Hierarchical Neural Network Based Product Quality Prediction of Industrial Ethylene Pyrolysis Process
A two-layer hierarchical neural network is proposed to predict the product qualities of an industrial KTI GK-V ethylene pyrolysis process. The first layer of the model is used to classify these changes into different operating conditions. In the second layer, the process under each operating condition is modeled using bootstrap aggregated neural networks (BANN) with sequential training algorithm. The overall output is obtained by combining all the trained networks. Results of application to the actual process show that the proposed soft-sensing model possesses good generalization capability.
KeywordsIndividual Network Good Generalization Capability Hierarchical Neural Network Process Operating Condition Feedstock Property
Unable to display preview. Download preview PDF.
- 1.Zhou, Q., Xu, Y.M.: Predict the Product Yields Profile of Ethylene Pyrolysis Furnace by Multi-Models. In: Proceedings of the IFAC International Conference on Intelligent Control and Signal Processing, Faro, Portugal, April 8-11 (2003)Google Scholar
- 6.Zhang, J.: Sequential Training of Bootstrap Aggregated Neural Networks for Nonlinear System Modeling. In: Proceedings of the American Control Conference, Anchorage, Alaska, U.S.A, pp. 531–536 (2002)Google Scholar