EAGE Conference on Digital Innovation for a Sustainable Future

Development of data-driven models has been one of the epitomes of the growth of digitalization in the oil and gas industry. These models can be established by using machine learning techniques. In reservoir engineering, they have been employed to solve reservoir management problems, including production optimization, history matching, and geological uncertainty quantification. One of the benefits of such data-driven models is the lower computational cost while preserving the accuracy of results. In this work, we have demonstrated the workflow for the construction of data-driven models (which are termed “proxy models” here) that act as a substitute of numerical reservoir simulation to conduct the well control optimization. For illustrative purposes, we have utilized the OLYMPUS model as benchmark. The results yielded from this work showcase the reliability of implementing the proxy models in resolving the reservoir engineering problems.

The latest 12th Malaysia Plan is directing the nation aggressively towards low-carbon nation. Current application system and infrastructure for most of the operators simply may not be able to support this green growth. This will cause delay in planning and execution along the value-chain and risking most operators in Malaysia to miss their decarbonization targets. These challenges can be solved by migrating operators’ aging architecture and workflows completely onto DELFI E&P cognitive environment, a new modern cloud-based solution which unlocks significantly higher computing and simulation power leveraging on latest cutting-edge technology.

Through project transition planning, petabytes of data including enterprise applications and databases can be migrated from on premise to cloud. Strengthened by a robust enabling environment in cloud E&P environment, petrotechnical experts can be onboarded with minimal interruption to their technical delivery included their complex key integrated workflows. In fact, higher efficiency was recorded with users’ satisfaction scores regularly exceeding 90% from deployment surveyed.

Cloud digital solution has proven to significantly accelerate its efforts towards decarbonizing its assets and build a strong, solid foundation for a fully digitalized end-to-end asset modelling and management platform. This will unlock new opportunities to leverage on other technologies such as AI to study new possibilities for production optimization, cost reduction and sustainable development.

The use of Deep Learning on unstructured data, like images and text, has recently made some significant advancement both in capability but also on simplicity of utilisation. As a contribution to E&P, we present three practical cases where these technologies are producing an excellent contribution to geoscience. In this study, we would like to highlight how Deep Learning, and more specifically computer vision, can support geoscientists in their activities. The first case is an identification tool that allow to classify macroscopic rocks sample into lithological classes, useful for both support to non-geoscientists in industrial setting and to promote geodiversity to citizens. The second case is a detection and categorization algorithm for microfossils on thin sections, which can be used on fast identification from microfossils or adapted for mineral to serve as lithological recognition. The third case is a lithofacies identification algorithm on core images that helps, based on expert identification, to perform an automatic recognition for the rest of the core. All these technologies can be easily adjusted to the needs of sustainable georessources and integrated to the E&P workflows.

The move towards digitalization of the oil industry is a must for the time and cost saving. All actual methods for the wettability determination have a negative impact on altering the in-situ wettability and longer time of determination. Our methodology provides instant determination without altering the initial wettability.

This article proposes a new approach to well modeling in integrate modelling of oil and gas fields. Using machine learning describe the well as the dependence of bottomhole pressure on fluid flow rate, water cut, GOR and wellhead pressure.

The well model was implemented using the “Random Forest” assembly of “Decision Trees” using the gradient boosting technique. The model was tested on synthetic and real data from various fields.

The proposed approach outperforms linear interpollation of VFP tables in terms of speed and prediction score.

Let’s take a break from our daily high-tech race, zoom out a little and fallback. In this paper, we would share a case study with our integrated data strategy (A tweak of classical way) – where we would visually present some of the simple yet effective way how we condense our highly complex information (multi-disciplinary data: PP, Geological, Geophysical and RE production data) into the most condense form of visualizations that would pack the most information in a simplistic manner, aiding interpreter to conceptualize geology, screening for infill/NFE opportunity, while bridging communications - giving ability to interpreters to connect the dots from the most fundamental perspective of geology and rock physics, and how our integrated data visualization strategy could bridge both worlds where Interpreter can intuitively analyze and describe these visuals, and how insights and opportunity is generated in a simplistic, classical manner.

Method and/or Theory

• Problem Statement - The complex nature of GnG dataset
• Methodology – Fundamental Rock Physics to Geomorphology Visualization
• Application – Integrated Dataspace for opportunity screening and Visualization
• Integrated Seismic to Well calibrations
• Integrated Litho-Fluid Atlases
• Classical multi-attribute visualization, validation and interpretations
• Result – Opportunity and Way Forwards

The objective of this case study is to predict the extension of sand reservoir intervals to guide the location of infill wells and their trajectories. The reservoirs are deposited in fluvial system of channel point bar and deltaic complexes. The field suffers the problem of uncertain reservoir connectivity and complex fluid distribution due to multiple sedimentation processes and different rock provinces. An efficient approach of machine assisted interactive classification was taken to effectively classify the lithologies and fluids of different intervals to assist in infill wells placement based on the seismic and its attribute volumes available.

A new EOR monitoring system was internally developed for the large-scaled polymer injection projects in the Thailand’s largest onshore oilfield. The EOR monitoring system comprises an EOR monitoring dashboard and an EOR daily report dashboard, which improve the work process, save manpower in data preparation and analysis, and more importantly provide the insights of ongoing flood performance leading to improved decision making and prompt actions. This system helps assure and maximize the incremental oil gain from EOR.

For digital field management in the oil and gas industry, digital transformation opens up a world of new possibilities. The objective is to increase operation efficiency of sand monitoring and control by using web-based interfaces to reference with operation efficiency improvement. Vardoulakis I. and et al (1996) were the first proposed the hydro-erosion model based on rigid porous media in 1996. The erosion models were extended to include the effect of the deformation of porous media in a consistent manner ( Wan and Wang, 2002 ) and geomechanics in a consistent manner ( Wang et al., 2004 ). Methodology was starting from data collection, conducted Simplified Representative Elementary Volume (SREV), model calibration, monitoring, integration of well, facility for prediction of sand accumulation. In addition, a data driven was imbedded to integration as well as workflow automation and user surveillance dashboard in web-based interfaces. The calibration process has been proved the models results with greater than 87% accuracy of sand production. It was found that SREV model approach to sand prediction and control monitoring (SPCM) successful gave an improvement in operations efficiency by reducing time spent on manual analysis and decision-making process through digital web-based dashboards.

The technical abstract is aimed at highlighting an innovative approach to reduce the sub-surface uncertainty for prospect imaging through a fluid simulation model based approach in integration with advanced geophysical properties to evaluate the preferred migration scenario coupled with the seal integrity analysis. This approach helps in mitigating the risk associated with seal and charge in addition to the trap and reservoir condition to augment the existing conventional approach for prospect evaluation. This modeling approach integrates the data from regional scale to the prospect scale through a seamless connection through scalability. But, in order to capture the original resolution of the model, the key challenge comes in terms of the slow and high computing power intensive Darcy migration model and limitation in hardware capability. In this abstract we have highlighted the cloud elasticity feature which helps in retaining the original resolution of the model without compromising the geological consistency. Additionally, the advanced Invasion Percolation migration model has been used to have the robust migration scenario in lesser time and investigate uncertainty analysis for more informed decision.