Machine Learning on Permian Seismic?

Pioneer Natural Resources is improving the accuracy of its Midland basin depth-models by up to 40%, using a machine-learning algorithm to re-calibrate its seismic from well logs. Faster drilling and better production rates should follow.


Pioneer Natural Resources has patented a new methodology in 2018, to calibrate its seismic images in the Permian, with reference to its well-logs. Ordinarily this task would be challenging and time-intensive. But the new algorithm employs machine-learning. This places it at the cutting edge of Permian data-science, where just 2% of technical papers have used ML in the past year (chart below).

machine learning on permian seismic

Specifically, a multi-layer neural network model iteratively improves the estimates of key seismic parameters from the log data (e.g., impedance, sonic velocity, Young’s modulus, Poisson’s ratio) (chart below). This algorithm improves the vertical accuracy of seismic interpretations by up to 40%.

The neural network creates different inversion volume estimates (208) from the well logs (202) and their attributes (204)

Improved well-placement and geo-steering. The patent cites how “reflectors that were previously unmappable on conventional seismic data can be mapped so horizontal wells can be more accurately placed”. This will be used to target wells into larger-capacity reservoirs and to inform well completion parameters.

Improved drilling-times. The company also cited a need to avoid drilling through carbonate debris flows in the Midland basin. They are excessively hard, damage drill-bits and lead to costly ‘trips’. Instead, it is intended to use the better-calibrated seismic to steer well-paths through brittle organic facies. Thus, we expect the innovation to lower costs and improve well-economics

Pioneer screens as one of the top quartile operators, across all the technologies we have diligenced so far (chart below). Although, please note, we are still “early” in our project to categorize who has the best technologies in oil and gas.

If you would like to read our latest deep-dive note on shale-technology it is linked here. The full database, covering all 300 technical papers is available here.

Patent Source: Meek, R., (2018). High Resolution Seismic Data Derived from Pre-Stack Inversion and Machine Learning. Pioneer Natural Resources USA, patent WO2018201114


Please log in to view this content

Copyright: Thunder Said Energy, 2019-2024.