Advanced Machine Learning Models for Well Placement and Completion Optimization

As so much in the oil and gas industry relies on expert interpretation over unstructured data and understanding of elaborate geological concepts, tracking the production and consumption of conceptual knowledge and information is crucial. Systems need to be developed that can capture these interpretative trails in order to meet the needs of exploration.

IBM’s Hyperknowledge Trails may just be that system.

“It is a technology that supports the registration and promotes reasoning about geo-experts’ particular approaches when handling knowledge-intensive interpretative tasks over complex and unstructured data,” said Renato Cerqueira, senior research manager of natural resources solutions for IBM Research Brazil.

He said that was one of the goals of the company’s research group when it designed Hyperknowledge: to advance the state-of-the-art in knowledge engineering for AI systems.

Specifically, Hyperknowledge is a hyperlinked representation and knowledge-base solution that supports association among symbolic and non-symbolic semantics and data fragments. Its representational approach takes advantage of combining, in a single rationale, user interaction, data segments, including sentences of a text document, fragments of images, segments of seismic data, frames of a video file, fragments of executable code and semantic representations. Its foundation is the usual hypermedia concepts of nodes and links.

Please log in to read the full article

As so much in the oil and gas industry relies on expert interpretation over unstructured data and understanding of elaborate geological concepts, tracking the production and consumption of conceptual knowledge and information is crucial. Systems need to be developed that can capture these interpretative trails in order to meet the needs of exploration.

IBM’s Hyperknowledge Trails may just be that system.

“It is a technology that supports the registration and promotes reasoning about geo-experts’ particular approaches when handling knowledge-intensive interpretative tasks over complex and unstructured data,” said Renato Cerqueira, senior research manager of natural resources solutions for IBM Research Brazil.

He said that was one of the goals of the company’s research group when it designed Hyperknowledge: to advance the state-of-the-art in knowledge engineering for AI systems.

Specifically, Hyperknowledge is a hyperlinked representation and knowledge-base solution that supports association among symbolic and non-symbolic semantics and data fragments. Its representational approach takes advantage of combining, in a single rationale, user interaction, data segments, including sentences of a text document, fragments of images, segments of seismic data, frames of a video file, fragments of executable code and semantic representations. Its foundation is the usual hypermedia concepts of nodes and links.

“Subsurface characterization is a knowledge-intensive process that requires experts from different fields of geosciences for several months analyzing and interpreting huge amounts of data,” said Cerqueira, who is leader of IBM’s Natural Resources Solutions area, where, in recent years, artificial intelligence and other digital technologies have been successfully applied to different data interpretation activities to address challenges in subsurface characterization.

Making the Abstract Concrete

“However, the understanding and tracking of contextual information during interpretation processes, as well as the capture and curation of experts’ domain knowledge, are still open problems to enable a broader applicability of AI and a continuous evolution of AI systems,” he said.

That’s the challenge the Hyperknowledge Trails technology – to address some of these problems.

“Registering geologists and geoscientists’ interpretive trails is not a trivial task,” he said.

That is because the process commonly relies on advanced symbolic constructions, logical observations and numerical processing techniques to analyze complex surface and subsurface data.

Cerqueira, whose work is focused on addressing major business and technical challenges of the natural resources industries, said, “Our research agenda implies a constructivist take, where basic events of support systems are encapsulated and connected to user intent and actions.”

This system will allow the user to structure different levels of information in a knowledge graph, relating low-level system events to a high-level abstraction perspective, while considering facts about the expert’s activity.

“The capturing process generates valuable information that can be consumed by AI systems, which can potentially detect patterns and provide recommendations to experts in their activities. Such information can also support knowledge curation in shared knowledge graphs,” he said.

Team Effort

Artificial intelligence, in addition to solution leveraging advanced machine learning for cost evaluations, scheduling, KPI planning, and measurement and risk profiling, Cerqueira said, are fundamentally designed to enhance the teamwork approach.

“Collaboration is a key factor in the design of AI systems, considering the creation of structured and consistent labeled data (in an automatically, semi-automatically or manually manner) and knowledge curation. The collaborative production of structured data and shared knowledge promotes applicability of machine .earning models, giving users better results in terms of prediction accuracy over time,” he said.

The hope is that if such data is described in a common representation for all coworkers, designers of predictive models can take advantage of that, and reapply their models to different but related problems.

“By using AI systems, coworkers may benefit from each other’s contribution over time,” he said.

Cerqueira said that the ideal AI system, like that of Hyperknowledge, should be designed to act as transparently as possible, while supporting experts in their regular activities. Keeping track of the consumption and production of conceptual knowledge and data is crucial to structuring such investigative processes.

“The disconnection observed in the O&G industry in terms of technologies and lack of crosscutting knowledge representation approaches is a critical issue to be addressed,” he said.

Hyperknowledge Trails provides an alternative perspective with a holistic representation geared toward the connection of user actions and intent, data fragments and meanings. This representational approach promotes applicability of AI systems, while enhancing communicability among teams and the interoperability of the tools they use.

“In practice though, it is not uncommon that even opportunistic approaches have some impact on user practices. Sure, there is always a learning curve to overcome when adhering to new dynamics of work. However, supporting experts with AI systems is a win-win situation, if they are consistently designed through a human-centered and collaborative perspective,” he said.