Soft-Sensing Modeling And Optimization For Ethylene Pyrolysis Furnace Based On Neural Networks

Complex industrial process modeling and optimization is one of the hottest research points in the process control fields and also a hard work to apply control theory to the practical industrial process control. Cracking furnace is the key facility in ethylene productions, whose stability, safety and effectiveness will play an extremely important role on the industrial ethylene process. Therefore, application of APC (Advanced Process Control) and Operation Optimization to the cracking furnace of ethylene has great theoretic and practical significances. In order to realize APC and Operation Optimization for the cracking furnace of ethylene, it always need to on-line measure the yields of industrial cracking furnace by using the on-line chromatogram analysis instrument. However, the on-line analysis instrument exist the shortcoming of long time delay, high investment and easy trouble and the like. In accordance with these problems, the soft-sensing technique which is appeared recent years is adopted in this paper, and soft-sensing methods are discussed. The intelligent soft-sensing method based on NN (neural network) is proposed to adopt in this paper.The research topic of the article comes from the improvement and expanding of the ethylene control system for a certain large-scale factory, and the product yields of cracking furnace is selected as the control object. For the industrial ethylene cracking process is complex, the article starts off from analyzing the internal mechanism of cracking process, and then studies the modeling and optimization methods with NN and WNN (Wavelet Neural Network). At last, the article discusses how to apply it to the practical producing process with the combination of engineering application problems. The main research work is described as follows:1. A survey of current researching status of modeling, control and optimization of ethylene process is summarized. The modeling methods of soft-sensing technique are discussed.2. Deeply analysis is developed to the process mechanism of cracking process and the influence of every factor in the process so as to find out secondary variables for the soft-sensing model.3. Deeply Analysis of main modeling methods with NN is conducted. Researches of construction of classical BP NN, RBF NN and pop Elman NN are emphasized. 4. Owing to using spot data, Methods of the preprocessing of input data are discussed. The elimination of gross errors and random errors use statistic methods and filtering respectively. Through soft-sensing modeling, these methods are compared and the effectiveness of these methods is illustrated through modeling effect.5. On-line estimation soft-sensing models of yields of industrial cracking furnace were constituted based on BP NN, RBF NN and Elman NN respectively. And the corresponding program has been compiled in Matlab.6. Researches of popular Wavelet Analysis in nowadays are made. Analysis of the combination of wavelet with NN is conducted, and a kind of wavelet-NN modeling algorithm is proposed, which shows its performance is better than conventional NN modeling methods.7. Considering the problems of cascade control of the degree of pyrolysis in the project, a preliminary research of combining modeling with DCS to realize the intelligent control of the degree of pyrolysis is conducted.