基于人工神经网络的透射水体亚硝酸盐含量模拟估计  

Simulation Estimation of Nitrite Content in Transmission Water Based on Artificial Neural Network

作　　者：王彩玲[1] 张国浩 闫晶晶 WANG Cailing;ZHANG Guohao;YAN Jingjing(College of Computer Science,Xi'an Shiyou University,Xi'an,Shaanxi 710065,China)

机构地区：[1]西安石油大学计算机学院,西安710065

出　　处：《中国无机分析化学》2024年第7期857-865,共9页Chinese Journal of Inorganic Analytical Chemistry

基　　金：陕西省重点研发计划项目(2023-YBSF-437);国家自然科学基金资助项目(31160475,61401439)。

摘　　要：亚硝酸盐是水体的重要测试指标,对水体质量的评估有着重要意义。采用透射高光谱结合人工神经网络(ANN)建立水体亚硝酸盐含量估算模型。首先采用试剂配制10种浓度的亚硝酸氮标准溶液(0.02、0.04、0.06、0.08、0.10、0.12、0.14、0.16、0.18和0.20 mg/L),并使用OCEAN-HDX-XR微型光纤光谱仪扫描10次各浓度亚硝酸盐溶液在181.1~1023.1 nm的透射光谱,取平均值作为各浓度亚硝酸盐溶液原始透射光谱,分别使用最大最小均一化(MMN)、标准正态变化(SNV)、多元散射校正(MSC)、以及二阶差分(SOD)四种光谱预处理方法,并结合ANN方法建立水体亚硝酸盐含量估算模型,通过比较模型的精度来选择最优的模型进行水体亚硝酸盐含量的估计。结果显示,基于二阶差分预处理下的BP-ANN神经网络预测模型中的均方根误差RMSE为0.032367,平均绝对误差MAE为0.016895,决定系数R^(2)为0.987403,与二次有理高斯过程回归(QR-GPR)和二次支持向量机(Q-SVM)预测模型相比,该模型的拟合效果更好,精确度更高。提出了反向传播人工神经网络(BP-ANN)高光谱水质亚硝酸盐参数的反演方法,为水质亚硝酸盐参数动态检测提供了新方法。Nitrite is an important test index for water bodies and has great significance for the evaluation of water quality.In this paper,transmission hyperspectral combined with artificial neural network(ANN)was used to establish a water nitrite content estimation model.In this experiment,10 concentrations of nitrogen nitrite standard solution(0.02,0.04,0.06,0.08,0.10,0.12,0.14,0.16,0.18 and 0.20 mg/L)were prepared by reagents,and the transmission spectrum of each concentration of nitrite solution in the range of 181.1—1023.1 nm were scanned 10 times by OCEAN-HDX-XR microfiber spectrometer,and the average value was taken as the original transmission spectrum of each concentration of nitrite solution.Four spectral preprocessing methods of maximum and minimum uniformization(MMN),standard normal change(SNV),multivariate scattering correction(MSC),and second-order differential(SOD)were used respectively,and the water nitrite content estimation model was established by combining the ANN method,and the optimal model was selected to estimate the nitrite content of water by comparing the accuracy of the model.The results showed that the RMSE was 0.032367,the MAE was 0.016895 and the R^(2) was 0.987403 in the BP-ANN neural network prediction model based on second-order differential preprocessing,which had better fitting effect and higher accuracy than the quadratic rational Gaussian process regression(QR-GPR)and quadratic support vector machine(Q-SVM)prediction model.Based on the above experimental results,an inversion method combining ANN hyperspectral nitrite parameters was proposed,which provided a new method for the dynamic detection of nitrite parameters in water quality.

关 键 词：高光谱 人工神经网络 亚硝酸盐 数据预处理 估算模型 

分 类 号：O657.31[理学—分析化学] P237[理学—化学]