Numerous scale types normally deposit inside oil production wells; however, sulfate scales are probably the most alarming types due to their high strength and insolubility. Several company cases of slickline scratching and coiled tubing milling fail to clean and remove heavy depositions of barium and strontium sulfates. Observations of the current study show that these sulfate scales deposit due to cooling of super-saline formation waters inside offshore producers and pipelines, besides the mixing of incompatible waters.
Prediction of sulfate scale deposition is challenging. Many of the currently-available prediction software products have drawbacks in sulfate prediction due to the limited experimental data, the uncertainty of ion pair interactions, and the extremely-low solubility of these minerals. Therefore, more experimental work is still needed to investigate extreme field conditions and complex water chemistries.
Different Machine Learning (ML) algorithms are being used in the oil industry with successful applications that are adding and/or replacing the traditional methods. Therefore, the scope of the current study is to utilize ML algorithms in scale prediction. The study investigated actual field scale depositions that were collected from multiple offshore fields from 1998 through 2020 with more than 1400 data records. The available database contains 14 input features including water chemistry, water production rate, oil production rate, gas production rate, pressure, and temperature. Feature engineering was adopted to define the most important features to build the ANN models. The available data was split into training and testing datasets. Several Artificial Neural Network (ANN) models were developed to predict barium and strontium sulfate scales downhole in production wells. A comparative analysis was performed between the developed ANN models against a commercial scale prediction software and empirical correlations. The ANN models outperformed the other traditional methods concerning deposition probability or classification of scale type. Moreover, the ANN models could also predict the amount of scale with accuracy of 93% and 75% for strontium sulfate and barium sulfate, respectively. The Mean Absolute Error (MAE) of scale percentage was 3.6% and 8.2% for strontium sulfate and barium sulfate, respectively.
The paper novelty is the inclusion of actual scale deposits from different fields to build ANN algorithms capable of predicting the real composition of sulfate scales not detecting their super-saturation level. Predicting the type of scale leads to optimize company resources and dedicate personnel efforts to severe cases of hard scale depositions without the need of well intervention.