Stephane Gesbert (Shell)
Exploration has been facing headwinds. Discovery sizes are on the decline and the exploration success rate has dropped from 40 to 35% over the last decade. Emerging plays are often modest in size, with more challenging, riskier reservoirs. This is sometimes summarized with the tagline, “the easy oil is gone”: we go deeper, into harder rocks, and chase increasingly subtle plays. Our seismic interpreters are faced with the challenge to detect new plays and subtle prospects in ever increasing amounts of data.
Recent breakthroughs in Artificial Intelligence, and more specifically Deep Learning, have caught the attention of the industry and fostered new hopes in assisting interpreters to deliver prospects with increased speed, accuracy, or scope. There have been encouraging, even exciting results, but it is still early days and Machine Learning comes with very specific challenges that must be addressed – state of-the-art neural nets are very well suited for many tasks, but not all.
In this presentation, I will showcase applications of machine Learning and Deep learning in the realm of seismic interpretation at Shell, from fault interpretation to various seismic attributes, all the way to Quantitative Interpretation. I will reflect on the strengths and weaknesses of various approaches and discuss challenges and opportunities ahead.