The rapid and accurate coupling interference analysis and concentration detection of the multiple pollutants in complex water bodies are significantly important for the in-situ real-time monitoring of field water quality. To address the problems of characteristic coupling and the interference of spectral peaks in the synchronous detection of chemical oxygen demand (COD) and turbidity using ultraviolet spectroscopy, which significantly affect the detection accuracy, a decoupling method for predicting water pollutant concentration was developed in this study based on the continuous projection algorithm combined with support vector regression. The continuous projection algorithm was used to screen the characteristic wavelengths of the ultraviolet absorption spectra of water quality samples and eliminate irrelevant redundant numbers, to improve the iteration rate and accuracy of the model. Based on the concept of the multi-classification support vector machine, the support vector regression algorithm was improved via multi-regression fitting, and the ultraviolet coupling analysis of COD and turbidity, as well as the simultaneous prediction of concentration, was realized. The test results for actual water samples reveal that the maximum relative errors are reduced to less than 4%, and the improvement rate of the root mean square error of predictions before the coupling analysis reaches 76%. Thus, the proposed method offers a better detection accuracy, as compared with similar methods. Notably, this work is expected to serve as a reference for the application of ultraviolet spectroscopy in water-quality multi-coupling parameter detection.