ABSTRACT
Natural gas has been considered a fuel of choice for increasing the security of energy supply and reducing environmental pollution across the world. Accurate prediction of natural gas density is essential for calculation of other important physical properties and establishing reliable and economical product transfer across pipelines in the oil and gas industry. This work presents a hybrid machine learning model based on coupled Sparse Autoencoder (SAE) and Artificial Neural Network (ANN). The objective is to achieve accurate predictions of natural gas density of gas mixtures that represent five natural gas reservoirs in Qatar’s North Field. Our proposed workflow uses the unsupervised learning capability of SAE to learn and extract important features of the input data and represent them in a low-dimensional space. These important features were mapped to their respective outputs then used to train and test an ANN that predicts the densities of the natural gas mixtures. We demonstrated that the SAE can provide a low-dimensional replica of the input data and reconstruct the original data from the reduced features with a minimal root mean squared error of 0.0048. Furthermore, the performance of the hybrid SAE-ANN model was evaluated by comparing its results to that of an uncoupled ANN and a Support Vector Machine (SVM) algorithm using R-squared score and root mean squared error. Results show that the hybrid SAE-ANN model outperforms the other two algorithms with the highest prediction accuracy and least root mean squared error.
Keywords:
Artificial Neural Network, Natural Gas Density, Machine Learning, Sparse Autoencoder.
