Prediction of Heating Value of Natural Gas from Ghana's Oil Fields Using Supervised Machine Learning Techniques

Abstract

The heating value of natural gas is used to determine the quality of the gas sample, hence accurate prediction of heating value helps in controlling the issue of under billing and overbilling between a gas aggregator and an off-taker. The current method of determining heating value with gas chromatograph comes with many limitations as there can be carrier gas leaks, calibration gas issue and many more which can affect the prediction accuracy. The research focused on predicting the high heating value of natural gas based on percentage gas compositions obtained from Ghana’s offshore oil fields using different algorithms with the aid of Colab Notebook Software and selecting the best performing model from the algorithms used. Seven Algorithms namely Artificial Neural Networks (ANN), AdaBoost, XGBoost, Linear Regression, Random Forest, Bagging Regressor and Stacking Regressor (Hybrid model) were modelled to determine the best predictive model using 2021 sample data on Gas specifications obtained from Ghana’s offshore field, of which 80% of the data was used for training and the remaining 20% was used for testing. The performance of each algorithm was evaluated using metrics such as Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), R2 and Adjusted R2 . Random Forest model performed better than all the other predictive models with an R2 and adjusted R2 of 91.66% and 91.43% respectively and RMSE, MAE and MAPE of 1.6821, 0.5517 and 0.57% respectively during the testing stage. The hybrid model and the XGBoost Model equally did very well during the testing and can be relied on for the prediction of heating values of natural gas. The incorporation of this method provides a diverse approach to the analysis of the pipeline dynamic results of the heating value of natural gas.