Proper control of distillation columns requires estimating some key variables that are challenging to measure online (such as\r\ncompositions), which are usually estimated using inferential models. Commonly used inferential models include latent variable\r\nregression (LVR) techniques, such as principal component regression (PCR), partial least squares (PLS), and regularized canonical\r\ncorrelation analysis (RCCA). Unfortunately, measured practical data are usually contaminated with errors, which degrade the\r\nprediction abilities of inferential models. Therefore, noisy measurements need to be filtered to enhance the prediction accuracy of\r\nthesemodels.Multiscale filtering has been shown to be a powerful feature extraction tool. In this work, the advantages ofmultiscale\r\nfiltering are utilized to enhance the prediction accuracy of LVR models by developing an integrated multiscale LVR (IMSLVR)\r\nmodeling algorithmthat integratesmodeling and feature extraction.The idea behind the IMSLVRmodeling algorithmis to filter the\r\nprocess data at different decomposition levels, model the filtered data fromeach level, and then select the LVRmodel that optimizes\r\na model selection criterion. The performance of the developed IMSLVR algorithm is illustrated using three examples, one using\r\nsynthetic data, one using simulated distillation column data, and one using experimental packed bed distillation column data. All\r\nexamples clearly demonstrate the effectiveness of the IMSLVR algorithm over the conventional methods.
Loading....