Battelle Institute on big data in oil and gas

TAMU big data conference paper compares different machine learning approaches.

Speaking at a Texas A&M University conference earlier this year, Srikanta Mishra from the Battelle Institute presented a paper on big data applications in E&P. Big data analytics (BDA) holds the promise of new insights into geoscience, reservoir management, operations and maintenance.

Mishra, citing work done by his colleague, Shuvajit Bhattacharya on machine learning-based lithofacies classification, compared various ML approaches including support vector machine, artificial neural network, self organizing maps and graph-based clustering. SVM proved successful at predicting sweet spots in the Marcellus shale. Another key reference for comparing the different ML approaches is Hastie’s seminal 2008 book on The elements of statistical learning. Hastie showed that some approaches are better at forecasting, while other provide more interpretable results. Occam’s razor (using the simplest approach you can) is a way of avoiding the ‘curse of dimensionality.’ Models should ideally be capable of feeding back into a better understanding of the process and of the sensitivity of inputs. Aggregating results over a set of competing models can provide better understanding and prediction. QV Beven’s ‘equifinality’ concept. Mishra wound up with a word of caution. Prediction supposes that the same physical processes will operate in the future. In shale, this will not be the case if the flow regime switches from transient to boundary dominated flow. More from Battelle.

Click here to comment on this article

Click here to view this article in context on a desktop

© Oil IT Journal - all rights reserved.