Soft sensing of non-Gaussian processes using ensemble modified independent component regression

Abstract

The modified independent component analysis (MICA) has been proposed to tackle some shortcomings which existed in the original ICA iterative procedures and has found wide applications in non-Gaussian data modeling. Motivated by the success of MICA, the modified independent component regression (MICR) method for predicting quality properties of non-Gaussian processes keeps drawing attention within the soft sensing circle. However, the determination logic for non-quadratic functions involved in the iterative procedures of MICA algorithm has always been empirical. Without enough prior knowledge, no theoretical investigation can be carried out to conclusively prove which non-quadratic function is optimal for improving the precision of regression models. The selection of non-quadratic functions is still a challenge that has rarely been attempted. Recognition of this issue activates the current study, which proposes a novel soft sensing approach through taking advantage of ensemble learning strategy. Instead of focusing on a single non-quadratic function, the proposed ensemble MICR (EMICR) method takes all three non-quadratic functions into account and combines multiple base MICR models into an ensemble through assigning different weights. The enhanced soft sensing performance is validated through case studies on three non-Gaussian systems.