Domain Compensation-Assisted Quality Inference Enhancement of Chemical Processes with Distributed Outputs
Jialiang Zhu, Yun Dai, Weiwei Guo, Hongying Deng, Yi Liu
Abstract
The prediction performance of data-driven soft sensors for chemical processes with distributed outputs tends to degrade when distribution discrepancies exist. To meet this challenge, an offset compensation Gaussian process regression model is proposed for the quality inference of chemical processes with distributed outputs. The model first captures the common molecular weight characteristics of different distributed outputs. Subsequently, the product distribution of the molecular weight between different operating conditions is adjusted by the offset compensation mechanism. From a statistical perspective, the conditional distribution discrepancy between the modeling inputs and outputs is reduced. Consequently, by exploring and transferring the distributed knowledge from the related operating conditions, the reliable prediction domain is enlarged. The molecular weight distribution prediction in a polymerization process indicates its feasibility and superiority.