Fast 3D Seismic Interpretation with Unsupervised Deep Learning: Application to a Potash Network in the North Sea

We present a new Artificial Intelligence (AI) tool enabling fast and accurate interpretation of 3D geological objects from seismic data. The tool is powered by deep learning, a set of techniques representing the latest breakthroughs in machine learning. The original method proposed in this work is built around generative models (unsupervised deep learning) which leads to several key advantages compared with other state-of-the-art AI approaches. Powerful interpretation models can be trained from the seismic data alone, without requiring a dedicated problem-specific training set. Moreover, this computationally intensive training phase can therefore be conducted automatically without human supervision (for instance as a background task during data importation) and produces powerful models which can be generically applied to a wide array of interpretation problems.

The result is a highly-versatile tool able to deliver fast and accurate predictions based on minimal user interactions while receiving real-time feedback from the AI. We demonstrate the practicality of the overall approach by performing the interpretation of a complex intra-salt potash network in the UK North Sea. By being directly applicable to a wide range of data and problems amongst others key features, we believe that our proposed solution has the potential to revolutionise interpretation workflows.