Perpustakaan Digital ITB

Sand production can lead to various production problems such that knowing whether sand will be produced or not is critical. However, conventional approaches to predicting sanding phenomena need complex methods and information that are difficult and/or expensive to obtain. Several studies show that there is an alternative approach to predicting sanding by utilizing computer models. A binary classification model/algorithm will be developed to predict sand production in a mature offshore oilfield in Indonesia. The needed parameters for the model are gathered from well, production, field, and sand sieve analysis data. A typical process of machine learning is implemented to develop the model. Logistic regression and linear discriminant analysis model is found to be the top candidate for the model’s purposes. Two cases for each two model will be compared. It is found that the case that rescaled several features and used a linear discriminant analysis model has the most optimal result. The model’s potential is not limited to only predicting sanding. It is found that the model can also be further developed to predict critical sand production rate and suggests the most optimal type of sand control completion in a well. It can be concluded that a model has been developed but with only decent performance on predicting when tested. It can be expected that with higher quality and quantity of data the model’s performance can be improved.