Title : Proactive sand management using machine learning: A predictive approach to well intervention optimization in a major brownfield of upper assam basin
Abstract:
Sand production in oil and gas wells leads to equipment erosion, increased maintenance costs, and production inefficiencies. Traditional sand control methods are reactive, resulting in unplanned shutdowns and high intervention frequency. This study leverages machine learning (ML) techniques to identify high-risk wells and forecast sand accumulation trends, enabling predictive intervention planning. By integrating data-driven insights with field validation, this research aims to optimize well interventions, reduce downtime, and enhance production efficiency in a major brownfield of Upper Assam Basin. This study applies a structured ML framework to analyse sand production trends and optimize intervention strategies. The process begins with feature selection, incorporating key production parameters such as oil rate, water cut, gas rate, well age, depth, and historical sand production data. Wells are categorized into low, medium, and high sand production risk using unsupervised clustering techniques, enabling risk-based intervention prioritization. To complement clustering, a regression-based predictive model is employed to estimate sand fill-up rates, allowing operators to anticipate sand accumulation trends and schedule cleanouts before production losses occur. Validation is performed using historical intervention records and field data. The effectiveness of clustering is assessed using the Silhouette Score and Davies-Bouldin Index, while regression models are evaluated with standard error metrics. The ML-driven classification aligns well with historical intervention records, validating its reliability in identifying sand-prone wells. The clustering model successfully differentiates wells based on sand accumulation risk, forming the basis for a data-driven intervention matrix. Regression-based forecasting provides accurate predictions of sand fill-up rates, enabling operators to schedule proactive cleanouts and optimize resource allocation. Early field implementations in a major brownfield of Upper Assam Basin have demonstrated that timely, data-driven interventions can substantially improve maintenance scheduling, reducing the frequency of unplanned shutdowns and mitigating production losses. This integrated framework, continuously refined through real-time field data feedback, leads to improved well performance, extended equipment lifespan, and overall operational efficiency, thereby supporting sustainable asset management.
