Soft Sensor Modeling And Cracking Severity Real-time Optimization For Ethylene Furnace

In recent years, with the rapid increase of the economic, the demands for petrochemical products such as ethylene are growing rapidly. Production scale of ethylene has become one important standard to evaluate the level of the oil industrial of a country. How to improve the ethylene production, reduce the consumption of the raw material, increase the benefit of the enterprises have been the focus of the researchers.Cracking furnace is not only the key facility but also the most energy-consuming facility in ethylene productions, so its safety, stability and effectiveness are extremely important. Operation optimization for the cracking furnace is a significant way to improve the product yields. In order to maximize the product benefit, the cracking severity has to be controlled in a limited range. Traditionally, the cracking severity is controlled by coil outlet temperature(COT), however, the value of the COT is determined by the feed type and experience, which is not precise enough. In this paper, the cracking severity is defined by the ratio of the propylene and ethylene yield in pyrolysis products. As the on-line analysis instrument has shortcoming of high investment and long time delay, so it is proposed that build products forecasting model with soft-sensing technique. The optimal cracking severity is real-time corrected according to the process data in DCS, so that the economical optimal index is achieved with different features and operational parameters.This paper is based on the project of Sinopec Petrochemical Company, with the view of achieving optimal economical benefit. The main research work is described as follows:The process mechanism of ethylene cracking, the current researching status of modeling and the optimization of ethylene process is summarized. Deeply analysis of the factors that influence the product yields is discussed. It is described how to deal with the process data, how to cluster the feeds with fuzzy C-means algorithm, how to batch computing with COILSIM1D software, and how to build product yield models with BP neural network algorithm. From the perspective of maximizing the economic benefits, it is described how to optimize the cracking severity with the neural network models, and how to overcome the operating disturbance.