Study On The Detection Method Of Nitrate Nitrogen Concentration In Water Based On Ultraviolet Absorption Spectrum

In recent years,the problem of water pollution is serious increasingly.As the second major pollutant in industrial wastewater,nitrate nitrogen is an important index to measure the toxicity and eutrophication degree of water.The traditional detection of nitrate nitrogen?NO₃^--N?in water quality is complicated,time-consuming,costly and prone to secondary pollution,which makes it difficult to meet the requirements of water quality monitoring.In this day of water shortage,improving water quality monitoring is an important part of ensuring water security.Therefore,this paper conducts the following research work on the detection of nitrate nitrogen in water quality by ultraviolet absorption spectrum.In this paper,an ultraviolet spectrum water quality detection system is proposed,and the nitrate nitrogen standard solution preparation,spectrum collection,the rational partition of training set and prediction set are made.In order to effectively improve the prediction performance of the model,The linear prediction models of ultraviolet full-spectrum method are established by using six pretreatment methods,including smoothing,standard normal transformation,detrend and others,combined with Partial Least Squares Regression?PLSR?and Multiple Linear Regression?MLR?.The experimental results show that the prediction error of the linear models is large for low-concentration samples,and the prediction performance of PLSR is better than that of MLR model.Due to the complex nonlinearity of the spectral data and concentration of nitrate nitrogen,The nonlinear prediction models of BP neural network and RBF neural network are established based on the principal component extracted from PLS.The experimental results show that PLS dimension reduction pretreatment can effectively improve the prediction accuracy of the model.Compared with the PLS-BPNN model,the PLS-RBFNN model has stabler performance,fewer adjustable parameters and smaller errors,and is more suitable for nonlinear modeling of nitrate nitrogen in water quality.The critical concentration value of 3mg/L is screened out by the spectral area integral,and then the establishment of composite model is built that prediction of simple linear regression model between the low concentration sample concentration and spectral integral area and PLS-RBF neural network nonlinear prediction model based on high concentration samples,whose result is compared with PLSR model and PLS-RBFNN model.The result shows that the composite model has the best performance,the root mean square error?RMSE?between the predicted value and the true value is 0.3831,the average absolute percentage error?MAPE?is 0.99%,the average absolute error?MAE?is 0.2967,the three parameters are the smallest in the models,and the relative error is small as a whole.Compared with the single model,the combined model can predict the concentration of nitrate nitrogen more accurately with the upper predicted concentration of several hundred mg/L.Besides,it has certain applicability in the detection of the other compoent nitrite nitrogen among three nitrogen.In view of the spectral similarity between nitrate nitrogen and nitrite nitrogen,a binary classification model is established with PLS-ELM,and the recognition rate is up to 100% to some extent.