Modelling and Simulation of Natural Gas Condensate Production Using Artificial Neural Network

Abstract

Abstract Natural gas condensates are hydrocarbon liquid streams separated from natural gas when it is cooled to temperatures within the upper and lower limits of the hydrocarbon dewpoint curve at a definite pressure in a cryogenic gas plant. They find useful applications in the petroleum and petrochemical industry for production of high octane-petrol, jet, diesel and boiler fuels as well as in production of aromatics, olefins and other monomers used in the production of plastics, synthetic rubbers, resins and fibers. This research paper focused on modelling the process involved in obtaining natural gas condensates from raw natural gas in a cryogenic plant. Stationary-state process data were obtained from a natural gas processing facility in southern Nigeria. The data were pre-processed, five inputs and three output variables were then carefully chosen and arranged in cell arrays in Microsoft Excel before being incorporated into a written MATLAB m-file script, which was ran to generate time-series input-output datasets in Microsoft Excel via a developed Simulink transfer function model. The generated chaotic time series dataset was then fed into the neural network graphical user interface of MATLAB R2021a software and optimized using Levenberg Marquardt, Bayelsian regularization and conjugate gradient algorithms respectively to develop neural network models that represented the production process. Two key indices, namely the mean squared error (MSE) and regression value were used to evaluate the level of accuracy of the developed neural network models. The results obtained revealed that the neural network models developed could effectively capture the underlying trend in the time-series dataset with the Levenberg-Marguardt optimized-neural network having a faster convergence time of 10 seconds, higher regression value of 0.999 and lower MSE value of 0.0489.