Study On COD Predicted Model Of Water Quality Based On Multi-spectral Information Fusion

The high concentration of organic matter content in water body is harmful to human health as well as serious environmental pollution.The traditional chemical method for the determination of the chemical oxygen demand in water is cumbersome and time-consuming,which is not conducive to the rapid quantitative detection of COD in water.Spectrometric determination of COD has the advantages of real-time,on-line and continuous monitoring.It can reflect the dynamic situation of organic matter content in water in time.It is a real green detection technology.Based on the theory of ultraviolet spectroscopy,near infrared spectroscopy and water quality COD spectroscopy detection technology,combined weight algorithm and generative adversarial networks algorithm are introduced to construct the combined weight COD concentration prediction model of single spectral source and the COD concentration prediction model of multi-spectral data fusion GANs algorithm,and the prediction accuracy and generalization ability are compared and analyzed.First of all,based on the study of the micro-mechanism of ultraviolet and near infrared spectroscopy,the relationship between spectral absorbance and COD concentration is analyzed.The spectral pretreatment algorithm,characteristic spectral interval screening algorithm,quantitative modeling algorithm and model evaluation parameters in the process of building COD quantitative analysis model of water quality are introduced,which provides theoretical basis for the construction of COD quantitative analysis model.Secondly,a fast quantitative COD detection method is proposed,which combines the ultraviolet spectrum of a single spectral source with the combined weights algorithm.The two-stage screening mechanism of backward interval PLS and synergy interval PLS is used to screen the feature intervals,and the weights formula is used to screen the feature intervals.Subinterval weights are calculated,and the final COD concentration prediction model of combined weights is established,and the prediction accuracy of combined weights model is analyzed.Finally,a COD concentration prediction model based on multi-spectral data fusion is proposed.The data level fusion and feature level fusion of ultraviolet and near infrared spectra can cover a wider range of spectral intervals and reflect the difference of pollutants in water from two levels of ultraviolet and near infrared spectroscopy.The influence of different normalization methods on spectral data fusion modeling is discussed.By comparing the performance of the multi-spectral model with that of the single spectral source prediction model,it is verified that the multi-spectral model can reveal the pollutant level in water comprehensively,improve the prediction accuracy and generalization ability,and have higher research value.