Regarding labeling in our solution, how can we get accurate labels to effectively train supervised models? EGML assumes event labels are often missing or incorrect. It recognizes that labels are easier to create using a platform that gives experts all the information they need to review and capturing everything they notice about the event. Any logs and workover reports generated by operators or engineers are visible and searchable in context of all the signals, production and state of the well at that time.
Each label can be edited with full tracking of the tags and comments of each subject matter expert (SME). Labels can be prioritized and statused allowing SMEs to express their opinions without requiring absolute decisions. In many cases experts do not know exactly what happened. Different experts can review and collaborate on the same information and reach different opinions. Using a crowd of experts helps determine what most likely happened by combining different opinions.
Most importantly, all this should happen quickly and efficiently. EGML ensures that each type of expert is notified of events as they are identified and brings together all the information needed to assess what happened. Pressures, temperatures, frequencies, messages, well bore diagrams, well configurations, well history, workover reports and any other information needed to diagnose the well is visualized with a couple clicks or taps in an intuitive user interface allowing rapid note taking and tagging. See the entire life of the well or drill into the last hour of operation with the drag of a mouse. Review a list of potential events in minutes instead of hours.
With improved workflow and intuitive access to all information about the well, events are discovered and diagnosed more quickly and the accuracy of the diagnosis improves. There is no need to manually enter the date and time the problem started and stopped, when the pump was shut down and restarted, or if the workover was started and completed. Point, touch and click on what you see and each interaction is captured by the system. No more mistyping information or incorrectly entering wells or characteristics. Anything that can be detected or calculated is automatically logged along with expert observations. Accuracy and timeliness improve so that labels become the true source of information necessary to fuel artificial intelligence.
To learn more about expert-guided machine learning, access our white paper entitled “Expert-Guided Machine Learning: Engaging Petroleum Experts’ Know-How in Petroleum Engineering and Oilfield Operations” here.