Study On The Methods Of The Measurement Of COD In Water Using UV-VIS Spectroscopy Method

In recent years, frequent water pollution incidents affected the living and socio-economic healthy development seriously, so water quality testing(monitoring) has become urgent. Chemical oxygen demand(COD) is one of the most important parameters to measure water quality, as well as one item must be measured of water quality testing. For reagents easily lead to secondary pollution, long cycle and other shortcomings, the traditional chemical testing can’t be suitable for real-time water quality testing. So, UV-Vis spectroscopy that can test COD online, in-situ catches the world’s attention and has good prospects for development.This paper makes Sichuan Province Science and Technology Support Program " real-time automatic monitoring of water quality multifunction technology development and product development"(2012SZ0111) and Sichuan Belam Technology Co. horizontal cooperation project "spectrum multi-parameter water line automatic detector development "(contract Number: 1042012920140453) jointly funded project as basis, and makes UV- visible spectroscopy study of COD technical as background to carry out the research of UV- visible spectroscopy water COD detection. The main work of this paper is:① Wavelet de-noising algorithms study of the raw spectral data of water quality testing. There are advantages of wavelet de-noising algorithm in locality of time domain and frequency domain, and singularity and mutation structure of detection signal, additional, other characteristics such as the processing of signal of non-stationary and signal including broadband noise is more advantageous. As to the high-frequency noise interference problem with the raw UV- visible absorption spectroscopy data, we decompose noisy spectrum signal, then select improved threshold function to filter out low-frequency coefficients, and recover the signal through the inverse wavelet transform. The results show that wavelet de-noising algorithm can effectively suppress noise.② Pretreatment of dimension reduction for the spectral data of water quality testing based on principal component analysis(PCA). There is a serious nonlinear overlap, more samples and high dimension in water quality spectral data. Additional, there is a large number of redundant, hidden correlations of important relationships in high-dimensional data. In this paper, using PCA algorithm, which is simple, theoretical improve and so on, to complete pretreatment of dimension reduction for the spectral data before building prediction model by extracting effective feature information and reducing the dimension of the input variables, in order to improve the efficiency of machine learning.③ Prediction model study of water COD based on least squares support vector machine(LS-SVM) algorithm. Whether UV- visible online COD detector accurately determine the value of water COD, the key is how to build a mathematical model between the water spectral data with the value of water COD. By comparison with the existing software measurement method, combining features of spectral data, we choose LS-SVM, which need less samples, less initial parameter setting, good extrapolation performance and so on, to build COD prediction model. The results show that COD prediction model base on LS-SVM is suitable for spectral information processing of the principle type prototype. To improve predicted accuracy of LS-SVM prediction model, using PSO algorithm to optimize the LS-SVM model parameters, and obtaining PSO-LS-SVM prediction model. Comparing predicted results of water COD of the PCA-PSO-LS-SVM,LS-SVM and PSO-LS-SVM prediction model with the evaluation criteria of average relative error(MRE) and root mean square error(RMSE),which show that PCA-PSO-LS-SVM prediction model is better with its high accuracy and fast convergence.④ Mixing Programming Techniques study of Lab VIEW and C language. Using the standard C language to write source code of the above algorithm, and generating dynamic link library files(DLL) that called by LABVIEW, which abandon the weak of data processing function of Lab VIEW software and the low operating efficiency and poor portability of Matlab.