-DA)算法联合快速辨识矿井涌水水源类型的方法。实验使用405 nm激光对被测水体进行激发,获取矿井5个不同含水层100组水样的荧光光谱,根据光谱曲线特征,对数据进行压缩处理,获取合适的光谱数据。每种水样使用15组共75组光谱数据作为建模集,剩余的25组水样的光谱数据作为测试集。为验证实验结果,设计了簇类的独立软模式(SIMCA)算法与 PLS-DA算法构建的实验模型进行对比。实验发现矿井不同含水层水样的荧光光谱差异较大,在不进行任何预处理的情况下,以PLS模型为基础的PLS-DA算法较SIMCA算法的建模正确率高,达到了100%,其校正及验证结果与实际分类变量的相关系数均大于0.951,校正集均方根误差(RMSECV)和验证集均方根误差(RMSEP)均小于0.123,利用模型对测试集中五种水样样本的识别正确率均为100%。