Revolutionizing Lower Completions Design Through Unsupervised Machine Learning Clustering: An Integration of Well Construction Data and Reservoir Characteristics

Abstract This paper presents a novel approach that utilizes unsupervised machine learning clustering (UMLC) to enhance the completions process in reservoir management. By integrating real-time data streams from multiple sources, this method harmonizes and transforms data into a unified format for seamless cross-disciplinary analysis. The goal is to optimize zone selection for completions design, enabling better decision-making in a shorter timeframe and avoiding data pitfalls between Well Construction and Completions stages. Additionally, this paper compares the zones identified using the same wells from previous studies. The proposed integration focuses on the ingestion and processing of real-time data from various sources. This data undergoes harmonization and transformation into a standardized format, which allows for seamless integration and analysis through sophisticated UMLC algorithms that delineate zones based on the data characteristics throughout the lateral well section. Subsequently, these zones are assessed by interpreters to pinpoint defined potential pay zones, non-productive intervals, and optimal locations for completions design equipment, emphasizing production and isolation. This analytical framework fosters a collaborative decision-making environment that promotes synergy among multiple disciplines, including geology, petrophysics, reservoir engineering, and completions, ensuring that all pertinent insights are incorporated into the completions process. The comparative evaluation between interpreter-generated results and UMLC outcomes highlights notable discrepancies in zonation accuracy. The data-driven clustering results yield a more precise and nuanced assessment of zones, enabling interpreters to distinguish between high-quality and low-quality productive zones, as well as non-productive zones. This enhanced accuracy is attributed to the incorporation of data science and machine learning methodologies, which provide users with a deeper understanding of the well and reservoir characteristics. Additionally, the methodology includes a quality control mechanism to establish the feature importance scores of the clustering and to compare the analytical drivers. The amalgamation of UMLC with conventional reservoir mapping techniques signifies a substantial leap forward in reservoir management practices. The data-centric clustering outcomes deliver a degree of granularity and precision that bolsters interpretation, ultimately yielding optimal zonation and extended productivity of lateral wells. By harnessing the capabilities of data science and machine learning, this approach ensures more cost-effective and timely completions processes.