--> --> Applied Insight From Machine Learning Applications Using Analog Data

2019 AAPG Annual Convention and Exhibition:

Datapages, Inc.Print this page

Applied Insight From Machine Learning Applications Using Analog Data


Applied Insight from Machine Learning Applications using analog data

Artificial intelligence. Industry 4.0. Big Data. Machine learning. What does all this mean? Is it a series of buzz words or something that will transform the industry? If so when? A vision is a great thing if it can drive progress. However, the gap between a vision and implementation in the oil and gas industry is substantial. In the perfect view of AI a few simple rules would help us predict the location of the next big discovery and we would not need sub-surface teams. The prediction tasks across E&P are however very hard to codify. The sub-surface is a multi-dimensional problem and the advent of the latest analytical techniques permits the chance to see through and combine empirical observations. Delivering progress and tangible value requires momentum. Momentum comes from small steps and small successes, and that is why one-way C&C Reservoirs are responding to the opportunities new machine learning code is offering, is through the development of pragmatic forecasting algorithms based on data structured by experienced geoscientist. Machine learning algorithms offer an automated approach to regression to find a best fit algorithm to predict an observed parameter. The observed parameter we have worked upon is recovery factor based on structured input knowledge from proven global oil and gas reservoirs. We will share our observations on which algorithms have provided the best results and the controlling factors we have found for predicting recovery factor in different depositional environments. Using a standardised classification system and structured data we will show how the pragmatic application of machine learning can solve real challenges. Geoscientists and Reservoir Engineers can do this themselves, if they have structured data, thus ensuring the results are grounded in sound sub-surface principles. There is so much more than can be done to integrate these approaches into portfolio management, play assessment, uncertainty calibration and reservoir simulation. This is not the first-time attempts have been made to predict recovery factor and the learnings from old approaches will be reviewed. We should also not forget that machine learning is not a new tool in geoscience. Seismic interpretation uses machine learning with some success. However, it is not a without challenges. Who hasn’t drilled a well on inverted seismic to find an “unexpected result” which post well is then regarded as an excellent additional calibration point, rather than the failure! At some point the success of all of these initiatives needs to be measured in tangible value add. We believe improved recovery factor forecasting and identifying the optimal development plan can deliver production performance upside and support reserve bookings. Both offer near term tangible value.