Leveraging Data-driven Approach for Sand Production Prediction and Management in Oil and Gas Wells

Sand production poses a major challenge in wells completed within sandstone reservoirs, despite their inherent benefits. It could lead to erosion of surface and subsurface equipment, well plugging from sand grain deposition and accumulation, and potential collapse of sections in horizontal wells, resulting in non-productive time, costly cleanup, and remedial or work-over operations. Various researchers have proposed strategies and models to tackle sand production and its associated problems. While these models have successfully predicted the onset of sand production, cavity stability, and rock collapse, they often struggle to accurately forecast the sand production rate due to the complexity of influencing parameters and the difficulty in fully capturing all reservoir and wellbore processes. Anticipating sand production from a reservoir or well prior to drilling and field development is essential to implement appropriate equipment and measures for optimal performance. Thorough investigation, analysis, and modeling of sand occurrence and production are crucial for developing an effective field development plan. This study introduces a comprehensive approach that utilizes artificial intelligence to address sand occurrence, production, and control. To enhance the prediction of sanding conditions, sand production rates, and critical drawdown pressure, a Nu-Vector classification model, an Ada-boost model, and a Random Forest Regression model were developed. The effectiveness of these models was assessed using publicly available field data, demonstrating their capability to accurately estimate sand generation in real-world settings. The study’s findings suggest that completion designers can leverage these models to devise timely sand management plans that minimize production degradation.