“…To explore and utilize useful information hidden in the data, several empirical soft-sensing approaches were applied to online predict the silicon content. Existing popular data-driven methods include artificial neural networks [ 7 , 8 , 9 , 10 , 11 , 12 , 13 , 14 ], multivariate regression [ 14 , 15 ], time series analysis [ 16 , 17 , 18 , 19 ], fuzzy systems [ 20 ], subspace identification [ 21 ], support vector regression (SVR) and least squares SVR (LSSVR) [ 22 , 23 , 24 , 25 ], and multi-scale and multiple models [ 26 , 27 , 28 , 29 , 30 ]. These data-driven empirical models for short-term silicon content prediction can be constructed in a quick way [ 31 , 32 , 33 ].…”