Real-Time Well Status Prediction Using Artificial Intelligence Techniques for Accurate Rate Allocation

Abstract

Abstract Knowing well operating conditions can help to allocate the rate accurately; however, there are several factors that govern the well status such as wellhead or downhole temperature and pressures. In this study, artificial intelligence techniques will be used to estimate and predict well status using combination of surface and subsurface parameters in offshore areas. Artificial intelligence (AI) techniques have proven their robustness in tackling petroleum engineering problems. Several techniques can be customized to what each problem requires in terms of accuracy and utilization. In this paper, four Machine Learning algorithms (ML) were used to estimate and then predict well operating status. The four algorithms were Gradient Boosting Machine (GBM), Random Forest (RF), Decision Tree (DecT), and Support Vector Machine (SVM). Surface parameters were fed into each model to estimate well operating status. Data were sub-categorized based on well type. Upstream wellhead pressure, downstream wellhead pressure, choke valve position and upstream wellhead temperature were used as features to create each model. Moreover, a prediction model was developed to specify well status at specific circumstances. The four Machine Learning (ML) algorithms were utilized with datasets covering tens of gas wells. The ML models were optimized in terms of its unique parameters for better results. Random Forest (RF) was proved to provide better results with least average absolute relative error and accuracy of 99% between actual and predicted well status although the other two methods gave reasonable errors. Additionally, the selected model was integrated with a real-time dashboard along with all attainable well parameters. Also, accuracy of the rate allocation was achieved after considering the well status from the selected model including all wells. The prediction model demonstrated acceptable results when comparing to the actual well operating status considering the optimization runs for each dataset. The well status prediction model can assist in monitoring well performance proactively by studying the relationship between well parameters which can further enhance the rate allocation process. Also, this can help with identifying opportunities for well profitability and maximizing revenue by avoiding production loss.