A hybrid two-layer characteristic wavelength selection for COD prediction based on UV-Vis

Chemical Oxygen Demand (COD) is a critical indicator for assessing water quality. The UV-visible spectroscopy method for detecting COD has advantages such as being non-polluting, having a short cycle, and requiring fewer consumables. However, because its absorption spectrum data is high-dimensional and highly collinear, the choice of models and characteristic wavelengths directly affects the prediction results. Moreover, the turbidity caused by suspended particles also limits detection accuracy. To address these issues, this paper proposes a novel characteristic wavelength selection method for absorption spectra, constructing a two-layer model based on the Artificial Fish Swarm Algorithm (AFSA) and Elastic Net (EN) by combining wrapper and embedded methods. AFSA acts as a wrapper to explore the optimal predictor subset, and then an embedded method is applied using EN to eliminate irrelevant predictors in the subset. Based on the extracted characteristic wavelengths, turbidity compensation is performed on the original spectrum according to Lambert-Beer’s law. The compensated absorbance values are then input into a Multilayer Perceptron to fit the COD. The results show that the proposed AFSA-EN model outperforms other characteristic wavelength extraction models. The wavelengths extracted by AFSA-EN achieve the best turbidity compensation for the full spectrum. The final COD prediction results show an increase in R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> from 0.4656 to 0.9172, with reductions in both RMSE and MAE.