Prediction of C3 Concentration in FCCU Using Neural Estimator Based on Dynamic PCA

Prediction of C₃ concentration, the most important parameter in determining the product´ s grade and quality control of liquid gas produced in FCCU, was studied. A neural estimator model based on improved dynamic principal component analysis (DPCA) and multiple neural networks (MNN) was proposed to infer the C₃ concentration from real process variables. DPCA was carried out to select the most relevant process features and to eliminate the correlations of the input variables. To reduce the large computing work of DPCA, the arithmetic of DPCA was predigested by constructing a compressed augmented data matrix on the basis of the autocorrelation analysis for input variables. Neural network model was established and used to characterize the nonlinearity of the process. To improve the robustness and accuracy of the neural networks, the MNN was obtained by stacking multiple neural networks which were developed based on the reorganization of the original data. The implementation of the model was presented and the model was applied to fluid catalytic cracking unit (FCCU) to predict the C₃ concentration. Research results show that the proposed method provides promising prediction reliability and accuracy.