Optimizing Depth Selection for Formation Pressure Testing in Complex Reservoirs – A Machine Learning Approach

Abstract Formation pressure testing is crucial for characterizing complex reservoirs, providing insights into pressure regimes, fluid mobilities, and fluid contacts. The effectiveness of these tests depends on selecting appropriate depths that are most likely to yield meaningful pressure data. This selection process is influenced by several factors such as hole conditions, reservoir rock properties, and downhole dynamics. This paper introduces a machine learning approach that integrates well logging data to enhance depth selection, thereby increasing the likelihood of obtaining accurate and valuable formation pressure results. In this study, well log and formation testing data were gathered from various wells in the Mangahewa Field, New Zealand. The well logs included Caliper, Gamma Ray (GR), Density, Neutron, Resistivity, Acoustics, and Nuclear Magnetic Resonance (NMR) logs. These logs were used to account for borehole conditions, rock properties, and mudcake characteristics. Multiple supervised machine learning classification models were tested and developed to identify the best fit-for-purpose model predicting which depths would yield meaningful formation pressure data. Although several models performed reasonably well, our selection was guided by their specificity, particularly the true negative rate. Our analysis revealed that the model achieved 94% specificity in identifying depth locations likely to yield invalid pressure data, which significantly cuts down on unnecessary trials. While the model’s performance in predicting valid pressure data depths was comparatively lower, this trade-off highlights its ability to effectively minimize wasted resources and operational costs. Ultimately, the model enhances depth selection and overall efficiency in formation testing, demonstrating that machine learning can be a valuable tool for optimizing depth selection in complex reservoirs. This paper presents a novel application of machine learning to address the challenge of depth selection in formation pressure testing, offering a data-driven approach that can be integrated into current reservoir evaluation workflows. Additionally, the model is designed to be easily retrained and adapted to new fields, allowing for continuous enhancement based on field-specific data.