Abstract: Three-layers BP neural network was established using MATLAB to predict the hydrocracking conversion, kerosene product endpoint and pressure drop of high pressure heat exchanger shell of middle pressure hydrocracking plant. The results revealed that the sample data quality and number of network hidden layer nodes affect the BP network accuracy evidently, while the network could predict the process parameters, product properties, and heat exchanger state primely. The prediction accuracy for hydrocracking conversion is the lowest, the relative error is ±(5%—10%); the accuracy of the prediction of jet fuel endpoint is high with a relative error of ± (0.15%—2.0%); The absolute error of the pressure drop prediction of the shell side of the heat exchanger is within ±0.03 MPa, indicating that the network established satisfies the requirements for heat exchanger condition monitoring.

Key words: BP neural network, middle pressure hydrocracking, prediction