Abstract:
Based on the design of ethylene cracking gas compressors, a new prediction model for compressor performance is established by modifying and correcting the design data according to the practical production situations. Based on the idea of learning rate varying self-adaptively with the error change and combined with genetic algorithm (GA) which contains global optimization characteristics, an improved BP algorithm (LR-GA-BP) is proposed for compressor performance prediction. The simulation calculation of a four-stage compressor system in the ethylene unit is carried out. The relative errors between the calculation values and actual values of main components in cracking gas from the fourth-stage outlet of the compressor are less than 2%, and the relative errors between the calculation values and actual values of the outlet temperature and pressure of the compressor are less than 1%, which verifies the reliability of the model. Based on the above model and thermodynamic basis, the effect factors of the higher outlet temperature at the fourth-stage of the compressor are analyzed by regarding the process of cracked gas compressing as a kind of adiabatic compression and the corresponding cooling measures are adopted. The simulation results show that increasing the return flow between compressor segments is beneficial to decrease the compressor outlet temperature and slow down coking in the compression system. The result in this paper is of some certain value for slowing down the coking of compression system and optimizing operation of compressors.