A COD measurement method with turbidity compensation based on a variable radial basis function neural network

Abstract

In recent years, ultraviolet-visible spectrometry has been widely used to measure sewage's chemical oxygen demand (COD). However, most methods that use UV-vis spectroscopy for COD measurement have not eliminated the influence of turbidity. Therefore, this article proposes a new COD measurement method using UV-vis spectroscopy. This method includes a new turbidity compensation algorithm and an algorithm for COD measurement using a variable radial basis function (VRBF) neural network. Our turbidity compensation algorithm first utilizes principal component analysis (PCA) to extract the characteristic wavelengths of the spectrum. Then, turbidity is used to fit the absorbance difference caused by turbidity at the characteristic wavelength, and a turbidity compensation model is obtained. The turbidity compensation model is used to remove the influence of turbidity from the absorbance spectrum, thereby compensating for its effect on the COD measurement. Secondly, the VRBF neural network model is used to measure the COD concentration. The results show that, compared with the traditional partial least squares regression model, the R² coefficient of determination increases from 0.27 to 0.88, and the root-mean-square deviation decreases from 5.56 to 1.69. Compared with the improved bagging algorithm and MLP algorithm, this method can measure COD concentration from absorption spectra faster, more directly, and more accurately.