[1]杨曼孜,王晓东,刘长青,等.基于紫外-可见光谱的COD预测模型优化方案研究[J].中国给水排水,2022,38(21):113-119.
　YANGMan-zi,WANGXiao-dong,LIUChang-qing,et al.Optimization of COD Prediction Model Based on Ultraviolet-visible Spectrum[J].China Water & Wastewater,2022,38(21):113-119.
点击复制

基于紫外-可见光谱的COD预测模型优化方案研究

中国给水排水[ISSN:1000-4062/CN:12-1073/TU] 卷: 第38卷 期数: 2022年第21期 页码: 113-119 栏目: 出版日期: 2022-11-01
Title:
Optimization of COD Prediction Model Based on Ultraviolet-visible Spectrum
作者:
杨曼孜， 王晓东， 刘长青， 塔哈·马哈巴
（青岛理工大学 环境与市政工程学院，山东 青岛 266000）
Author(s):
YANG　Man-zi， WANG　Xiao-dong， LIU　Chang-qing， TAHA　Marhaba
（School of Environmental and Municipal Engineering， Qingdao University of Technology， Qingdao 266000， China）
关键词:
紫外-可见光谱;  全光谱分析;  化学需氧量;  主成分回归;  偏最小二乘回归;  模型优化
Keywords:
ultraviolet-visible spectrum;  full-spectrum analysis;  chemical oxygen demand;  principal component regression;  partial least squares regression;  model optimization
摘要:
水质在线监测的精确性、实时性和可靠性是制约污水处理厂提高自动化水平和实现智能化管理的瓶颈问题。将光谱仪置于青岛某污水处理厂获取污水的紫外-可见（UV-Vis）全光谱（200～737.5 nm）数据，同时在实验室化验分析对应样品的COD浓度，基于主成分回归法（PCR法）和偏最小二乘回归法（PLSR法）构建COD预测模型，并采用两次回归分析的思想选取特征波长重新建模，深入讨论采用不同方法、不同类型波长建模对提高模型预测能力的影响。结果表明：与PCR法相比，采用PLSR法构建的模型具有更优的预测能力；基于PLSR法，选取COD特征吸收波长以及硝酸盐和浊度特征吸收波段的波长用于校准，构建的COD模型获得了最优的预测能力，模型拟合优度（R2）为0.992，预测均方根误差（RMESP）为1.979 mg/L。因此，采用COD特征吸收波长与非预测目标特征吸收波长组合建模的方式能够进一步提高模型的预测精度。
Abstract:
The accuracy, real?time and reliability of on-line water quality monitoring are the bottlenecks that restrict the improvement of automation level and the realization of intelligent management of wastewater treatment plants. The spectrometer was placed in a wastewater treatment plant in Qingdao to obtain the UV-Vis full-spectrum (200-737.5 nm) data of sewage. Meanwhile, the COD of corresponding samples was analyzed in the laboratory. COD prediction models were constructed based on principal component regression (PCR) method and partial least squares regression (PLSR) method, characteristic wavelengths were selected to reconstruct the model by using the idea of twice regression analysis, and the influence of different methods and different types of wavelength on improving the prediction ability of the model was discussed in detail. Compared with PCR method, the model constructed by PLSR method had better predictive ability. Based on PLSR method, the characteristic absorption wavelengths of COD and the characteristic adsorption wavelengths of nitrate and turbidity were selected for calibration, and the COD model obtained the optimal prediction ability. The goodness of fit (R2) of the model was 0.992, and the root mean square error of prediction (RMESP) was 1.979 mg/L. Therefore, the combined modeling method of COD characteristic absorption wavelength and non-predicted target characteristic wavelength could further improve the prediction accuracy of the model.
更新日期/Last Update: 2022-11-01