Data Driven and AI Methods to Enhance Collaborative Well Planning and Drilling Risk Prediction

Abstract

ADNOC is continuously enhancing its capabilities to manage its oil and fields efficiently by better planning, execution and operations that drives field development decisions, well performance, and safe operations. In this regard, ADNOC envisages to leverage the evolving Oil and Gas 4.0 technologies to enhance the well planning decisions of the sub-surface and drilling team through data-driven and AI methods. Effective well planning and operations require collaboration between different subsurface teams and drilling team leveraging multidisciplinary data, historical events and risks and constructing integrated drilling and sub-surface model for collaborative planning and keeping the model live. This requires having a live sub-surface model that is kept close to the field reality while reducing uncertainties. However, extracting key learnings, knowledge and experience from a variety of sources and reports is intense and requires lot of manual processing of data. An AI-based solution leveraging data analytics, natural language processing and machine learning algorithms is developed to automatically extract knowledge from a variety of data sources and unstructured data in building a live intelligent model that enables effective well planning, predicting operational hazards and plan mitigation. The solution systematically extracts, collects, validates, integrates, and processes a variety of data in different formats such as well trajectory, completion, historical events, risk offset well information, petrophysical data, geo-mechanical data, and technical reports. Newly acquired data comprising drilling events, geological and reservoir properties are integrated continuously to keep the model live and digital representation.

StackGBM: Stacked Gradient Boost Machine for Accurate Lost Circulation Prediction

“…The application of artificial intelligence methods in predicting and diagnosing lost circulation accidents has become a trend in the field of drilling engineering [3]. Recent studies mostly adopt neural networks, support vector machine and random forest, etc., to establish lost circulation prediction models based on machine learning algorithms, demonstrating excellent application prospects [4][5] [6]. Generally speaking, these machine learning algorithms have achieved significant improvement in accuracy compared with human analysis.…”