الفهرس | Only 14 pages are availabe for public view |
Abstract This study presents two novel machine learning implementations in petroleum engineering fields. The primary objective was to diagnose water. production mechanisms in oil wells using machine learning algorithms. Unlike previous studies, which addressed the problem as a classification. task, this study applied concepts of computer vision to achieve better. performance and overcome limitations and constraints that previous work. faced. Methodology, in detail, is demonstrated where mean average. precision was 0.848 on testing dataset. Model performance also showed. the inevitability to remove anomalies from production data accurately prior. to diagnostics or forecasting. Hence, the secondary objective of this study emerged. A detailed evaluation of existing unsupervised techniques that detect novelty in oil production data ascertained the limitations of them in large oil fields. Thereafter, a novel ensemble learning model, which. combines different algorithms, was implemented, and proved to outperform other algorithms. The proposed model achieved accuracy of 84.02% on simulated and real-field datasets. In addition, the approach of statistics was employed to optimize cut-off threshold of outliers. automatically without human interference. Overall, the introduced approaches could be implemented successfully in large digital oilfields for the sake of surveillance and reservoir monitoring. |