Geology Driven EUR Prediction Using Deep Learning
- L. Crnkovic-Friis (Peltarion Energy) | M. Erlandson (Peltarion Energy)
- Document ID
- Society of Petroleum Engineers
- SPE Annual Technical Conference and Exhibition, 28-30 September, Houston, Texas, USA
- Publication Date
- Document Type
- Conference Paper
- 2015. Society of Petroleum Engineers
- 5.8.1 Tight Gas, 5 Reservoir Desciption & Dynamics, 6.1.5 Human Resources, Competence and Training, 7.6 Information Management and Systems, 1.6.6 Directional Drilling, 5.8.4 Shale Oil, 7 Management and Information, 4.1.2 Separation and Treating, 4 Facilities Design, Construction and Operation, 4.1 Processing Systems and Design, 6 Health, Safety, Security, Environment and Social Responsibility, 1.6 Drilling Operations, 6.1 HSSE & Social Responsibility Management, 5.8 Unconventional and Complex Reservoirs
- deep neural network, EUR, prediction, hydraulic fracturing, OGIP
- 9 in the last 30 days
- 879 since 2007
- Show more detail
- View rights & permissions
|SPE Member Price:||USD 8.50|
|SPE Non-Member Price:||USD 25.00|
We present a geology driven deep learning Estimated Ultimate Recovery (EUR) prediction model for multi-stage hydraulically fractured horizontal wells in tight gas and oil reservoirs. The novel approach was made possible by recent development in the field of deep learning and the use of big data (200,000+ geological data points and 800+ wells). A Deep Neural Network (DNN) was trained to learn the relationship between geology and the average EUR (estimated by decline analysis). The model was validated on wells from other geological regions to show its generalization capabilities.
The DNN model we present significantly outperforms both volumetric estimates and type curve region averages (even on highly developed acreage). It generalizes well across geological areas with limited loss in accuracy. On a test region not used during model creation it produces a mean absolute percentage error of 33.1% compared to 69.7% for type curve averages. Oil and gas recovery are treated separately and the model outputs the oil to gas ratio. The model was trained and tested on data from the Eagle Ford Shale but the general methodology should be applicable to other resource plays.
The model is applicable in the exploration stage, as it only requires geological data. This is important as type curve regions require production data to be constructed, and are thus not available until the area has been in production for some time.
|File Size||3 MB||Number of Pages||10|
LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep Learning. Nature, doi:10.1038/nature14539.