Data driven soft sensor development for complex chemical processes using extreme learning machine

摘要

In this paper, a novel double parallel extreme learning machine with Pearson correlation coefficient based independent subnets (PCCIS-DPELM) was proposed for accurately modeling complex chemical processes. Compared with traditional ELM, PCCIS-DPELM has two salient features. One feature is that there are two independent subnets based on the Pearson correlation coefficient (PCC) between the input attributes and output attributes. Another feature is that PCCIS-DPELM has a double parallel structure. The PCCIS-DPELM model can well deal with the highly nonlinear data generating from complex chemical processes. In order to test the performance of PCCIS-DPELM, two complex processes of the Tennessee Eastman (TE) and the purified terephthalic acid (PTA) were selected. Then PCCIS-DPELM based soft sensors were developed for modeling the two complex processes. Compared with double parallel ELM (DPELM) and ELM, the experimental results of the two applications demonstrate that the PCCIS-DPELM model with less number of parameters can achieve smaller predicted relative errors. And the PCCIS-DPELM model can respond faster than the other two models. It is proved that the proposed PCCIS-DPELM is a promising method for accurately modeling complex chemical processes. (C) 2015 The Institution of Chemical Engineers. Published by Elsevier B.V. All rights reserved.