Predictive Control Of An Ethylene Oxide Reactor On Neural Net-based Model

Concerning the predictive control of an EO (Ethylene Oxide) reactor, this paper presents and accomplishes the problems as follows: First, based on the fact that acquirement data can reflect the law and characteristics of the object, the idea is advanced that the model can be founded by using real-time measuring data. Second, according to the system structure of TDC-3000, the means is brought forward that the real-time measuring data can be collected by the data-sequal function of DCS and that the data can be transmitted from the printer interface of the DCS workstation to internal computers. After experiments, the real-time transmission between DCS and Internet is achieved. Third, the real-time data exchange between different applications of different operating systems such as UNIX, Windows etc. is studied and resolved by using memory share and dynamic data exchange technique. Fourth, the method of real-time data disposal and the generalized distributing distance rule and algorithm of samples are presented by which few can be selected from plenty of measuring data. Fifth, the ANN model is presented, whose main structure is constructed by off-line- trained RBFN and expresses the object's wide-change law, whose secondary structure is constructed by on-line-trained BPN, which expresses the main structure's error and can eliminates the model error that was the result of the object's slow-change. The model is founded by linear addition of the main structure and the secondary one. Simulation results show that the maximum error between the predictive data and the measuring ones is no bigger than 2 percent and that the method is efficient. The double-ANN model, connected with the algorithm of real-time data collection and sample selection, is of great practical value. Sixth, the algorithm theory of predictive control on the model is investigated and the theory and algorithm of the predictive process of real-time data, the predictive calculation of parameters, the predictive calculation of slow tide, the real-time optimization, the calculation of boundaries and the restrictive process is brought forward. Besides, the involved parameters are founded. Finally, the MPC predictive error amendment method of the ANN model and the switch problem for different control means are discussed and presented. With the settlement of the problems before, it's sure that the algorithm and methods of the predictive control for an EO reactor on ANN-based model have been resolved.