MACHINE LEARNING APPROACH TO OPTIMIZE WATERFLOODING WHITE TIGER BASEMENT OILFIELD OFFSHORE VIETNAM

Abstract

The predominant oil production offshore Vietnam comes from White Tiger (Bach Ho) Basement reservoir of which the flow regime is very complicated due the complexity of the spatial distribution of petrophysical parameters such as porosity, permeability, and water saturation. Consequently, the traditional reservoir simulation method for oil production forecasting is somewhat not accurate or takes a lot of efforts and time to optimize the dynamic parameters. Recent development of Machine Learning (ML) algorithms would help to predict the oil rate from water injection rates of each injection wells faster and more acceptable. Once the oil rate prediction can be done by ML approach, the waterflooding optimization can further implemented by any optimization algorithms such as random search, grid search or gradient based one. In this research, the Random Forest algorithm will be used as it shows the most acceptable results as the correlation coefficients between the predicted and actual values is 0.98 and 0.95 for training and testing datasets, respectively. After that the grid search optimization algorithm is applied to find the reasonable water injection rate for each injection wells that increase the oil productivity and net present values (NPV). The results shows that the oil productivity increase of average 2.5% while the NPV increase of average 1.2% by using newly optimized injection schemes.