Answers are grounded in this paper and cite their sources.
Application of machine learning predicts well logs in the Gulf of Thailand, suggesting potential cost savings.
_ This article, written by JPT Technology Editor Chris Carpenter, contains highlights of paper SPE 222299, “Unlocking the Potential of Mud Logs and LWD in the Gulf of Thailand Using Machine Learning,” by Sethawut Palviriyachote, Siddharth Misra, SPE, and Sandro Moelyono, SPE, Texas A&M University, et al. The paper has not been peer reviewed. _ The subject gas field in the Gulf of Thailand (GOT) stands as one of the largest natural gas reserves in Thailand, with over 30 years of development history and more than 1,000 penetrated wells. Use of machine learning (ML) for log synthesis can help reduce expenses and operational risks associated with traditional well-logging methods, including service fees, rig time, and potential retrieval challenges. This study aims to use ML techniques to predict well logs by analyzing mud-log and logging-while-drilling (LWD) data. Introduction Drilling in the GOT faces challenges involving high temperatures, pressures, and complex geological structures, demanding robust logging techniques and careful interpretation to ensure successful outcomes. Many wells feature incomplete well-log data, which complicates development planning. With the advent of ML technologies, however, a transformative approach to predict crucial well logs is presented. The potential cost savings from reducing well-log fees and rig time in one of the fields could amount to $1 million per month. Data-Driven Methodology This section outlines the process of developing ML models for predicting three key targets from openhole logs: bulk density (RHOB), neutron porosity (NPHI), and compressional sonic travel time (DTC). The first two of the five steps of the methodology are detailed in this synopsis. 1. Data Collection. The initial step involves gathering a comprehensive data set of well logs, including LWD, mud-log and openhole-log data. This data set should be scrutinized for completeness, correctness, and relevance to ensure that it provides a solid foundation for subsequent analysis. 2. Data Preprocessing. In this stage, the data is prepared for developing predictive models. First, anomalies that may distort the analysis are identified and removed. Next, new features are engineered, or existing ones are modified to enhance predictive performance. Finally, numerical features are normalized and scaled to ensure generalizable, unbiased predictive performance. 3. Model Selection and Training. This phase involves choosing appropriate regression and ML techniques and then tuning their hyperparameters to obtain highly generalizable prediction models. Key tasks include defining hyperparameters to be tuned and their tuning ranges that ultimately control model generalization. The authors tune random forest, gradient boosting, neural networks, and support vector regression techniques. The overall objective in this study is to process mud logs and LWD logs to predict RHOB, NPHI, and DTC. 4. Model Evaluation. Once the model is trained, its performance is evaluated on test data to determine its effectiveness. Metrics such as R-squared, mean absolute error(MAE), and mean absolute percentage error (MAPE) are calculated to gauge generalization performance. 5. Deployment. The final step involves integrating the trained and tested model into a real-world deployment environment to evaluate the applicability of the predictive model.
No takes yet. Share an insight, caveat, or question.
Chris Carpenter (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: