A COD measurement method with turbidity compensation based on Multilayer perceptron

Chemical Oxygen Demand (COD) in water can be measured using ultraviolet-visible (UV-Vis) spectroscopy, an optical measurement method solely based on physical principles. It has benefits over conventional chemical detection techniques, including shorter testing periods, less waste, and no pollution. This research suggests a turbidity-compensated COD prediction model based on Partial Least Squares (PLS) and Multilayer Perceptron (MLP) to address these problems. First, spectral data is processed using PCA to extract characteristic wavelengths. Second, by fitting the absorbance changes due to turbidity at the characteristic wavelengths, a turbidity compensation model is derived based on PLS. Last, an MLP model is applied to the turbidity-compensated data to estimate COD. The results show that compared with the traditional models, such as Support Vector Machine and Random Forest, the mean square error of the proposed model for COD prediction results is reduced from 13.7 to 4.5, and the determination coefficient is increased from 0.21 to 0.81, which significantly improves the accuracy of the COD prediction model and provides a new idea for the detection of COD concentration in water quality.