Gas- Oil Ratio Prediction Using Machine Learning Procedures for Niger Delta Region

The laboratory measurement of Gas-Oil Ratio (GOR) is highly expensive and time consuming, hence the use of predictivemodels like empirical correlations, equation of state and artificial intelligent tools. The solution gas–oil ratio (GOR) is thequantity of gas dissolved at reservoir pressures in reservoir fluids. This study adopted two machine learning procedures ofArtificial Neural Network (ANN) and Support Vector Machine (SVM) to predict GOR. A total number of 852 data set wasobtained from PVT report from Niger-Delta, out of which, 70% (596) were used to train the models, 15% (127) for testing and15% (127) for validation. Quantitative and qualitativeanalysis were carried out to compare the performance and reliability of thenew developed machining learning models with some selected empirical correlations. The result revealed that the ArtificialNeural Network performed better than the Support Vector Machine (SVM) as well as some common selected GORcorrelations.ANN performed better than other evaluated tool with the best rank of 0.139, highest correlation coefficient of 0.98,Mean Absolute Error (Ea) of 0.41, with a better performance plot, followed by Support Vector Machine model with correlationcoefficient of 0.95, Mean Absolute Error (Ea) of 0.163 and the rank of 0.1616. Obomanu and Okpobiri (1947) performed betterthan other evaluated empirical correlations with the Rank of 0.1751 and correlation coefficient 0.95.This study recommendsObomanu and Okpobiri (1947) correlation to be used to predict GOR for Niger Delta region in absence of this new intelligenttool developed in this research.The new developed Artificial Neural Network model can potentially replace the empiricalmodels for gas-oil ratio predictions for Niger Delta region for quick predictions and higher accuracy.