Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
6589410 | Chemical Engineering Science | 2015 | 12 Pages |
Abstract
The crystal size distribution in crystallization processes, the molecular weight distribution in polymerization processes, the particle size distribution in powder industries, and the pulp fiber length distribution in paper industries are all distributed product outputs. Reliable online quality prediction of these chemical processes with distributed outputs is important but challenging. In this work, the kernel learning (KL) framework is introduced to model and online predict the distributed product outputs. First, the KL method is proposed to construct a global distributed shape. Then, without resorting to a KL-based global distributed model, a just-in-time KL (JKL) model is presented for better description of local distributed shapes with more accurate and quick prediction performance. Moreover, an ensemble JKL (EJKL) modeling approach is developed to obtain more reliable prediction performance of the distributed outputs. The proposed modeling methods are applied to online prediction of the molecular weight distribution in polymerization processes and of the crystal size distribution in crystallization processes. The prediction results show the proposed method is superior to the traditional counterparts.
Keywords
CSDRVMRBF-NNSVRGPRJITLSSVRRMSEMWDPSDleave-one-outParticle size distributionCrystal size distributionMolecular weight distributionRoot-mean-square ErrorBatchSupport vector regressionGaussian Process RegressionNeural networksSimilarity factorchemical processesJust-in-timeLOORelevance vector machineCSFSystems engineeringLeast squares support vector regressionProcess controlKernel learning
Related Topics
Physical Sciences and Engineering
Chemical Engineering
Chemical Engineering (General)
Authors
Yi Liu, Zhengjiang Zhang, Junghui Chen,