Hybrid Modeling Approach For Complex Process Combining Mechanism Model And Intelligent Model

Three modeling approaches of complex process are mechanism modeling, empirical modeling and hybrid modeling. Hybrid modeling approach, which combines process mechanism model and empirical model, uses the known process mechanism model as the basic model while the empirical model to make up the mechanism model. Considering that the hybrid model has both advantages of mechanism model and empirical model, it has a promising future and also has received extensive attentions of numerous scholars. For the past few years, hybrid model has been successfully applied in the field of process modeling, process control, process optimization, process monitoring and so on. Although some achievements have been obtained, hybrid modeling technology also has many shortcomings because of the development of hybrid modeling has just started. The main study contents of the hybrid modeling are aiming at the characteristics of the process to improve the whole performance of the process model which uses the various kinds of information sources sufficiently. This paper mainly studied the hybrid modeling approach of chemical engineering process and biology engineering process under some different circumstances, and also studied the correction of laboratory hybrid model to the industrial hybrid model. The main research achievements of this paper can be summarized as follows:(1) Considering that the reaction rate mechanism model is missing in the complex reaction process, a hybrid kinetic model based on the artificial neural network reaction rate model was proposed and used for the modeling of p-xy\ene (PX) oxidation to terephalic acid (TA). Artificial neural network model was used to develop the reaction rate intelligent model of PX oxidation process. Based on the reaction rates of various reactant at different reaction times predicted by the reaction rate intelligent model, the process mechanism model (mass conservation equation) was combined with it. The process hybrid model was developed based on the reaction rate intelligent model and the mechanism model. The experimental data which obtained from a laboratory semi-batch reactor of PX oxidation was used to develop and verify the hybrid model. Results show that the obtained hybrid model can accurately predict the concentrations of various reactant at different reaction times of PX oxidation process.(2) Considering that the reaction rate mechanism model is known while the influence law of different reaction factors on the reaction rate constants is unknown in the complex reaction process, an embedded hybrid kinetic model based on the support vector regression reaction rate constant model was proposed and used for the modeling of PX oxidation to TA. Because of the reaction rate constants of the process reaction kinetic model have the problems of uncertainty and inaccuracy, a two-step parameter estimation method was proposed to estimate the reaction rate constant of the reaction kinetic model. Considering that the missing knowledge of the reaction rate constants, an intelligent model was developed to represent the relationship between the reaction factors and the reaction rate constants. This intelligent reaction rate constants model was embedded into the process reaction kinetic model. The process embedded hybrid kinetic model was developed based on the intelligent reaction rate constants model, the reaction kinetic model and the mass balance equation. This process hybrid model was used to model the PX oxidation process. Three kinetic model of PX oxidation in the literatures were considered, and then the intelligent model of artificial neural network model and support vector regression model were embedded into the three kinetic models. Six embedded hybrid kinetic model were developed. The experimental data which obtained from a laboratory semi-batch reactor of PX oxidation was used to develop and verify the embedded hybrid kinetic model. Eventually, the best embedded hybrid kinetic model was obtained for the PX oxidation process.(3) Considering that the part reaction rate mechanism model is inaccuracy while the influence law of different reaction factors on the reaction rate constants is unknown in the complex reaction process, a hybrid kinetic model based on the parallel-to-series of two artificial neural networks and the mechanism model was proposed and used for the sodium gluconate fermentation process. Considering the lack of some mechanism knowledge on mycelium growth in the fermentation mechanism model, the artificial neural network model was utilized to develop the mycelium growth rate intelligent model. The three-layer feed-forward network combined with the Alopex-differential evolution (Alopex-DE) algorithm was employed to develop a model of kinetic parameters given that mechanism model mismatch exists in different fermentation batches (i.e. different kinetic parameters on different operation conditions). These two artificial neural network models were combined with the mechanism model of gluconate fermentation into an overall hybrid model. The proposed hybrid model solved the two problems of mechanism model (missing mechanism knowledge and model mismatch) well. Obtained results show that the hybrid model performed better than the pure mechanism model and pure intelligent model.(4) Considering that the large difference between the complex reaction process in the laboratory process and that in the industrial process, a correction model was proposed to adjust the laboratory hybrid model to the industrial hybrid model and used for the modeling of industrial process of PX oxidation to TA. The performance of the laboratory hybrid kinetic model is limited when we looks for the industrially application due to the large difference between the nature of PX oxidation in the laboratory semi-batch reactor and that in the industrial continues stirred tank reactor. The artificial neural network was used to develop the correction model to modify the reaction rate constants of laboratory hybrid model to be suitable for the industrial process. Due to the lack of entire training data of reaction rate constants, a new optimization strategy was designed. In order to reduce the output error of the model, genetic algorithm was used to optimize the weights and thresholds of the artificial neural network. Eventually the hybrid model for the industrial PX oxidation process was developed. Results show that the artificial neural network correction model can adjust the laboratory kinetic model parameters effectively and the performance of obtained industrial hybrid model are better than the laboratory hybrid model. The industrial PX oxidation process can be simulated accurately by the proposed industrial hybrid model.