Development of Well Placement Workflow for Infill Drilling

Abstract Infill drilling has already been implemented in major basins. Optimizing well placement is critical to enhance production and economic returns. Traditional methods, which depend on physics-based models or empirical data, can be slow and yield suboptimal results. This study explores a machine learning framework for reservoir simulation to improve decisions regarding well spacing and stimulation design, which are pivotal for maximizing hydrocarbon recovery and reducing development costs. Our approach involved a comprehensive development of advanced optimization techniques and algorithms, forming the foundation for our machine learning framework tailored for shale production. The framework development included: 1) collecting and preparing data from various sources, 2) selecting crucial features like production rates, geological characteristics, and well interference (frac-hit), 3) training Deep Convolutional Neural Networks (DCNN) with the selected data and features to forecast the outcomes of the completion process, 4) validating these models with a separate dataset to ensure accuracy and reliability, and 5) optimizing critical parameters and evaluating the predicted performance of each configuration. This method provides a complete, machine-learning-driven workflow for modeling shale production. A multiple-layer model structure was implemented on data from the Delaware Basin for both county-wide and regional fine-tuned models. In the process, drainage interference was quantified to optimize the space of new placement. The framework demonstrated its robustness by closely matching the validation dataset with 94% accuracy and reducing evaluation time by over 85% compared to traditional expert assessments. The rest of the work for integrating time-series analytics to enhance data handling and automate adjustments for sensor calibrations and offsets has been planned. The effectiveness of this methodology, verified by current datasets, is ready an extended pilot test. This research introduces a novel workflow employing Deep Convolutional Neural Networks (DCNN) to address the unique challenges of shale development. This approach leverages specialized algorithms and techniques to optimize infill drilling strategies accurately, minimizing risks associated with well interference from hydraulic fracturing and enhancing overall productivity.