Study On Prediction Models Of Determination Of Chemical Oxygen Demand (COD) In Wastewater By Near-Infrared Spectroscopy Based On Matlab

The near-infrared(NIR) spectrum analysis technique is a high-tech, which is developed fast in recent years. The analysis speed is fast. It doesn't destroy the samples. The operation is simple. The stability is good. It is high efficient. It synthesized the physics, chemomeritics, mathematics and computer science, etc.It belongs to the cross techniques. All materials with -H group will have the absorption in the near-infrared spectral region. It contains vast amount of information. It concentrates the information of overtone area and combinations area. Large amount of information feature peak is overlapped seriously. There is no obvious peak. So it belongs to the weak spectrum information. Using the general spectrum analysis means , we can't sort out available information from the spectrum. So we should use the way of computer and the chemomeritics.According to the experiment data, we found the correlation between the pollution degree of some organic contaminations (-H group---OH, CH, NH, SH, PH) and the near-infrared spectrum, and formed their quantative model. All these made a preparation for the further study of the near-infrared spectrum transducer which can measure organic contaminations on-line, and the research of on-line system which inspect the water quality COD's real-time situation.The dissertation emphasized on studying the predicting model of the near-infrared spectrum of water quality COD. We used multiple variable analysis method to form the model. Such as Stepwise linear regression (SMLR), Pincipal component regression(PCR) and Partial least squares(PLS).The best result of stepwise regression is 15 points smooth.The determination coefficient (R~2)is 95.01 pecent, the number of best regression factor is 6, Standard Error Calibration(SEC) is 19.03 mg/l, Standard Error Prediction(SEP) is 18.72mg/l.The best result of PCR is the first derivative mean. When the contribution is 98 percent , the determination coefficient (R~2) is 97.69 percent, the number of principal component is 6, the SEC is 12.26 mg/1, the SEP is 18.72 mg/l. The best result of PLS is baseline covariation mean. The determination coefficient (R~2) is 99.42 percent, the number of factors is 5, the SEC is 8.93 mg/1, the SEP is 7.56 mg/l.Comparing the results of three ways of the multiple variable analysis method, we found the way of PLS is best, and then it is the way of PCR and SMLR. We can try to use as many methods as possible to find a best way of predicting mean.The dissertation also exploit the way of mixing programme of VB and MATLAB.It is the software system of engineering calculation. Users can use them conveniently to make programme by using the compute function of MATLAB and the convenient interface. The research has practical value.