Research On Control Strategy Of Catalytic Gasoline Adsorption Desulfurization Device

With the development of automobile industry, the increase of the consumption of fossil energy, the environment problem becomes more and more serious. Production of clean gasoline with low sulfur becomes the focus of the work of petrochemical enterprises. This paper takes the S zorb device used in production of low sulfur gasoline in SINOPEC Luoyang Department as the research object. Through the analysis of the principle, technological process and desulfurization of device to determine the main technological conditions affecting the product quality, and based on the current problems existing in the feeding hydrogen oil ratio and the heating furnace outlet temperature control scheme, a new control scheme is designed.In view of the problems existing in the hydrogen-oil ratio control scheme cann't guarantee feed gas flow and the circulating hydrogen flow rate ratio constant, this paper designs a double closed-loop ratio control scheme with blend station which combine the advantanges of parallel structure of double closed-loop ratio control and serial structure of double closed-loop ratio controller. And according to the demand of circulating hydrogen flow need to fast track the catalytic gasoline in ratio control system, three PID controller is designed. Simulation analysis shows that the fuzzy PID hybrid controller has the smallest instantaneous error, the single neuron PID controller has the fastest instantaneous error amount decreases speed, this two solutions are better than current PID control.Based on the large lag characteristics of heating furnace temperature control, analyzing the present stage of heating furnace temperature control plan, this paper proposes a cascade-Smith with gain compensation control scheme. For Smith forecast controller need accurate mathematical model of controlled object, this paper use RBF neural network instead of Smith forecast controller, to overcome the problem of model mismatch correction function. For the k-means clustering algorithm are clustering slow in the process of RBF neural network online modeling, this paper proposes a improved k-means clustering method in the pretreatment of the initial clustering center, improving the speed of clustering. Finally, the cascade-Smith with gain compensation control scheme and cascade-RBF neural network predictive control scheme are simulated, the simulation results show that the influence of model mismatch in cascade RBF neural network predictive control scheme is less than the cascade-Smith with gain compensation control scheme.