Hybrid Estimator for Chemical Product Composition

A. Ahmad, W.P. Lim, and W.S. Chen (Malaysia)


Inferential Estimation, PLS, Neural Networks, HybridModel.


Successful implementation of partial least square (PLS) regression model in estimating product composition in a fatty acid distillation process has motivated further improvements to improve both accuracy and robustness properties. Initially, a hybrid technique that combines PLS regression and artificial neural networks (ANN) was implemented. The results show that the hybrid method provides better estimation properties and is able to generate reasonably accurate estimation of the product composition. To further improve the accuracy and robustness of the hybrid estimators, two strategies are introduced. Firstly, a bias model using multiple linear regressions (MLR) is employed to provide corrections to estimation errors of the ANN-PLS estimator. Secondly, on-line update of network weights in the ANN model is introduced. Successful implementations of both strategies were obtained. The results obtained in the study confirmed the potentials of the ANN-PLS model a viable estimator for product properties in chemical industry.

Important Links:

Go Back