Abstract

In the current environment, leaner oil and gas exploration teams must deliver increasingly reliable evaluations at a faster pace. At the same time, data of varying qualities are more abundant and available than ever. Petroleum system analysis workflows are essential to de-risk opportunities but remain lengthy. This limits iterations within and between disciplines and the completeness of the resulting volume, risk, and value/economic scenarios.

We present, in collaboration with This is Petroleum Systems (t!Ps), a novel approach, XCaliber, for fast iterations so that more time can be spent on the final technical and economic integration, illustrated in the Gulf Coast basin context. It is based on an interactive thermal framework, powered at its core by a machine learning (ML)–based basin simulator that can compute in seconds high-resolution basin-scale temperature and standard thermal stress results.

The ML–based basin model is trained end-to-end on full physics basin simulation results that account for key physical variables and transient processes. The trained model is a straight replacement of a traditional basin simulator, while being thousands of times faster (Fig. 1)

In the Gulf Coast Basin, this speed-up has allowed for the simultaneous inversion of crustal radiogenic heat production (RHP) at 32,000 well locations within a few minutes, given a regional structural model and a curated temperature database.

The calibrated RHP values are used to predict temperature and thermal stress maps (Fig. 2), which are then combined with source rocks kinetics to compute expulsion history, accumulation volumes, and hydrocarbon quality. Multiple high-resolution scenarios can be delivered in seconds to support risk evaluation workflows, at basin or prospect scale.