Technical Session – Digital E&P: Deep Learning Seismic Inversion

Abstract:

Halliburton – Digital E&P: Deep Learning Seismic Inversion

Big Data and Digital Technology has created significant value in various industries such as the travel industry, transport industry and online shopping. The E&P industry is still in an embryonic stage regarding the exploitation of Big data and data science. We at Halliburton are applying Digital Technology, Big Data Analytics, Artificial Intelligence and Machine Learning to the various disciplines of the Exploration and Production Industry including  Subsurface applications such as data driven Reservoir Engineering and integrating Geology to predict wear on drillbits to reduce NPT and automated well offsetting.

We are using methods using Deep Learning and Neural networks to create Seismic Inversion volumes and comparing these volumes to inversion products created using traditional workflows. The dataset used for the comparison is from the NorthSea’s Hedron Field, an oil and gas field located 190km offshore in the Haltenbanken region of the Norwegian Sea in water depths of around 350m. The field comprised of two blocks, 6507/7 and 6507/8, covering an area of 38km². The Heidrun field situated in the Halten Terrace of the Mid-Norwegian Continental Shelf consists of sandstone reservoirs of four Jurassic age formations namely Garn, Ile, Tilje and Åre. The depth of the reservoir is up to 2,300m beneath the seabed.

The workflow used for this exercise initially consisted of sonic log calibration and well tie analysis to create a velocity model design for the inversion job parametrization, once this was done the results were fed into a Deep Learning Program and processed via neural networks to create an inversion Acoustic Impedence product. Other inversion properties and volumes were also modeled.

The proposed seismic inversion method proves to be much easier for geophysicists who are not seismic inversion specialists.  The method involves learning from examples (acoustic impedance well logs vs. seismic post stack data and velocity models) using Deep Learning techniques to create a pseudo-acoustic impedance volumes. In constrast with inversions based on forward modeling, the proposed method works well with depth based seismic data. As a result, wavelet extraction and low-frequency models are unnecessary. Using Deep learning Seismic Inversion can be executed with very good results.

Leave a Reply

Your email address will not be published. Required fields are marked *