Forecasting Dead Oil Viscosity Using Machine Learning Processes for Niger Delta Region

Prediction of Dead oil viscosity using experimental measurements is highly exorbitant and time consuming, hence the use of forecasting models. Dead oil viscosity is a very important PVT parameter that solve numerous reservoir engineering problems and one of the most required factors for enhanced oil recovery processes. This study utilized two machine learning algorithms of Artificial   Neural Network (ANN) and Support Vector Machine (SVM) to predict dead oil viscosity. A total number of 243 data set was obtained from PVT report from Niger-Delta, out of which, 70% were used to train the models, 15% for testing and 15% for validation.  Quantitative and qualitative analysis was carried out to compare the performance and reliability of the new developed machine learning models with some selected empirical correlations. The result revealed that the Artificial Neural Network Outperformed Support Vector Machine (SVM) as well as common dead oil viscosity empirical correlations with the best rank of 0.144, highest correlation coefficient of 0.984, Mean Absolute Error (Ea) of 0.205, with a better performance plot, followed by Support Vector Machine model with correlation coefficient of 0.926, Mean Absolute Error (Ea) of 0.199 and the rank of 0.176. The new developed Artificial Neural Network model can potentially replace the empirical models for dead oil viscosity predictions for Niger Delta region.