Support Vector Regression based Spectral Calibration Model Building for In-situ Process Measurement via Near-infrared Spectroscopy

For in-situ measurement of biochemical process variables such as the multi-component concentrations of a fermentation process via near-infrared spectroscopy, an improved spectral calibration model building method is proposed in this paper, based on the support vector regression (SVR) in combination with a grey wolf optimizer for parameter setting. To address the nonlinear properties and high-dimensional problem of infrared spectral variables, the Gaussian kernel function is adopted to establish an SVR based spectral calibration model so as to improve fitting accuracy with respect to the training dataset. Moreover, a grey wolf optimization algorithm is given to determine the model parameters for obtaining good prediction performance. The application to a fluidized bed drying (FBD) process for prediction of moisture content and another lactic acid fermentation process for multi-component prediction are performed, well demonstrating the effectiveness and superiority of the proposed modelling method over the traditional partial least-squares modeling algorithms.