Abstract: The product yield estimation is a crucial part of the whole plant design in the early stage of petrochemical engineering. Previous research has shown the neural network technologies can perform higher efficient yield prediction than traditional manual estimation and is one of the major trends of the petrochemical engineering informatization in the early stage. In this paper, we designed and preliminarily implemented a petrochemical neural network system using Python and PHP as the implementation means, and established a combination model for the yield prediction of fluid catalytic cracking (FCC) unit (MIP process) applicable to the early stage of petrochemical design.The results show that the yields of MIP process predicted by the neural network models are consistent with actual yields in production process.

Key words: design of petrochemical engineering, the early stage, catalytic cracking, BP neural network, open source language, Python, PHP