The Study On Near-Infrared Spectroscopy Quantitative Analysis Model For The Chemical Oxygen Demand (COD) Of Wastewater

The discharge of industrial wastewater from sugar plant will cause serious environmental pollution. The chemical oxygen demand(COD) is a general indicator to estimate the quality of the water. Howerver, nowadays conventional detection methods of wastewater need to use a lot of corrosive chemicals reagent, which is not only high-cost, time-consuming, difficult to operate, but also cause secondary pollution. It is important to establish a newly direct and rapid quantitative determination method of wastewater COD from sugar industry without any chemical reagent. This paper focused on the relative mathematical methods.50 different divisions of calibration set and prediction set were made to select the stable prediction result and model parameters. The optimal parameters of model were selected according to the average effect of 50 different divisions of calibration set and prediction set. This paper adopts Savitzky-Golay smoothing, and method to establish prediction models on different waveband. Using root mean squared error of prediction (RMSEP) and correlation coefficient of predication (RP) as model estimate indicators to select suitable wavebands to measure COD of wastewater. Then MWPLS were applied to select the waveband which is optimal to measure the COD, which is high signal to noise ratio.It indicated that the model established on the short-wave near-infrared (780-1100nm) has good predictive effect, the model established on the short-wave near-infrared (780-1100nm) with the derivative spectra of the order of 4 combined with smoothing modes(5 polynomial,5 smoothing points) has the best result, whose predictive RMSEPAve, RP,Ave were 25.4(mg/L), and 0.969, the corresponding band for the model was 810-1096nm, the number of adopted points was 144, the model established on the waveband 810-872nm with the derivative spectra of the order of 2 combined with smoothing modes(6 polynomial,7 smoothing points) has good result, whose predictive RMSEPAve, RP,Ave were 28.7(mg/L), and 0.959. The selection of wavebands 810-872nm with high signal to noise ratio can be made to design the spectroscopy system of Near-Infrared spectroscopy.