Application of artificial neural network genetic algorithm in inferential estimation and control of a distillation column

Abstract

Adaptation of network weights using Genetic Algorithm (GA) was proposed as a mechanism to improve the performance of Artificial Neural Network (ANN) inferential estimator. This is particularly useful for cases involving changing operating condition as well as highly nonlinear processes. As a case study, a fatty acid distillation process was considered. The ANN model trained using GA, employed as inferential estimator was successful in providing on- line estimates to a reasonable accuracy. Comparisons were also made to the feedforward network model trained using Levenberg-Marquardt (LM) training algorithm as well as Elman network. When implemented on-line, GA-based ANN model was proved to be more efficient. The use of on- line retraining further improved the estimator performances. To avoid drastic changes of network weights, a partial network on- line retraining strategy was introduced. In this case, the estimator model did not undergo on-line retraining, but a newly introduced bias model, attached to the main estimator was used for the fine-tuning purposes. Significant improvements were obtained especially when assessing from the perspective of model generalization. The results obtained in this work confirmed the potential of using model update strategy for neural network process estimator.