TLDR: PETRONAS, in collaboration with Earth Science Analytics, has launched a pioneering digital initiative to integrate AI and machine learning into its exploration and production (E&P) workflows. This project, focusing on East Coast Peninsular Malaysia, aims to improve subsurface data analysis, reduce uncertainties, and identify new hydrocarbon opportunities through advanced data harmonization, log infilling, and reservoir characterization.
The oil and gas industry is undergoing a significant digital transformation, with artificial intelligence (AI) and machine learning (ML) emerging as crucial tools for faster and more accurate decision-making, particularly in subsurface analysis. A groundbreaking project, a collaboration between PETRONAS MPM and Earth Science Analytics, is at the forefront of this shift, integrating PETRONAS myPROdata with EarthNET to create an AI-powered geoscience and subsurface reservoir characterization ecosystem in East Coast Peninsular Malaysia.
This initiative is structured into two primary tasks. Task One focuses on the seamless integration and transfer of data between PETRONAS myPROdata repositories and EarthNET. This ensures that geoscientists can efficiently visualize, analyze, and interpret extensive well and seismic data. The integration allows for direct selection of wells and seismic surveys from myPROdata dashboards for instant visualization in EarthNET’s browser-based environment, with logs displayed on the fly. EarthNET also provides pre-trained ML models for fault prediction, offering immediate structural insights.
Task Two is dedicated to subsurface analysis using machine learning. This involves applying modern data management and AI-driven analytics for well data harmonization, log infilling, and reservoir property prediction. Phase One of this task encompassed the ingestion, quality control, harmonization, and reservoir characterization of over 2,000 wells, including approximately 1,600 development wells and 550 exploration wells. A significant challenge addressed was standardizing diverse data formats and units from multiple sources.
A key component of Phase One was supervised machine learning for log infilling, designed to enhance data coverage and reliability for consistent reservoir characterization. The methodology involved training ML models on available logs, blind testing for optimal performance, and applying a prioritization framework. This approach led to a substantial expansion of data coverage; for instance, DTC (sonic) log coverage improved from 57% to 87%, and density logs increased from 4 million to 6 million data points.
Following expanded log coverage, the project moved into reservoir property prediction. ML models were utilized to derive critical properties such as porosity, water saturation (Sw), and clay volume (Vcl). These predictions facilitated one-dimensional pay analysis and the classification of wells based on reservoir quality. A crucial aspect of this phase was the quantification of uncertainty, allowing geoscientists to distinguish between high-confidence predictions and those requiring further scrutiny.
EarthNET’s visualization tools, including a Well Viewer (myPROdata module) and cross-plots, enable comprehensive analysis of both measured and ML-predicted datasets. Map plots further illustrate the spatial distribution of wells with applied filters (e.g., porosity >0.4, Sw <0.2, low uncertainty) and allow zooming into specific intervals. Pay class analysis, developed in collaboration with PETRONAS MPM, categorizes reservoirs based on quality, from high-quality (porosity > 0.4, Sw < 0.2) to progressively lower quality classes.
Also Read:
- UK Government’s AI Investment Surges Amidst Persistent Data Quality Challenges
- Norwegian Potato Processor Hoff SA Pilots Generative AI for Factory Optimization
A critical outcome of this project is its emphasis on traceability and uncertainty quantification. Every predicted interval can be traced back to the specific machine learning model used, the contributing features, and the associated uncertainty (standard deviation). This transparency ensures that AI-driven predictions are not ‘black boxes’ but are accompanied by context, fostering trust and enabling geoscientists to assess their reliability. This comprehensive approach transforms raw well data into actionable insights, accelerating exploration and unlocking new hydrocarbon opportunities in East Coast Peninsular Malaysia.
