Predictive Analysis Requires Quality Data Across a Well Cohort
A well cohort describes the group of wells that OspreyData uses to create a modeled solution for your oilfield. Quality source data among well cohorts are essential to predictive analysis. In an artificial intelligence project, it is important to understand that all wells might not be candidates to participate in the building and training of the solution. While it is possible to have different cohorts for different diagnostic, failure or optimization conditions, it is best to have a single cohort. Teams will spend a substantial amount of time to gain a better understanding of the design, implementation and operation of a set of wells. Therefore, a clear and consistent set of known wells is critical to our methodology.
For example, the chart above is an output of a recent data quality assessment performed by OspreyData. It outlines the number of wells that were initially considered for this project, then through a series of evaluation steps, wells were removed. In this case, we started with 174 wells for consideration and ended with a single cohort of only 100 wells.
The project teams worked collaboratively to establish the criteria for evaluation in this assessment. Not all data quality assessments have the same or as strict a set of guidelines, but it is normal to see a reduction in the number of viable wells. In some cases, 30 to 40% of the available wells are removed during the selection process.
To clarify, the cohort is used during the construction of machine learning models. Once a set of models is built, they can be used on all of the wells available. There is a possibility that some wells may still not participate, typically due to missing sensors on the specific well. For instance, if we are attempting to detect “holes in tubing” for a well, then “casing pressure” and “tubing pressure” are required.
Over the last several weeks, our blogs have covered how to get quality source data, and it is clear that it is important. For production, if an organization is consistently using allocated production, actual production may vary by roughly 10%. At OspreyData, we have seen much higher rates when the allocated wells have a number of up and down cycles on production. This can cause problems with a true-up of the actual decline in the well and can ultimately lead to inaccurate reserve determinations. The downside is that reserves drive a company valuation and ensure ongoing employment for all of the teams.
There may be an electric submersible pump (ESP) that sees spikes in overheating, which may not be seen without adequate frequency in the polling of data. This will most certainly shorten the life of those very expensive pumps. Similarly, on a rod pump well, it is easy to see “pounding” on Dynacards, but the identification of “friction” is not always as noticeable. Quality data can show “friction” and help to avoid some of the issues that friction causes. Repairing holes in tubing or casing can be very expensive, but quality data leads to faster response times, and time is money.
If you would like to communicate with sales or engineering about how we can help you evaluate your source data and build a cohort model that leads to faster response times and a higher ROI, then feel free to contact us here. We will respond to your inquiry to answer your questions as soon as possible.