Application of Gauss process regression modeling based on NN-MIV for marine enzyme fermentation process

Abstract

To overcome the problems of variable redundancy, long training time and low prediction accuracy in the soft sensing model for marine enzyme fermentation, a Gauss process regression (GPR) model based on NN-MIV is presented, which is named as GPR-NNMIV soft sensing model. Firstly, the NN-MIV variable selection method, combining neural network (NN) and mean impact value (MIV), takes into account both the internal contribution rate and the external contribution rate to get the most suitable input variables with the highest contribution rate, and reduces the number of variables and simplifies soft sensing model. Secondly, based on the NN-MIV method, a new Gauss process regression model is proposed, which does not only give out the soft sensing results but also gives the corresponding uncertainty simultaneously. Results show that the proposed GPR-NNMIV soft sensing model has higher accuracy of results and small confidence intervals compared with single Gauss process model.