When we consider the good source data required to power machine learning, we must consider the full life-cycle of the well. Most use-cases that OspreyData is building have moved beyond utilizing just a set of time-series sensor values. Well designs and completion reports provide insight to the physical aspects of the well. In some cases, they can also provide a measure of the performance expectations for the well.
Understanding what maintenance history or which chemical treatments were performed aids in model effectiveness. Changes in well behavior that correlate to the maintenance or chemical applications provide different insight to well behavior and failures. Lastly – this can’t be stressed enough for failure diagnostics and detection – detailed failure reports including work-over and tear- down reports are critical. The understanding and categorization of failure details can be of great consequence in constructing models. The more detailed data picture that is provided, the more accurate the models become.
It is clear that many field based priorities are dependent on the understood or perceived production of a well. Developing a stronger understanding of production allocations can allow systems to construct better interpretations in the differences in production across wells in the same tank battery.
Better yet, a triangulation of production allocations methodology coupled with tank levels to time series sensor stream can better pinpoint which wells are underperforming versus a general reduction of all wells in the battery.
The concept that “more data is always better” must be tempered with a thoughtful assessment of how that source data provides additional information. For more information about quality source data for oil and gas, request our white paper on data quality.