The application of data science and machine learning is transforming petroleum geoscience workflows. Routine, yet time-consuming and important tasks can be made more efficient by the application of machine learning-based assisted interpretation, freeing the geoscientist to carry out tasks with greater value. Accuracy, reproducibility, and understanding of uncertainty are also improved and greater insight can be gained. Biostratigraphic data is very common in the industry but requires deep specialised knowledge and significant time to interpret, hence it can be underutilized. However, the form of the data makes it suitable for the application of machine learning techniques. The applications of machine leaning have been tested on biostratigraphic data from a set of typical industry wells to facilitate the interpretation of biozone/age and paleoenvironment. Application of Random Forest and Naïve Bayesian algorithms achieved results comparable to standard human interpretation, although pre-processing of the data (e.g. removal of spurious reworked or caved data) proved beneficial. Critical to the success of the project was the close working relationship between data scientists and subject matter experts in order to capture the nuances of biostratigraphic data and its interpretation. The work forms a case study for application to other geoscience data types.
|Original language||English (US)|
|Title of host publication||81st EAGE Conference and Exhibition 2019|
|Publisher||EAGE Publishing BVfirstbreak@eage.org|
|State||Published - Jun 3 2019|