Hybird Modeling Method Based On Structual Separation And Parameter Separation

Along with the development of science and technology, chemical industry has become more and more important in our entire economy. Most chemical process is very complicate, the structure and parameters of the system are hard to gain. And most of them are non-linear and time-variant, part of the process data is hard to be obtained accurately on-line. With the above reasons, the modeling of chemical process becomes a bottle neck of the development of chemical industry. Thus, research on chemical process modeling has become a hot spot in chemical area. It will help to enhance the efficiency of chemical industry, harden the security in the productive process, decrease the loss of rough stock, and make the enterprise more competitive.This paper is mainly focus on the modeling method of chemical process, a hybrid modeling method based on structural separation and parameter separation is proposed.Chemical reactor is the key equipment of chemical process. This paper's research is started from CSTR. Many CSTR has the below characters:it has known mechanism, which can be described by the principles of mass conservation and energy conservation, and along with the Arrhenius equation. But the key parameters are unknown. The modeling method in this paper fully utilizes the prior knowledge of the system, use the above characters to approximate the model's structure. And only runs identification to the unknown parameters, keeps the value of known ones. Through this hybrid modeling method, the accuracy and generalization ability of the model is enhanced.This paper develops a hybrid model of a certain type of objects (Simple CSTR), which only has one reaction rate in the model. Then a hybrid model for a complex CSTR is set up through the same method, which has several reaction rates. Finally, the new hybrid modeling method for Polymerization Process's molecular weight distribution model is proposed in this paper.The simulation proves that this hybrid method is effective. The structure approaching hybrid model has high reliability. The result of comparison with other hybrid model illustrates that this approach can enhance the accuracy and generalization ability of the model.