Improved boosting and self-attention RBF networks for COD prediction based on UV-vis

Abstract

Chemical Oxygen Demand (COD) is crucial for assessing water quality. Compared to traditional chemical detection methods, UV-vis spectroscopy for measuring COD offers advantages such as speed, reduced consumption of materials, and no secondary pollution. Considering the impact of suspended particles in water, this paper proposes an optimized boosting model based on a combination strategy for turbidity compensation, using absorption spectra obtained from reservoir water samples via UV-vis. A self-attention mechanism is introduced into the radial basis function (RBF) network, resulting in a COD detection model based on the saRBF framework. This model facilitates comprehensive optimization of the entire process, from turbidity compensation of the original absorption spectrum to the subsequent COD prediction. Experimental results show that the proposed COD measurement model achieves a coefficient of determination of 0.9267, a root mean square error of 1.2669, and a mean absolute error of 1.0097, outperforming other COD measurement models. This work provides a new approach for turbidity compensation and COD detection research.