Process Data Modeling Based on Orthogonal Independent Component Analysis

Based on independent component analysis（ICA), a multivariate linear regression（MLR) method combined with orthogonal signal correction（OSC), which is called orthogonal independent component regression（O-ICR), is proposed for regression prediction of non-Gaussian processes. First, the O-ICA is conducted on an original input data matrix for removing disturbing variation that is not correlated to Y from the extracted high-order statistics in ICA. Then, independent components are extracted X from after correction. The regression prediction model is derived using these components instead of the original input data and Y. Compared with the traditional ICR, the proposed method has a more superior performance because independent components are corrected. Finally, the validity of the method is verified though quality prediction simulation in the Tennessee Eastman（TE) process.