Development of Artificial Neural Network Model to Predict the Performance of the Fractionation Towers in Gas Processing Plant

Abstract

Abstract This paper presents a methodology for the application of the artificial neural networks associated with an optimization technique to predict the performance of the fractionation towers in the gas processing plants. The proposed methodology was applied to nine fractionation towers in the Western Desert Gas Complex (WDGC). The gas plant is owned and operated by the Egyptian Natural Gas Company (GASCO), and located at Alexandria - Egypt. It includes three cryogenic trains (Trian A, B and C) and two fractionation areas. The nine towers are (1) three De-Methanizer towers of Train A, B and C; and (2) two De-Ethanizer towers, two De-Propanizer towers, and two De-Butanizer towers of the two fractionation areas. Nine artificial neural networks were built with a specific target for each fractionation tower. The goal of Networks 1, 2 and 3 is to predict the top vapor analysis and the performance of the De-Methanizer towers of Train A, B and C; respectively. Moreover, the target of Networks 4 and 5 is to provide the analysis and the expected flow rates of the produced ethane/propane (C2/C3) mixture from the two De-Ethanizer towers of the two fractionation areas. In addition, the target of Networks 6 and 7 is to provide the purity degree of the produced commercial propane from the two De-Propanizer towers of the two fractionation areas. Finally, Networks 8 and 9 aim to predict the vapor pressure of the produced Liquified Petroleum Gas (LPG) from the De-Butanizer towers of the two fractionation areas. The collected data of each fractionation tower ranges between 1000 to 1450 records. The data covers a wide range of conditions for a full year of operation. The main parameters and data for the input nodes are the operating conditions (temperatures, pressure, and flow rates) of the fractionation towers. Furthermore, the correspondent outputs data are the compositions and the flow rates of the end products. All the data were reviewed and filtered to ensure better accuracy in the results of the prediction. After the verification of the neural network model, the nine artificial neural networks were associated with an optimization technique to predict the optimum operating conditions of each fractionation tower at the WDGC. The results showed that the mean squared error between the actual measurements and the outputs of the nine artificial neural networks reached a maximum of less than 6% for each network. Applying the recommendations of the developed neural networks improved the performance of the fractionation towers in the WDGC and achieved products with acceptable specifications. The results indicate that applying artificial neural network model with an optimization technique algorithm for a large multivariable non-linear data sets is proven to be a reliable tool for decision-making in the Western Desert Gas Complex. Such methodology can be applied to other gas processing facilities and could aid in overcoming the disturbances of the operating conditions in the gas processing plants.