Application and comparison of several modeling methods in spectral based water quality analysis

Detection of the organic pollutants in wastewater based on spectroscopy is important for water environment protection. And it is significant for researchers to improve the prediction precision of water spectral analysis model. Based on the ultraviolet absorption spectrum and the fluorescence spectrum, four common approaches in spectral analysis, such as Least Squares, Principal Component Regression, Partial Least Squares and Least Squares Support Vector Machine, are adopted in the modeling of two criterions for water quality evaluation. Comparisons are conducted by computing the root mean square error of prediction values and the relativity of prediction errors. Simulation results show that the prediction models for TOC are superior to those for COD, and the method of LSSVM has the best prediction precision compared with other three linear modeling methods.