PREDICTION OF GAS-OIL-RATIO BELOW THE BUBBLE POINT PRESSURE USING MACHINE LEARNING FOR NIGER DELTA REGION

Determination of solution gas−oil ratio (GOR) is a very important requirement that helps in multiple production engineering and reservoir analysis issues. Gas-oil ratio is the ratio of the gas volume that comes from the produced oil at atmospheric pressure measured in standard cubic feet (SCF) to the volume of oil produced after the dissolved gas has evolved from it at the surface, measured in (STB). Before now, some empirical correlations existed that determines the solution gas−oil ratio however, they still prove unreliable due to the applied assumptions and their specification to operate only under a particular range of data. In this research, Machine learning models were trained to predict solution gas-oil ratio using Niger Delta reservoir fluid properties precisely for the pressure below bubble-point. The three-machine learning algorithm adopted are Deep neural network (DNN), Supper learner (SL) and Extreme Gradient Boost (XGBoost). A total number of 1083 of data set was obtained from PVT report from Niger-Delta and validated, out of which, 70% (758) were used to train the models, 15% (162) for testing and 15% (162) for validation. Quantitative and qualitative statistical analysis were carried out as to compare the performance and reliability of the new developed machining learning models with commonly used gas-oil ratio empirical correlations. The proposed Super Leaner algorithm model predicted better than the other selected machine learning models and GOR empirical correlations validated. The super leaner model gave the best result with the Rank of 0.144, followed by DNN and XGBoost having the Ranks of 0.419 and 0.503. GOR empirical models Glaso (1980), Standing (1947), Obomanu and Okpobiri (1987) having the Ranks of 0.41, 0.45 and 0.48 respectively. The findings from this research can be applied for estimation of the volumetric properties of hydrocarbon reservoir fluids without the need for conducting routine laboratory analyses.