Oil and gas (O&G) companies spend considerable time and resources to identify and evaluate prospects before they drill an exploration well. ReservoAIr combines geological and geophysical methods with machine learning (ML) techniques to enhance reservoir interpretation quality and efficiency. The novelty of ReservoAIr is to combine prediction of sands based on generated geomorphological features in synthetic seismic with the prediction of lithology (rock types) derived from well logs and partial stacks (angle dependent seismic amplitudes) to predict reservoir sands and enhance reservoir characterization. In order to validate and train the sand prediction, ray-based seismic modeling will be used to generate synthetic seismic. Qualitative validation of the ML model predictions will be performed by industry partners and by the modeling approach of the synthetic seismic. The project partners will develop an interactive human-machine feedback loop allowing geoscientists (users) to improve the underlying ML models. The most optimal networks and methods for detecting sand deposits and lithology will be utilized. RagnaRock's proprietary techniques of predicting horizons and faults will be used as the starting point for this research. ML models are only as good as the data they are trained on. Utilization of diverse training data sets, together with the consortium of domain expertise in both ML and geoscience, lays the foundation for ReservoAIr to exceed the reliability, usability, and performance of recently presented methods. ReservoAIr will leverage ML’s superior ability of pattern recognition with abundant amounts of data, guided by cross-disciplinary geoscience expertise, to extract information not apparent to the human eye. ReservoAIr represents a new end-to-end workflow for reservoir characterization, enabling O&G companies in faster and more informed exploration decisions, saving them millions in exploration.
Project leader: Åsmund Heir
Institution: RAGNAROCK GEO AS