Water Quality pH Value Determination for Visible-Near Infrared Spectroscopy Based on SPA and PSO-LSSVM

To improve the detecting precision and robustness in the determination of water pH value using visible near-infrared (Vis-NIR) spectroscopy, a multivariate calibration model is constructed based on successive projections algorithm (SPA) and particle swarm optimization-least squares support vector machine (PSO-LSSVM). The Vis-NIR spectra data of 60 water samples with different pH values are collected, and the original spectral data are preprocessed by Savitzky-Golay smoothing and standard normal variate. Based on the characteristic wavelength of SPA screening and PSO algorithm the modeling parameters of LSSVM are automatically optimized and a multivariate nonlinear calibration model is established. The results show that the SPA-PSO-LSSVM model has higher accuracy and stability than the comparison models. For the verification set, the root mean square error is 0.67, the coefficient of determination is 0.91, and the residual predictive deviation is 3.10.