Expert: Shreshth Srivastav
Senior Data Scientist
Operators are inundated with vast and growing volumes of digital borehole data as the number of logs, cores, surveys and petrophysical analyses per well is growing. Which makes it difficult to maintain data quality and consistency. Borehole metadata (data about data) is crucial throughout the cycle of data governance and thus can impact user workflows or strategic decisions. Conventional methods of controlling borehole metadata quality are solely based on business rules and are therefore missing hidden insights stored in historical metadata.
What if there is a way to fully automate the borehole metadata quality management process with tracking the confidence level of the system? What if the system not only learns from existing business KPIs but also integrates hidden patterns from the historical metadata?
The schematic below shows the process of model building which is applied to QC the borehole metadata. The data is ingested from a high-quality database and pre-processed to build a machine learning (ML) model. At this point the ML model, with its data-driven rules, is stored along with the traditional module with its business-driven rules. A new set of borehole metadata is passed through a preliminary QC done using Landmark’s Recall Raven™ software to identify quality issues with the metadata such as missing or invalid values, among others.
Figure: The solution workflow* includes 1) data pre-processing, 2) model building, 3) metadata prediction, 4) prediction applied in the decision process, and fed back to the master database for continuous learning, and 5) intermediate results reports. (CI in Process 5 is Confidence Index.)
Next, the ML model and the traditional module are applied and compared to suggest the correct value (for example, in the case of a missing value). In the decision module you can define the model confidence level. Depending on this threshold, a certain custom condition is activated to automatically apply the fix.
Once the issue is fixed, the corrected value is fed back to the master data repository for the ML model to re-train and improve its performance. This ensures continuous model updates in a very short maintenance window.
Finally, a QC report is generated to help you track the ML model’s performance and monitor metadata health.
Since this solution combines data-driven and business rules, users gain an additional resource to help overcome the classic limitations associated with data management. It boosts the data manager’s confidence to quickly correct the data quality issue by providing data fix options with high levels of reliability, while handling uncertainty in a systematic way. Moreover, this is a fully automated data driven or machine learning-enabled borehole metadata management technique that can enable real time model updates, resulting in saved time and expense.
As part of our SmartDigital® co-innovation service, we can collaborate alongside you to help you learn how AI and ML can help to enhance your data management and make a demonstrable impact on your E&P business. contact us today at firstname.lastname@example.org.
*Workflow covered by US Patent (pending).