Deep Learning Based Horizon Interpretation

Attempts to fully automate seismic interpretation date back to the earliest days of interpretation workstations and have met with limited success. Even with state of art automated and semi-automated tracking approaches, horizon interpretation of 3D seismic data remains a challenging and time consuming task.

This effort is compounded when seismic data is reprocessed, or time lapsed data is made available and the original tracked horizon needs reinterpreting. To that end, an improvement in overall efficiency could be achieved if previously interpreted horizons could be autonomously morphed to fit new datasets.

A new artificial intelligence workflow is proposed that is capable of transferring a degree of geological understanding between similar 3D seismic datasets (4D, reprocessed) in order to morph horizons picked on one dataset to another.

The proposed workflow uses a deep learning neural network to learn the geological characteristics of an event in one dataset and recognise the same event in another dataset, even when the event is visibly different or has shifted location.

Deep learning neural networks have demonstrated the ability to learn and distinguish subtle differences in events between multiple volumes and automatically adjust previous tracked horizons, which would be time consuming to identify using traditional interpretation techniques.