Predicting NGL Recovery from Flared Gas Using Nonlinear Multiple Regression and Artificial Neural Network Model

Abstract Recovery of natural gas liquid (NGL) from flared gas is considered one of the techniques of emissions reduction, the recovered quantity depends on several factors including the natural gas composition, operating pressure, and temperature. Process simulation is required to measure the recovery of NGL from the flared gas, and process simulation cases are required to predict the NGL productivity, these cases should be generated at several operating conditions for optimization purposes. In this paper, 12873 data sets are generated using Aspen HYSYS for different natural gas compositions collected from 11 natural gas fields. These huge data sets are used to develop an artificial neural network (ANN) model for predicting the NGL recovery. The developed model was trained based on 70% of the data sets, validated based on 15%, and tested using 15 % of all the data sets. The results show that the developed NN model can accurately the NGL recovery based on the natural gas composition, and the operating pressure and temperature with a coefficient of determination of 0.9999 and an absolute average error of 4%. The proposed model for estimating NGL production rates is a feasible environmental issue. The availability of changing the operating pressure and temperature could help in selecting the proper equipment such as recovery compressors and refrigeration unit size. This approach also could allow production companies to join the project quickly, eliminate carbon dioxide emissions, and start to gain money in a short time.