Artificial neural network and semi-empirical modeling of industrial-scale Gasoil hydrodesulfurization reactor temperature profile

The sulphur content of the desulfurization reactor product is strongly related to the severity of the reaction, which is determined by the reactor bed temperatures. In this study, the potential of neural network and semiempirical regression models to estimate temperature variation in industrial-scale gasoil HDT reactor was examined to achieve improved prediction ability. The reactor consists of 4 beds with 10 temperature indicators to control and monitor the temperature. Model developing Inlet temperature, H2/HC ratio, LCO flowrate, cracked gasoline flowrate, and straight run gasoil flowrate as input layer parameters and bed section outlet temperature as output layer was used. A database with 283 different data was built using daily records of the HDS process from an Iranian refinery. Feedforward backpropagation algorithm with learngdm learning function was used to develop ANN models. The number of neurons in the hidden layer varied between 7–13 to find the most reliable network. The optimum ANN models were selected for reactor temperature profiles on the trial-and-error method. The performance of designed models was tested by computing root mean square error (RMSE), and average absolute deviation (AAD). Comparing the neural network model and regression models, the artificial neural network model is the best model for the reactor bed temperature prediction. The value of AAD for temperatures sensors 1 to 10 in the artificial neural network model obtained 0.128, 0.094, 0.091, 0.1, 0.032, 0.067, 0.052, 0.031, 0.086, and 0.062 respectively, which show good agreement between the actual data and the artificial neural network predicted data for temperature profile in the length of the reactor.