Study on a fast non-contact detection method for key parameters of refractory organic wastewater treatment

In this paper, a fast non-contact detection method is proposed to test the key parameters of refractory organic wastewater in the treatment system intelligently and efficiently. The proposed method combines different preprocessing to the hyperspectral data sampled in the Polyvinyl Alcohol (PVA)-sizing wastewater with our independent-designed equipment and optimizes the models of neural network to predict the pH, COD and PVA of wastewater detection. The data preprocessing contains original spectrum (R), envelope removal (CR), original spectrum continuous projection (R-SPA) and envelope removal continuous projection (CR-SPA). The BP, RBF, SVR and RF networks are trained with the preprocessed hyperspectral data to obtain the optimal inversion prediction models. When the wavelength of the hyperspectral data for discipline ranges from 400 nm to 1700 nm, the trained RF network gets the optimal prediction model of pH after the data preprocessing by CR and its Root Mean Square Error (RMSE) is 0.0424. The RBF network trained by the data pretreated with R-SPA gets the optimal prediction model of COD and its RMSE is 40.4489. The RF network trained with the data preprocessed with the R-SPA gets the optimal PVA index prediction model and its RMSE is 1.5841. The results show that the proposed method can realize the accurate prediction for the pH, COD and PVA values of PVA-sizing wastewater and accomplish it smartly and fast.